test_20n_1024bs.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "af054338-49e9-44b4-8af6-d60a1b7e6b01",
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch\n",
    "from time import gmtime, strftime\n",
    "\n",
    "from attention_dynamic_model import AttentionDynamicModel, set_decode_type\n",
    "from reinforce_baseline import RolloutBaseline\n",
    "from train import train_model\n",
    "\n",
    "from utils import create_data_on_disk, get_cur_time"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "excellent-parent",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "2021-08-31 00:16:29 model initialized\n",
      "2021-08-31 00:16:29 validation dataset created and saved on the disk\n"
     ]
    }
   ],
   "source": [
    "# Params of model\n",
    "SAMPLES = 1280000 # 1024*1250\n",
    "BATCH = 1024\n",
    "START_EPOCH = 0\n",
    "END_EPOCH = 40\n",
    "FROM_CHECKPOINT = False\n",
    "embedding_dim = 128\n",
    "LEARNING_RATE = 0.0001\n",
    "ROLLOUT_SAMPLES = 10000\n",
    "NUMBER_OF_WP_EPOCHS = 1\n",
    "GRAD_NORM_CLIPPING = 1.0\n",
    "BATCH_VERBOSE = 625\n",
    "VAL_BATCH_SIZE = 1000\n",
    "VALIDATE_SET_SIZE = 10000\n",
    "SEED = 1234\n",
    "GRAPH_SIZE = 20\n",
    "FILENAME = 'VRP_{}_{}'.format(GRAPH_SIZE, strftime(\"%Y-%m-%d\", gmtime()))\n",
    "\n",
    "# Initialize model\n",
    "model_pt = AttentionDynamicModel(embedding_dim).cuda()\n",
    "set_decode_type(model_pt, \"sampling\")\n",
    "print(get_cur_time(), 'model initialized')\n",
    "\n",
    "# Create and save validation dataset\n",
    "validation_dataset = create_data_on_disk(GRAPH_SIZE,\n",
    "                                         VALIDATE_SET_SIZE,\n",
    "                                         is_save=True,\n",
    "                                         filename=FILENAME,\n",
    "                                         is_return=True,\n",
    "                                         seed = SEED)\n",
    "print(get_cur_time(), 'validation dataset created and saved on the disk')\n",
    "\n",
    "# Initialize optimizer\n",
    "optimizer = torch.optim.Adam(params=model_pt.parameters(), lr=LEARNING_RATE)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "98f77ae4-709f-401b-8689-b359769e2f8a",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  7.90it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Evaluating baseline model on baseline dataset (epoch = 0)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.12it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "2021-08-31 00:16:31 baseline initialized\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    }
   ],
   "source": [
    "# Initialize baseline\n",
    "baseline = RolloutBaseline(model_pt,\n",
    "                           wp_n_epochs = NUMBER_OF_WP_EPOCHS,\n",
    "                           epoch = 0,\n",
    "                           num_samples=ROLLOUT_SAMPLES,\n",
    "                           filename = FILENAME,\n",
    "                           from_checkpoint = FROM_CHECKPOINT,\n",
    "                           embedding_dim=embedding_dim,\n",
    "                           graph_size=GRAPH_SIZE\n",
    "                           )\n",
    "print(get_cur_time(), 'baseline initialized')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "8efbbcc3-0fb2-4d87-88b7-5fbcd66f54ef",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 0: 0it [00:00, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Current decode type: sampling\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 0: 1it [00:01,  1.16s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 6.027115345001221, clipped_norm = 0.9999998211860657\n",
      "Epoch 0 (batch = 0): Loss: -1.0668154954910278: Cost: 13.029962539672852\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 0: 626it [07:12,  1.48it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 2.394676923751831, clipped_norm = 0.9999995827674866\n",
      "Epoch 0 (batch = 625): Loss: 0.41713735461235046: Cost: 8.813619613647461\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 0: 1250it [15:00,  1.39it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  7.25it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Evaluating candidate model on baseline dataset (callback epoch = 0)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.17it/s]\n",
      "Rollout greedy execution:   0%|          | 0/10 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 0 candidate mean 7.130025863647461, baseline epoch 0 mean 10.678730010986328, difference -3.548704147338867\n",
      "p-value: 0.0\n",
      "Update baseline\n",
      "Evaluating baseline model on baseline dataset (epoch = 0)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.08it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  6.79it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "alpha was updated to 1.0\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.27it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Validation score: 7.142499923706055\n",
      "2021-08-31 00:31:36 Epoch 0: Loss: 0.05772543326020241: Cost: 8.126152038574219\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 625/625 [02:03<00:00,  5.06it/s]\n",
      "batch calculation at epoch 1: 0it [00:00, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Current decode type: sampling\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 1: 1it [00:00,  1.42it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 2.3114888668060303, clipped_norm = 0.9999996423721313\n",
      "Epoch 1 (batch = 0): Loss: -1.0338002443313599: Cost: 7.215219974517822\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 1: 626it [08:26,  1.29it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 1.5616954565048218, clipped_norm = 0.9999993443489075\n",
      "Epoch 1 (batch = 625): Loss: -0.04109835997223854: Cost: 7.0990190505981445\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 1: 1250it [16:48,  1.24it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  7.18it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Evaluating candidate model on baseline dataset (callback epoch = 1)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  7.67it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  8.03it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1 candidate mean 6.824679851531982, baseline epoch 1 mean 7.127614974975586, difference -0.3029351234436035\n",
      "p-value: 0.0\n",
      "Update baseline\n",
      "Evaluating baseline model on baseline dataset (epoch = 1)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  7.80it/s]\n",
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  7.96it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Validation score: 6.833199977874756\n",
      "2021-08-31 00:50:33 Epoch 1: Loss: 0.43523985147476196: Cost: 7.0198774337768555\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 625/625 [02:10<00:00,  4.80it/s]\n",
      "batch calculation at epoch 2: 0it [00:00, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Current decode type: sampling\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 2: 1it [00:00,  1.30it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 2.4715218544006348, clipped_norm = 0.9999996423721313\n",
      "Epoch 2 (batch = 0): Loss: -0.6143490672111511: Cost: 6.918206214904785\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 2: 626it [08:20,  1.31it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 2.28598690032959, clipped_norm = 0.9999995827674866\n",
      "Epoch 2 (batch = 625): Loss: -0.2358335554599762: Cost: 6.835482597351074\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 2: 1250it [16:25,  1.27it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  7.30it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Evaluating candidate model on baseline dataset (callback epoch = 2)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  7.01it/s]\n",
      "Rollout greedy execution:   0%|          | 0/10 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 2 candidate mean 6.712669372558594, baseline epoch 2 mean 6.847395420074463, difference -0.13472604751586914\n",
      "p-value: 3.860358203974445e-255\n",
      "Update baseline\n",
      "Evaluating baseline model on baseline dataset (epoch = 2)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.13it/s]\n",
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  7.65it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Validation score: 6.69290018081665\n",
      "2021-08-31 01:09:13 Epoch 2: Loss: -0.05244047939777374: Cost: 6.799927234649658\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 625/625 [02:12<00:00,  4.71it/s]\n",
      "batch calculation at epoch 3: 0it [00:00, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Current decode type: sampling\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 3: 1it [00:00,  1.21it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 2.1257359981536865, clipped_norm = 0.9999995827674866\n",
      "Epoch 3 (batch = 0): Loss: -0.3617994785308838: Cost: 6.717039585113525\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 3: 626it [08:08,  1.36it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 1.9289133548736572, clipped_norm = 0.9999995231628418\n",
      "Epoch 3 (batch = 625): Loss: -0.28829658031463623: Cost: 6.721423625946045\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 3: 1250it [16:07,  1.29it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  7.55it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Evaluating candidate model on baseline dataset (callback epoch = 3)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  7.38it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  8.14it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 3 candidate mean 6.625886917114258, baseline epoch 3 mean 6.673469543457031, difference -0.04758262634277344\n",
      "p-value: 4.1064769689789866e-45\n",
      "Update baseline\n",
      "Evaluating baseline model on baseline dataset (epoch = 3)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  7.88it/s]\n",
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  7.92it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Validation score: 6.648099899291992\n",
      "2021-08-31 01:27:38 Epoch 3: Loss: -0.21513542532920837: Cost: 6.705970287322998\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 625/625 [02:09<00:00,  4.81it/s]\n",
      "batch calculation at epoch 4: 0it [00:00, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Current decode type: sampling\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 4: 1it [00:00,  1.42it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 2.002413272857666, clipped_norm = 0.999999463558197\n",
      "Epoch 4 (batch = 0): Loss: -0.3305237591266632: Cost: 6.700907230377197\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 4: 626it [08:03,  1.35it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 1.8158410787582397, clipped_norm = 0.999999463558197\n",
      "Epoch 4 (batch = 625): Loss: -0.22202523052692413: Cost: 6.660002708435059\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 4: 1250it [16:04,  1.30it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  7.90it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Evaluating candidate model on baseline dataset (callback epoch = 4)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  7.61it/s]\n",
      "Rollout greedy execution:   0%|          | 0/10 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 4 candidate mean 6.586111068725586, baseline epoch 4 mean 6.64369535446167, difference -0.057584285736083984\n",
      "p-value: 2.4792885020810764e-73\n",
      "Update baseline\n",
      "Evaluating baseline model on baseline dataset (epoch = 4)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  7.49it/s]\n",
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  7.64it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Validation score: 6.585599899291992\n",
      "2021-08-31 01:45:57 Epoch 4: Loss: -0.16910643875598907: Cost: 6.647656440734863\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 625/625 [01:57<00:00,  5.31it/s]\n",
      "batch calculation at epoch 5: 0it [00:00, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Current decode type: sampling\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 5: 1it [00:00,  1.35it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 2.4170424938201904, clipped_norm = 0.9999997019767761\n",
      "Epoch 5 (batch = 0): Loss: -0.25215989351272583: Cost: 6.585050582885742\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 5: 626it [08:17,  1.37it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 2.3629395961761475, clipped_norm = 0.9999995231628418\n",
      "Epoch 5 (batch = 625): Loss: -0.24504508078098297: Cost: 6.614116668701172\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 5: 1250it [16:16,  1.28it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  8.46it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Evaluating candidate model on baseline dataset (callback epoch = 5)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.18it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  8.42it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 5 candidate mean 6.56804084777832, baseline epoch 5 mean 6.596716403961182, difference -0.028675556182861328\n",
      "p-value: 2.953747600076083e-24\n",
      "Update baseline\n",
      "Evaluating baseline model on baseline dataset (epoch = 5)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.18it/s]\n",
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  7.90it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Validation score: 6.557700157165527\n",
      "2021-08-31 02:04:16 Epoch 5: Loss: -0.21084755659103394: Cost: 6.606468200683594\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 625/625 [01:53<00:00,  5.48it/s]\n",
      "batch calculation at epoch 6: 0it [00:00, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Current decode type: sampling\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 6: 1it [00:00,  1.44it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 1.9304836988449097, clipped_norm = 0.9999995231628418\n",
      "Epoch 6 (batch = 0): Loss: -0.20161205530166626: Cost: 6.552918910980225\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 6: 626it [08:13,  1.35it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 2.2945096492767334, clipped_norm = 0.9999995827674866\n",
      "Epoch 6 (batch = 625): Loss: -0.23037846386432648: Cost: 6.586703300476074\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 6: 1250it [16:09,  1.29it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  8.09it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Evaluating candidate model on baseline dataset (callback epoch = 6)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  7.88it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  8.46it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 6 candidate mean 6.533603668212891, baseline epoch 6 mean 6.558847427368164, difference -0.025243759155273438\n",
      "p-value: 8.209737291949247e-21\n",
      "Update baseline\n",
      "Evaluating baseline model on baseline dataset (epoch = 6)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.16it/s]\n",
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  7.96it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Validation score: 6.533400058746338\n",
      "2021-08-31 02:22:24 Epoch 6: Loss: -0.20123356580734253: Cost: 6.579024314880371\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 625/625 [01:55<00:00,  5.42it/s]\n",
      "batch calculation at epoch 7: 0it [00:00, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Current decode type: sampling\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 7: 1it [00:00,  1.28it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 1.7838009595870972, clipped_norm = 0.9999994039535522\n",
      "Epoch 7 (batch = 0): Loss: -0.2341514229774475: Cost: 6.604394435882568\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 7: 626it [07:54,  1.39it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 2.0799198150634766, clipped_norm = 0.999999463558197\n",
      "Epoch 7 (batch = 625): Loss: -0.22457249462604523: Cost: 6.564850330352783\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 7: 1250it [16:01,  1.30it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  7.50it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Evaluating candidate model on baseline dataset (callback epoch = 7)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  7.48it/s]\n",
      "Rollout greedy execution:   0%|          | 0/10 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 7 candidate mean 6.510906219482422, baseline epoch 7 mean 6.523597717285156, difference -0.012691497802734375\n",
      "p-value: 2.586950588666648e-07\n",
      "Update baseline\n",
      "Evaluating baseline model on baseline dataset (epoch = 7)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  7.02it/s]\n",
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  7.05it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Validation score: 6.523499965667725\n",
      "2021-08-31 02:40:26 Epoch 7: Loss: -0.20451880991458893: Cost: 6.559495449066162\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 625/625 [01:56<00:00,  5.35it/s]\n",
      "batch calculation at epoch 8: 0it [00:00, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Current decode type: sampling\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 8: 1it [00:00,  1.24it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 2.0416557788848877, clipped_norm = 0.9999995827674866\n",
      "Epoch 8 (batch = 0): Loss: -0.18443427979946136: Cost: 6.581435203552246\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 8: 626it [07:53,  1.40it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 2.371631145477295, clipped_norm = 0.9999995827674866\n",
      "Epoch 8 (batch = 625): Loss: -0.18671269714832306: Cost: 6.544984340667725\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 8: 1250it [15:55,  1.31it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  8.20it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Evaluating candidate model on baseline dataset (callback epoch = 8)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.31it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  8.06it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 8 candidate mean 6.509038925170898, baseline epoch 8 mean 6.524211883544922, difference -0.015172958374023438\n",
      "p-value: 2.28342310032771e-10\n",
      "Update baseline\n",
      "Evaluating baseline model on baseline dataset (epoch = 8)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.22it/s]\n",
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.11it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Validation score: 6.505799770355225\n",
      "2021-08-31 02:58:23 Epoch 8: Loss: -0.17670832574367523: Cost: 6.542794704437256\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 625/625 [01:54<00:00,  5.46it/s]\n",
      "batch calculation at epoch 9: 0it [00:00, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Current decode type: sampling\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 9: 1it [00:00,  1.27it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 2.7055022716522217, clipped_norm = 0.9999997019767761\n",
      "Epoch 9 (batch = 0): Loss: -0.17154733836650848: Cost: 6.497735023498535\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 9: 626it [07:51,  1.36it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 3.116502523422241, clipped_norm = 0.9999996423721313\n",
      "Epoch 9 (batch = 625): Loss: -0.19612887501716614: Cost: 6.535821437835693\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 9: 1250it [15:37,  1.33it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  8.14it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Evaluating candidate model on baseline dataset (callback epoch = 9)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  7.45it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  8.13it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 9 candidate mean 6.497844219207764, baseline epoch 9 mean 6.519439220428467, difference -0.021595001220703125\n",
      "p-value: 4.9528113142628694e-21\n",
      "Update baseline\n",
      "Evaluating baseline model on baseline dataset (epoch = 9)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.10it/s]\n",
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.38it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Validation score: 6.487299919128418\n",
      "2021-08-31 03:15:59 Epoch 9: Loss: -0.17904146015644073: Cost: 6.530584812164307\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 625/625 [02:07<00:00,  4.90it/s]\n",
      "batch calculation at epoch 10: 0it [00:00, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Current decode type: sampling\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 10: 1it [00:00,  1.39it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 2.239109754562378, clipped_norm = 0.9999995827674866\n",
      "Epoch 10 (batch = 0): Loss: -0.15314054489135742: Cost: 6.525386333465576\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 10: 626it [07:48,  1.34it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 2.263974905014038, clipped_norm = 0.9999995827674866\n",
      "Epoch 10 (batch = 625): Loss: -0.20949597656726837: Cost: 6.5220627784729\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 10: 1250it [15:32,  1.34it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  6.53it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Evaluating candidate model on baseline dataset (callback epoch = 10)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  7.07it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  7.35it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 10 candidate mean 6.51194429397583, baseline epoch 10 mean 6.507481098175049, difference 0.00446319580078125\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  7.42it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Validation score: 6.49399995803833\n",
      "2021-08-31 03:33:43 Epoch 10: Loss: -0.19809743762016296: Cost: 6.519343852996826\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 625/625 [02:07<00:00,  4.92it/s]\n",
      "batch calculation at epoch 11: 0it [00:00, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Current decode type: sampling\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 11: 1it [00:00,  1.43it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 2.3465282917022705, clipped_norm = 0.9999995827674866\n",
      "Epoch 11 (batch = 0): Loss: -0.1591038703918457: Cost: 6.518631458282471\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 11: 626it [07:44,  1.44it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 1.9023748636245728, clipped_norm = 0.999999463558197\n",
      "Epoch 11 (batch = 625): Loss: -0.16161136329174042: Cost: 6.510589122772217\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 11: 1250it [15:26,  1.35it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  7.90it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Evaluating candidate model on baseline dataset (callback epoch = 11)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  7.80it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  8.12it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 11 candidate mean 6.494513034820557, baseline epoch 11 mean 6.507481098175049, difference -0.012968063354492188\n",
      "p-value: 5.367208512256011e-08\n",
      "Update baseline\n",
      "Evaluating baseline model on baseline dataset (epoch = 11)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  7.75it/s]\n",
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  7.70it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Validation score: 6.478499889373779\n",
      "2021-08-31 03:51:21 Epoch 11: Loss: -0.15416735410690308: Cost: 6.508296012878418\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 625/625 [01:53<00:00,  5.49it/s]\n",
      "batch calculation at epoch 12: 0it [00:00, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Current decode type: sampling\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 12: 1it [00:00,  1.45it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 2.3118860721588135, clipped_norm = 0.9999995827674866\n",
      "Epoch 12 (batch = 0): Loss: -0.13856112957000732: Cost: 6.503539085388184\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 12: 626it [08:00,  1.41it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 3.272122383117676, clipped_norm = 0.9999997019767761\n",
      "Epoch 12 (batch = 625): Loss: -0.16633811593055725: Cost: 6.499916076660156\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 12: 1250it [15:44,  1.32it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  7.80it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Evaluating candidate model on baseline dataset (callback epoch = 12)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.07it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  8.21it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 12 candidate mean 6.4845685958862305, baseline epoch 12 mean 6.4756622314453125, difference 0.008906364440917969\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  7.25it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Validation score: 6.486999988555908\n",
      "2021-08-31 04:09:02 Epoch 12: Loss: -0.16621172428131104: Cost: 6.49953556060791\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 625/625 [01:52<00:00,  5.57it/s]\n",
      "batch calculation at epoch 13: 0it [00:00, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Current decode type: sampling\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 13: 1it [00:00,  1.35it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 2.5574100017547607, clipped_norm = 0.9999996423721313\n",
      "Epoch 13 (batch = 0): Loss: -0.2664288878440857: Cost: 6.501054763793945\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 13: 626it [07:43,  1.37it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 2.2577412128448486, clipped_norm = 0.9999995231628418\n",
      "Epoch 13 (batch = 625): Loss: -0.14383673667907715: Cost: 6.496207237243652\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 13: 1250it [15:37,  1.33it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  7.00it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Evaluating candidate model on baseline dataset (callback epoch = 13)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  7.71it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  8.30it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 13 candidate mean 6.472360610961914, baseline epoch 13 mean 6.4756622314453125, difference -0.0033016204833984375\n",
      "p-value: 0.09375605353275249\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  7.96it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Validation score: 6.476399898529053\n",
      "2021-08-31 04:26:35 Epoch 13: Loss: -0.13755233585834503: Cost: 6.494082450866699\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 625/625 [01:52<00:00,  5.56it/s]\n",
      "batch calculation at epoch 14: 0it [00:00, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Current decode type: sampling\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 14: 1it [00:00,  1.36it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 1.8702430725097656, clipped_norm = 0.9999995231628418\n",
      "Epoch 14 (batch = 0): Loss: -0.18117381632328033: Cost: 6.47097110748291\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 14: 626it [07:42,  1.40it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 2.8650879859924316, clipped_norm = 0.9999996423721313\n",
      "Epoch 14 (batch = 625): Loss: -0.12349574267864227: Cost: 6.489189624786377\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 14: 1250it [15:22,  1.36it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  7.92it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Evaluating candidate model on baseline dataset (callback epoch = 14)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  7.90it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  8.92it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 14 candidate mean 6.4587626457214355, baseline epoch 14 mean 6.4756622314453125, difference -0.016899585723876953\n",
      "p-value: 2.1023129801757364e-12\n",
      "Update baseline\n",
      "Evaluating baseline model on baseline dataset (epoch = 14)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.89it/s]\n",
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.61it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Validation score: 6.460599899291992\n",
      "2021-08-31 04:43:55 Epoch 14: Loss: -0.11532382667064667: Cost: 6.487607479095459\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 625/625 [02:05<00:00,  5.00it/s]\n",
      "batch calculation at epoch 15: 0it [00:00, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Current decode type: sampling\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 15: 1it [00:00,  1.41it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 1.5229368209838867, clipped_norm = 0.9999992847442627\n",
      "Epoch 15 (batch = 0): Loss: -0.08166652917861938: Cost: 6.488171577453613\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 15: 626it [07:37,  1.37it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 1.9161174297332764, clipped_norm = 0.9999995231628418\n",
      "Epoch 15 (batch = 625): Loss: -0.1460743546485901: Cost: 6.482095241546631\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 15: 1250it [15:16,  1.36it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  6.58it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Evaluating candidate model on baseline dataset (callback epoch = 15)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.22it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  8.41it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 15 candidate mean 6.459384918212891, baseline epoch 15 mean 6.46403694152832, difference -0.0046520233154296875\n",
      "p-value: 0.023140809495987763\n",
      "Update baseline\n",
      "Evaluating baseline model on baseline dataset (epoch = 15)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.58it/s]\n",
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  7.57it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Validation score: 6.458600044250488\n",
      "2021-08-31 05:01:20 Epoch 15: Loss: -0.14074893295764923: Cost: 6.480199813842773\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 625/625 [01:50<00:00,  5.64it/s]\n",
      "batch calculation at epoch 16: 0it [00:00, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Current decode type: sampling\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 16: 1it [00:00,  1.32it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 2.939380168914795, clipped_norm = 0.9999997019767761\n",
      "Epoch 16 (batch = 0): Loss: -0.10295625030994415: Cost: 6.4456586837768555\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 16: 626it [07:51,  1.39it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 2.353914737701416, clipped_norm = 0.9999996423721313\n",
      "Epoch 16 (batch = 625): Loss: -0.13890086114406586: Cost: 6.476585388183594\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 16: 1250it [15:31,  1.34it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  7.61it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Evaluating candidate model on baseline dataset (callback epoch = 16)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.29it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  8.18it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 16 candidate mean 6.446474075317383, baseline epoch 16 mean 6.448409557342529, difference -0.0019354820251464844\n",
      "p-value: 0.2082425275862821\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.18it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Validation score: 6.457300186157227\n",
      "2021-08-31 05:18:46 Epoch 16: Loss: -0.1312413513660431: Cost: 6.474483013153076\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 625/625 [01:53<00:00,  5.53it/s]\n",
      "batch calculation at epoch 17: 0it [00:00, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Current decode type: sampling\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 17: 1it [00:00,  1.24it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 2.3753247261047363, clipped_norm = 0.9999995827674866\n",
      "Epoch 17 (batch = 0): Loss: -0.1336967498064041: Cost: 6.49203634262085\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 17: 626it [07:37,  1.40it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 2.703552007675171, clipped_norm = 0.9999996423721313\n",
      "Epoch 17 (batch = 625): Loss: -0.11924676597118378: Cost: 6.472006797790527\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 17: 1250it [15:31,  1.34it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  7.19it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Evaluating candidate model on baseline dataset (callback epoch = 17)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  7.29it/s]\n",
      "Rollout greedy execution:   0%|          | 0/10 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 17 candidate mean 6.4396209716796875, baseline epoch 17 mean 6.448409557342529, difference -0.008788585662841797\n",
      "p-value: 8.244845001307312e-05\n",
      "Update baseline\n",
      "Evaluating baseline model on baseline dataset (epoch = 17)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  7.43it/s]\n",
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  7.23it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Validation score: 6.446100234985352\n",
      "2021-08-31 05:36:16 Epoch 17: Loss: -0.11284375190734863: Cost: 6.469790458679199\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 625/625 [01:50<00:00,  5.64it/s]\n",
      "batch calculation at epoch 18: 0it [00:00, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Current decode type: sampling\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 18: 1it [00:00,  1.43it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 1.9750452041625977, clipped_norm = 0.999999463558197\n",
      "Epoch 18 (batch = 0): Loss: -0.12269773334264755: Cost: 6.479109764099121\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 18: 626it [07:38,  1.44it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 2.088261604309082, clipped_norm = 0.9999995827674866\n",
      "Epoch 18 (batch = 625): Loss: -0.13232126832008362: Cost: 6.466843128204346\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 18: 1250it [15:13,  1.37it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  8.79it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Evaluating candidate model on baseline dataset (callback epoch = 18)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.29it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  8.33it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 18 candidate mean 6.460752964019775, baseline epoch 18 mean 6.467161178588867, difference -0.006408214569091797\n",
      "p-value: 0.00325299872111177\n",
      "Update baseline\n",
      "Evaluating baseline model on baseline dataset (epoch = 18)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.25it/s]\n",
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.57it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Validation score: 6.438000202178955\n",
      "2021-08-31 05:53:24 Epoch 18: Loss: -0.128618985414505: Cost: 6.464822769165039\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 625/625 [02:04<00:00,  5.04it/s]\n",
      "batch calculation at epoch 19: 0it [00:00, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Current decode type: sampling\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 19: 1it [00:00,  1.42it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 1.9301033020019531, clipped_norm = 0.999999463558197\n",
      "Epoch 19 (batch = 0): Loss: -0.1141408383846283: Cost: 6.466921329498291\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 19: 626it [07:38,  1.47it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 2.313491106033325, clipped_norm = 0.9999995827674866\n",
      "Epoch 19 (batch = 625): Loss: -0.13846318423748016: Cost: 6.460400104522705\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 19: 1250it [15:14,  1.37it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  8.20it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Evaluating candidate model on baseline dataset (callback epoch = 19)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.21it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:00,  9.16it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 19 candidate mean 6.436435699462891, baseline epoch 19 mean 6.440877914428711, difference -0.0044422149658203125\n",
      "p-value: 0.018370215595497084\n",
      "Update baseline\n",
      "Evaluating baseline model on baseline dataset (epoch = 19)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.86it/s]\n",
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.46it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Validation score: 6.433899879455566\n",
      "2021-08-31 06:10:48 Epoch 19: Loss: -0.13243529200553894: Cost: 6.459164142608643\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 625/625 [01:51<00:00,  5.60it/s]\n",
      "batch calculation at epoch 20: 0it [00:00, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Current decode type: sampling\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 20: 1it [00:00,  1.34it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 2.380662202835083, clipped_norm = 0.9999996423721313\n",
      "Epoch 20 (batch = 0): Loss: -0.11956396698951721: Cost: 6.524051666259766\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 20: 626it [07:52,  1.39it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 1.9079939126968384, clipped_norm = 0.999999463558197\n",
      "Epoch 20 (batch = 625): Loss: -0.13873715698719025: Cost: 6.456024646759033\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 20: 1250it [15:27,  1.35it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:00,  9.06it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Evaluating candidate model on baseline dataset (callback epoch = 20)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  7.88it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  9.00it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 20 candidate mean 6.430592060089111, baseline epoch 20 mean 6.429439067840576, difference 0.0011529922485351562\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.28it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Validation score: 6.438000202178955\n",
      "2021-08-31 06:28:09 Epoch 20: Loss: -0.13290825486183167: Cost: 6.453587532043457\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 625/625 [01:49<00:00,  5.69it/s]\n",
      "batch calculation at epoch 21: 0it [00:00, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Current decode type: sampling\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 21: 1it [00:00,  1.40it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 2.296560764312744, clipped_norm = 0.9999995827674866\n",
      "Epoch 21 (batch = 0): Loss: -0.11638560891151428: Cost: 6.491926193237305\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 21: 626it [07:37,  1.49it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 2.0692358016967773, clipped_norm = 0.9999995231628418\n",
      "Epoch 21 (batch = 625): Loss: -0.1260686218738556: Cost: 6.451298713684082\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 21: 1250it [15:10,  1.37it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  7.42it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Evaluating candidate model on baseline dataset (callback epoch = 21)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:02<00:00,  4.88it/s]\n",
      "Rollout greedy execution:   0%|          | 0/10 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 21 candidate mean 6.434387683868408, baseline epoch 21 mean 6.429439067840576, difference 0.004948616027832031\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:03<00:00,  3.16it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Validation score: 6.443600177764893\n",
      "2021-08-31 06:45:16 Epoch 21: Loss: -0.12190187722444534: Cost: 6.44996976852417\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 625/625 [02:02<00:00,  5.11it/s]\n",
      "batch calculation at epoch 22: 0it [00:00, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Current decode type: sampling\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 22: 1it [00:00,  1.55it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 1.9411216974258423, clipped_norm = 0.9999995231628418\n",
      "Epoch 22 (batch = 0): Loss: -0.19911369681358337: Cost: 6.464524269104004\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 22: 626it [07:33,  1.41it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 2.80659556388855, clipped_norm = 0.9999997019767761\n",
      "Epoch 22 (batch = 625): Loss: -0.11476957052946091: Cost: 6.4479079246521\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 22: 1250it [15:06,  1.38it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:00,  9.13it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Evaluating candidate model on baseline dataset (callback epoch = 22)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.48it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  8.49it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 22 candidate mean 6.419052600860596, baseline epoch 22 mean 6.429439067840576, difference -0.010386466979980469\n",
      "p-value: 2.4589847648483877e-06\n",
      "Update baseline\n",
      "Evaluating baseline model on baseline dataset (epoch = 22)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.17it/s]\n",
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.02it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Validation score: 6.421000003814697\n",
      "2021-08-31 07:02:29 Epoch 22: Loss: -0.11239375174045563: Cost: 6.447535991668701\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 625/625 [02:06<00:00,  4.95it/s]\n",
      "batch calculation at epoch 23: 0it [00:00, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Current decode type: sampling\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 23: 1it [00:00,  1.44it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 2.1981406211853027, clipped_norm = 0.9999995827674866\n",
      "Epoch 23 (batch = 0): Loss: -0.06402552872896194: Cost: 6.464924335479736\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 23: 626it [07:33,  1.44it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 2.2856457233428955, clipped_norm = 0.9999995231628418\n",
      "Epoch 23 (batch = 625): Loss: -0.13125038146972656: Cost: 6.444540977478027\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 23: 1250it [15:08,  1.38it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  8.94it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Evaluating candidate model on baseline dataset (callback epoch = 23)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.43it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  8.52it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 23 candidate mean 6.43772029876709, baseline epoch 23 mean 6.427553653717041, difference 0.010166645050048828\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.35it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Validation score: 6.432000160217285\n",
      "2021-08-31 07:19:47 Epoch 23: Loss: -0.12801744043827057: Cost: 6.443129062652588\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 625/625 [01:50<00:00,  5.64it/s]\n",
      "batch calculation at epoch 24: 0it [00:00, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Current decode type: sampling\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 24: 1it [00:00,  1.40it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 2.3011114597320557, clipped_norm = 0.9999996423721313\n",
      "Epoch 24 (batch = 0): Loss: -0.1693931668996811: Cost: 6.461979866027832\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 24: 626it [07:35,  1.47it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 2.038404703140259, clipped_norm = 0.9999995231628418\n",
      "Epoch 24 (batch = 625): Loss: -0.12140243500471115: Cost: 6.441113471984863\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 24: 1250it [15:21,  1.36it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  7.16it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Evaluating candidate model on baseline dataset (callback epoch = 24)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.10it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  8.47it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 24 candidate mean 6.4296674728393555, baseline epoch 24 mean 6.427553653717041, difference 0.002113819122314453\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  7.90it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Validation score: 6.427800178527832\n",
      "2021-08-31 07:37:02 Epoch 24: Loss: -0.11823546141386032: Cost: 6.440165996551514\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 625/625 [01:53<00:00,  5.48it/s]\n",
      "batch calculation at epoch 25: 0it [00:00, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Current decode type: sampling\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 25: 1it [00:00,  1.45it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 2.0857040882110596, clipped_norm = 0.999999463558197\n",
      "Epoch 25 (batch = 0): Loss: -0.1358443796634674: Cost: 6.412321090698242\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 25: 626it [07:35,  1.43it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 2.331122875213623, clipped_norm = 0.9999995231628418\n",
      "Epoch 25 (batch = 625): Loss: -0.11102128028869629: Cost: 6.437289714813232\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 25: 1250it [15:08,  1.38it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  8.47it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Evaluating candidate model on baseline dataset (callback epoch = 25)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.42it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  8.56it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 25 candidate mean 6.427104473114014, baseline epoch 25 mean 6.427553653717041, difference -0.00044918060302734375\n",
      "p-value: 0.4197641887754707\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.14it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Validation score: 6.418399810791016\n",
      "2021-08-31 07:54:07 Epoch 25: Loss: -0.10792962461709976: Cost: 6.436948776245117\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 625/625 [02:05<00:00,  5.00it/s]\n",
      "batch calculation at epoch 26: 0it [00:00, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Current decode type: sampling\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 26: 1it [00:00,  1.46it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 2.5164759159088135, clipped_norm = 0.9999995827674866\n",
      "Epoch 26 (batch = 0): Loss: -0.1606641411781311: Cost: 6.435614109039307\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 26: 626it [07:33,  1.43it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 1.8455162048339844, clipped_norm = 0.9999995231628418\n",
      "Epoch 26 (batch = 625): Loss: -0.09915708005428314: Cost: 6.43414306640625\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 26: 1250it [15:05,  1.38it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  7.27it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Evaluating candidate model on baseline dataset (callback epoch = 26)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.19it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:00,  9.66it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 26 candidate mean 6.43090295791626, baseline epoch 26 mean 6.427553653717041, difference 0.00334930419921875\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.84it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Validation score: 6.4274001121521\n",
      "2021-08-31 08:11:20 Epoch 26: Loss: -0.09892246127128601: Cost: 6.433500289916992\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 625/625 [01:51<00:00,  5.60it/s]\n",
      "batch calculation at epoch 27: 0it [00:00, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Current decode type: sampling\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 27: 1it [00:00,  1.45it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 2.4960951805114746, clipped_norm = 0.9999995827674866\n",
      "Epoch 27 (batch = 0): Loss: -0.12203453481197357: Cost: 6.44228458404541\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 27: 626it [07:47,  1.50it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 2.6113409996032715, clipped_norm = 0.9999997019767761\n",
      "Epoch 27 (batch = 625): Loss: -0.09711451083421707: Cost: 6.430648326873779\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 27: 1250it [15:18,  1.36it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  7.42it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Evaluating candidate model on baseline dataset (callback epoch = 27)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  7.67it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  8.65it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 27 candidate mean 6.416680335998535, baseline epoch 27 mean 6.427553653717041, difference -0.01087331771850586\n",
      "p-value: 2.956774128129409e-07\n",
      "Update baseline\n",
      "Evaluating baseline model on baseline dataset (epoch = 27)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.51it/s]\n",
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.84it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Validation score: 6.412399768829346\n",
      "2021-08-31 08:28:34 Epoch 27: Loss: -0.0926089808344841: Cost: 6.430895805358887\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 625/625 [01:51<00:00,  5.59it/s]\n",
      "batch calculation at epoch 28: 0it [00:00, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Current decode type: sampling\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 28: 1it [00:00,  1.42it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 1.9016788005828857, clipped_norm = 0.999999463558197\n",
      "Epoch 28 (batch = 0): Loss: -0.11298555135726929: Cost: 6.396769046783447\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 28: 626it [07:32,  1.44it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 1.7338236570358276, clipped_norm = 0.999999463558197\n",
      "Epoch 28 (batch = 625): Loss: -0.1126340702176094: Cost: 6.429904937744141\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 28: 1250it [15:16,  1.36it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:00,  9.11it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Evaluating candidate model on baseline dataset (callback epoch = 28)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.61it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  8.39it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 28 candidate mean 6.421536922454834, baseline epoch 28 mean 6.426906585693359, difference -0.005369663238525391\n",
      "p-value: 0.004457721077421751\n",
      "Update baseline\n",
      "Evaluating baseline model on baseline dataset (epoch = 28)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.30it/s]\n",
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  7.84it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Validation score: 6.407199859619141\n",
      "2021-08-31 08:45:47 Epoch 28: Loss: -0.11238245666027069: Cost: 6.429055213928223\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 625/625 [01:49<00:00,  5.71it/s]\n",
      "batch calculation at epoch 29: 0it [00:00, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Current decode type: sampling\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 29: 1it [00:00,  1.46it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 1.654449224472046, clipped_norm = 0.999999463558197\n",
      "Epoch 29 (batch = 0): Loss: -0.11179798096418381: Cost: 6.422333240509033\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 29: 626it [07:35,  1.38it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 2.094503879547119, clipped_norm = 0.9999995231628418\n",
      "Epoch 29 (batch = 625): Loss: -0.11884600669145584: Cost: 6.426951885223389\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 29: 1250it [15:07,  1.38it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  8.79it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Evaluating candidate model on baseline dataset (callback epoch = 29)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.28it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  7.98it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 29 candidate mean 6.400547027587891, baseline epoch 29 mean 6.405243396759033, difference -0.004696369171142578\n",
      "p-value: 0.008952716851746256\n",
      "Update baseline\n",
      "Evaluating baseline model on baseline dataset (epoch = 29)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.53it/s]\n",
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  7.87it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Validation score: 6.405799865722656\n",
      "2021-08-31 09:02:49 Epoch 29: Loss: -0.11275883764028549: Cost: 6.424092769622803\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 625/625 [02:03<00:00,  5.06it/s]\n",
      "batch calculation at epoch 30: 0it [00:00, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Current decode type: sampling\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 30: 1it [00:00,  1.58it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 2.147448778152466, clipped_norm = 0.9999995231628418\n",
      "Epoch 30 (batch = 0): Loss: -0.07073217630386353: Cost: 6.421289443969727\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 30: 626it [07:34,  1.43it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 1.864005208015442, clipped_norm = 0.9999995827674866\n",
      "Epoch 30 (batch = 625): Loss: -0.11079518496990204: Cost: 6.42382287979126\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 30: 1250it [15:08,  1.38it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  7.23it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Evaluating candidate model on baseline dataset (callback epoch = 30)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  7.86it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  7.90it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 30 candidate mean 6.409717559814453, baseline epoch 30 mean 6.412018775939941, difference -0.0023012161254882812\n",
      "p-value: 0.12928506431417658\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  7.68it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Validation score: 6.401199817657471\n",
      "2021-08-31 09:20:04 Epoch 30: Loss: -0.11065547913312912: Cost: 6.42165470123291\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 625/625 [01:50<00:00,  5.65it/s]\n",
      "batch calculation at epoch 31: 0it [00:00, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Current decode type: sampling\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 31: 1it [00:00,  1.46it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 1.90645170211792, clipped_norm = 0.999999463558197\n",
      "Epoch 31 (batch = 0): Loss: -0.0676814392209053: Cost: 6.403485298156738\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 31: 626it [07:29,  1.40it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 1.983119249343872, clipped_norm = 0.9999995827674866\n",
      "Epoch 31 (batch = 625): Loss: -0.10963853448629379: Cost: 6.419801712036133\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 31: 1250it [15:13,  1.37it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  5.67it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Evaluating candidate model on baseline dataset (callback epoch = 31)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.06it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  8.12it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 31 candidate mean 6.408082485198975, baseline epoch 31 mean 6.412018775939941, difference -0.003936290740966797\n",
      "p-value: 0.027102541754457383\n",
      "Update baseline\n",
      "Evaluating baseline model on baseline dataset (epoch = 31)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.25it/s]\n",
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  7.71it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Validation score: 6.397200107574463\n",
      "2021-08-31 09:37:13 Epoch 31: Loss: -0.10574161261320114: Cost: 6.418583393096924\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 625/625 [01:50<00:00,  5.68it/s]\n",
      "batch calculation at epoch 32: 0it [00:00, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Current decode type: sampling\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 32: 1it [00:00,  1.37it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 1.6155447959899902, clipped_norm = 0.9999994039535522\n",
      "Epoch 32 (batch = 0): Loss: -0.07607337832450867: Cost: 6.393620014190674\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 32: 626it [07:32,  1.39it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 1.7107068300247192, clipped_norm = 0.9999994039535522\n",
      "Epoch 32 (batch = 625): Loss: -0.11797300726175308: Cost: 6.416791915893555\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 32: 1250it [15:04,  1.38it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  8.19it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Evaluating candidate model on baseline dataset (callback epoch = 32)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.08it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:00,  9.19it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 32 candidate mean 6.402783393859863, baseline epoch 32 mean 6.398603916168213, difference 0.004179477691650391\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.43it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Validation score: 6.399099826812744\n",
      "2021-08-31 09:54:10 Epoch 32: Loss: -0.11683840304613113: Cost: 6.415804862976074\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 625/625 [02:00<00:00,  5.17it/s]\n",
      "batch calculation at epoch 33: 0it [00:00, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Current decode type: sampling\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 33: 1it [00:00,  1.41it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 2.302586078643799, clipped_norm = 0.9999996423721313\n",
      "Epoch 33 (batch = 0): Loss: -0.12117548286914825: Cost: 6.4626312255859375\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 33: 626it [07:31,  1.48it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 1.9971516132354736, clipped_norm = 0.9999995231628418\n",
      "Epoch 33 (batch = 625): Loss: -0.11300701647996902: Cost: 6.4138665199279785\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 33: 1250it [15:01,  1.39it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  8.09it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Evaluating candidate model on baseline dataset (callback epoch = 33)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  7.80it/s]\n",
      "Rollout greedy execution:   0%|          | 0/10 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 33 candidate mean 6.3927483558654785, baseline epoch 33 mean 6.398603916168213, difference -0.005855560302734375\n",
      "p-value: 0.0014807427448274482\n",
      "Update baseline\n",
      "Evaluating baseline model on baseline dataset (epoch = 33)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.18it/s]\n",
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.65it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Validation score: 6.389400005340576\n",
      "2021-08-31 10:11:18 Epoch 33: Loss: -0.11321704089641571: Cost: 6.413163661956787\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 625/625 [01:49<00:00,  5.72it/s]\n",
      "batch calculation at epoch 34: 0it [00:00, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Current decode type: sampling\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 34: 1it [00:00,  1.40it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 1.646062970161438, clipped_norm = 0.9999993443489075\n",
      "Epoch 34 (batch = 0): Loss: -0.09556068480014801: Cost: 6.403127193450928\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 34: 626it [07:47,  1.41it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 2.4534010887145996, clipped_norm = 0.9999995827674866\n",
      "Epoch 34 (batch = 625): Loss: -0.11878566443920135: Cost: 6.41259241104126\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 34: 1250it [15:16,  1.36it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  7.95it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Evaluating candidate model on baseline dataset (callback epoch = 34)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.22it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  8.69it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 34 candidate mean 6.411680221557617, baseline epoch 34 mean 6.407678127288818, difference 0.004002094268798828\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.82it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Validation score: 6.393499851226807\n",
      "2021-08-31 10:28:27 Epoch 34: Loss: -0.11725159734487534: Cost: 6.41105842590332\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 625/625 [01:50<00:00,  5.64it/s]\n",
      "batch calculation at epoch 35: 0it [00:00, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Current decode type: sampling\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 35: 1it [00:00,  1.53it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 2.8202273845672607, clipped_norm = 0.9999995827674866\n",
      "Epoch 35 (batch = 0): Loss: -0.1350281536579132: Cost: 6.429954528808594\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 35: 626it [07:31,  1.43it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 1.8516535758972168, clipped_norm = 0.999999463558197\n",
      "Epoch 35 (batch = 625): Loss: -0.11585825681686401: Cost: 6.409656524658203\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 35: 1250it [15:00,  1.39it/s]\n",
      "Rollout greedy execution:   0%|          | 0/10 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Evaluating candidate model on baseline dataset (callback epoch = 35)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:03<00:00,  3.33it/s]\n",
      "Rollout greedy execution:   0%|          | 0/10 [00:00<?, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 35 candidate mean 6.415002346038818, baseline epoch 35 mean 6.407678127288818, difference 0.00732421875\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:02<00:00,  3.49it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Validation score: 6.398099899291992\n",
      "2021-08-31 10:45:25 Epoch 35: Loss: -0.11503466963768005: Cost: 6.410204887390137\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 625/625 [01:58<00:00,  5.26it/s]\n",
      "batch calculation at epoch 36: 0it [00:00, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Current decode type: sampling\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 36: 1it [00:00,  1.34it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 2.054264783859253, clipped_norm = 0.9999995827674866\n",
      "Epoch 36 (batch = 0): Loss: -0.10297635197639465: Cost: 6.463846206665039\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 36: 626it [07:31,  1.45it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 1.8784699440002441, clipped_norm = 0.9999994039535522\n",
      "Epoch 36 (batch = 625): Loss: -0.1088085025548935: Cost: 6.4086174964904785\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 36: 1250it [14:59,  1.39it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  8.12it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Evaluating candidate model on baseline dataset (callback epoch = 36)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.00it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:00,  9.13it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 36 candidate mean 6.408956527709961, baseline epoch 36 mean 6.407678127288818, difference 0.0012784004211425781\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.68it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Validation score: 6.39300012588501\n",
      "2021-08-31 11:02:27 Epoch 36: Loss: -0.10925595462322235: Cost: 6.408689498901367\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 625/625 [01:58<00:00,  5.27it/s]\n",
      "batch calculation at epoch 37: 0it [00:00, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Current decode type: sampling\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 37: 1it [00:01,  1.18s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 1.7467371225357056, clipped_norm = 0.9999995231628418\n",
      "Epoch 37 (batch = 0): Loss: -0.04374488443136215: Cost: 6.429769039154053\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 37: 626it [07:38,  1.43it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 1.8659946918487549, clipped_norm = 0.999999463558197\n",
      "Epoch 37 (batch = 625): Loss: -0.10558528453111649: Cost: 6.406492710113525\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 37: 1250it [15:07,  1.38it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  8.16it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Evaluating candidate model on baseline dataset (callback epoch = 37)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.69it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  8.97it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 37 candidate mean 6.4077935218811035, baseline epoch 37 mean 6.407678127288818, difference 0.00011539459228515625\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.46it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Validation score: 6.3907999992370605\n",
      "2021-08-31 11:19:36 Epoch 37: Loss: -0.10428007692098618: Cost: 6.406471252441406\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 625/625 [01:48<00:00,  5.76it/s]\n",
      "batch calculation at epoch 38: 0it [00:00, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Current decode type: sampling\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 38: 1it [00:00,  1.53it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 2.0331599712371826, clipped_norm = 0.9999995827674866\n",
      "Epoch 38 (batch = 0): Loss: -0.12157275527715683: Cost: 6.3941264152526855\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 38: 626it [07:29,  1.43it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 1.496145248413086, clipped_norm = 0.9999993443489075\n",
      "Epoch 38 (batch = 625): Loss: -0.09738866984844208: Cost: 6.404356002807617\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 38: 1250it [15:12,  1.37it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  8.56it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Evaluating candidate model on baseline dataset (callback epoch = 38)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  7.99it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  7.74it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 38 candidate mean 6.40115213394165, baseline epoch 38 mean 6.407678127288818, difference -0.006525993347167969\n",
      "p-value: 0.0008919480082438409\n",
      "Update baseline\n",
      "Evaluating baseline model on baseline dataset (epoch = 38)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  7.93it/s]\n",
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  7.68it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Validation score: 6.38539981842041\n",
      "2021-08-31 11:36:41 Epoch 38: Loss: -0.09744840115308762: Cost: 6.40407133102417\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 625/625 [01:51<00:00,  5.63it/s]\n",
      "batch calculation at epoch 39: 0it [00:00, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Current decode type: sampling\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 39: 1it [00:00,  1.36it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 1.8033397197723389, clipped_norm = 0.9999994039535522\n",
      "Epoch 39 (batch = 0): Loss: -0.08509112149477005: Cost: 6.416381359100342\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 39: 626it [07:34,  1.40it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "grad_global_norm = 2.018285036087036, clipped_norm = 0.9999995827674866\n",
      "Epoch 39 (batch = 625): Loss: -0.10999958962202072: Cost: 6.4045329093933105\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "batch calculation at epoch 39: 1250it [15:00,  1.39it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  7.96it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Evaluating candidate model on baseline dataset (callback epoch = 39)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  8.58it/s]\n",
      "Rollout greedy execution:  10%|█         | 1/10 [00:00<00:01,  7.06it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 39 candidate mean 6.394413471221924, baseline epoch 39 mean 6.385366439819336, difference 0.00904703140258789\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Rollout greedy execution: 100%|██████████| 10/10 [00:01<00:00,  7.68it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Validation score: 6.3927998542785645\n",
      "2021-08-31 11:53:36 Epoch 39: Loss: -0.1081196740269661: Cost: 6.4034271240234375\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1080x648 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "train_model(optimizer,\n",
    "            model_pt,\n",
    "            baseline,\n",
    "            validation_dataset,\n",
    "            samples = SAMPLES,\n",
    "            batch = BATCH,\n",
    "            val_batch_size = VAL_BATCH_SIZE,\n",
    "            start_epoch = START_EPOCH,\n",
    "            end_epoch = END_EPOCH,\n",
    "            grad_norm_clipping = GRAD_NORM_CLIPPING,\n",
    "            batch_verbose = BATCH_VERBOSE,\n",
    "            graph_size = GRAPH_SIZE,\n",
    "            filename = FILENAME\n",
    "            )"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}