AdamwithClip(all_init).ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "AdamwithClip(all_init).ipynb",
      "provenance": [],
      "collapsed_sections": []
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    }
  },
  "cells": [
    {
      "cell_type": "code",
      "metadata": {
        "id": "4gLE-u90a-5l",
        "colab_type": "code",
        "outputId": "b4a839f9-b5dc-4d0f-94df-c680c4829357",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 118
        }
      },
      "source": [
        "import warnings\n",
        "import tensorflow as tf\n",
        "from keras_utils import SGDwithLR,AdamwithClip,RMSpropwithClip\n",
        "from keras import models\n",
        "from keras_focused import SimpleFocusedRNN\n",
        "from keras.optimizers import RMSprop\n",
        "from keras.datasets import imdb\n",
        "from keras.models import Sequential\n",
        "from keras.preprocessing import sequence\n",
        "from keras import layers\n",
        "from keras.layers import Flatten,SimpleRNN\n",
        "import numpy as np\n",
        "import numpy\n",
        "from keras import backend as K\n",
        "%matplotlib inline\n",
        "import pandas as pd\n",
        "import matplotlib.pyplot as plt\n",
        "\n",
        "def lr_settings(all_lr=0.01,sigma_lr=0.01,mu_lr=0.01,mom=0.9,decay_dict_lr=0.1):\n",
        "\n",
        "  lr_dict = {'all':all_lr,\n",
        "           'focus-1/Sigma_current:0': sigma_lr,'focus-1/Mu_current:0': mu_lr,'focus-1/kernel:0': all_lr,\n",
        "           'focus-1/Sigma_prev:0': sigma_lr,'focus-1/Mu_prev:0': mu_lr,'focus-1/recurrent_kernel:0': all_lr,\n",
        "           'dense-3/Weights:0':all_lr}\n",
        "        \n",
        "  #lr_dict = {'all':0.0001}\n",
        "\n",
        "  mom_dict = {'all':mom}\n",
        "  #decay_dict = {'all':0.9}\n",
        "  #mom_dict = {'all':0.9,'focus-1/Sigma:0': 0.25,'focus-1/Mu:0': 0.25,\n",
        "  #           'focus-2/Sigma:0': 0.25,'focus-2/Mu:0': 0.25}\n",
        "    \n",
        "  decay_dict = {'all':all_lr, 'focus-1/Sigma_current:0': decay_dict_lr,'focus-1/Mu_current:0':decay_dict_lr,'focus-1/Sigma_prev:0': decay_dict_lr,'focus-1/Mu_prev:0': decay_dict_lr}\n",
        "\n",
        "              #'focus-2/Sigma_current:0': 0.1,'focus-2/Mu_current:0': 0.1,'focus-2/Sigma_prev:0': 0.1,'focus-2/Mu_prev:0': 0.1}\n",
        "\n",
        "  clip_dict = {'focus-1/Sigma_current:0':(0.01,1.0),'focus-1/Mu_current:0':(0.0,1.0),'focus-1/Sigma_prev:0':(0.01,1.0),'focus-1/Mu_prev:0':(0.0,1.0)}\n",
        "\n",
        "             #'focus-2/Sigma_current:0':(0.05,1.0),'focus-2/Mu_current:0':(0.0,1.0),'focus-2/Sigma_prev:0':(0.05,1.0),'focus-2/Mu_prev:0':(0.0,1.0)}\n",
        "  return lr_dict,mom_dict,decay_dict,clip_dict"
      ],
      "execution_count": 1,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/html": [
              "<p style=\"color: red;\">\n",
              "The default version of TensorFlow in Colab will soon switch to TensorFlow 2.x.<br>\n",
              "We recommend you <a href=\"https://www.tensorflow.org/guide/migrate\" target=\"_blank\">upgrade</a> now \n",
              "or ensure your notebook will continue to use TensorFlow 1.x via the <code>%tensorflow_version 1.x</code> magic:\n",
              "<a href=\"https://colab.research.google.com/notebooks/tensorflow_version.ipynb\" target=\"_blank\">more info</a>.</p>\n"
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "The name for Record Variable has changed, use RecordWeights or RecordTensor instead\n",
            "Not working!\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "Using TensorFlow backend.\n"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "OTZ1KzXobAKw",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "from keras.callbacks import Callback\n",
        "class PrintLayerVariableStats(Callback):\n",
        "    def __init__(self,name,var,stat_functions,stat_names,num):\n",
        "        self.layername = name\n",
        "        self.varname = var\n",
        "        self.stat_list = stat_functions\n",
        "        self.stat_names = stat_names\n",
        "        self.num=num\n",
        "\n",
        "    def setVariableName(self,name, var):\n",
        "        self.layername = name\n",
        "        self.varname = var\n",
        "    def on_train_begin(self, logs={}):\n",
        "        all_params = self.model.get_layer(self.layername)._trainable_weights\n",
        "        all_weights = self.model.get_layer(self.layername).get_weights()\n",
        "        #print(\"self.model\",self.model)\n",
        "        #print(\"all_params\",all_params)\n",
        "        #print(\"self.layername\",self.layername)\n",
        "        #print(\"self.varname\",self.varname)\n",
        "        i=self.num\n",
        "        if(i == 0):\n",
        "            stat_str = [n+str(s(all_weights[i])) for s,n in zip(self.stat_list,self.stat_names)]\n",
        "            print(\"\\nStats for kernel:0 \", stat_str)\n",
        "        if(i == 1):\n",
        "            stat_str_1 = [n+str(s(all_weights[i])) for s,n in zip(self.stat_list,self.stat_names)]\n",
        "            print(\"Stats for Sigma_current:0 \", stat_str_1)\n",
        "        if(i == 2):\n",
        "            stat_str_2 = [n+str(s(all_weights[i])) for s,n in zip(self.stat_list,self.stat_names)]\n",
        "            print(\"Stats for Mu_current:0 \", stat_str_2)\n",
        "        if(i == 3):\n",
        "            stat_str_3 = [n+str(s(all_weights[i])) for s,n in zip(self.stat_list,self.stat_names)]\n",
        "            print(\"Stats for recurrent_kernel:0 \", stat_str_3)\n",
        "        \n",
        "        \n",
        "        \n",
        "\n",
        "        #def on_batch_end(self, batch, logs={}):\n",
        "        #    self.record.append(logs.get('loss'))\n",
        "\n",
        "    def on_epoch_end(self, epoch, logs={}):\n",
        "        all_weights = self.model.get_layer(self.layername).get_weights()\n",
        "        i=self.num\n",
        "        if(i == 0):\n",
        "            stat_str = [n+str(s(all_weights[i])) for s,n in zip(self.stat_list,self.stat_names)]\n",
        "            print(\"\\nStats for kernel:0 \", stat_str)\n",
        "        if(i == 1):\n",
        "            stat_str_1 = [n+str(s(all_weights[i])) for s,n in zip(self.stat_list,self.stat_names)]\n",
        "            print(\"Stats for Sigma_current:0 \", stat_str_1)\n",
        "        if(i == 2):\n",
        "            stat_str_2 = [n+str(s(all_weights[i])) for s,n in zip(self.stat_list,self.stat_names)]\n",
        "            print(\"Stats for Mu_current:0 \", stat_str_2)\n",
        "        if(i == 3):\n",
        "            stat_str_3 = [n+str(s(all_weights[i])) for s,n in zip(self.stat_list,self.stat_names)]\n",
        "            print(\"Stats for recurrent_kernel:0 \", stat_str_3)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "pODRn2NNbC3Y",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "from keras.optimizers import SGD\n",
        "def build_model(N=64,mod='dense', optimizer_s='SGDwithLR',dropout=0.2, recurrent_dropout=0.2, init_sigma_current= 0.1, init_sigma_prev=0.1,num_epochs=15,sgd_settings=None,dataset_percantage=100,\n",
        "                initer=None,distribution=None):\n",
        "    top_words = 5000\n",
        "    (X_train, y_train), (X_test, y_test) = imdb.load_data(num_words=top_words)\n",
        "    # truncate and pad input sequences\n",
        "    max_review_length = 500\n",
        "    if dataset_percantage == 100:\n",
        "      print(\"100\")\n",
        "      X_train = sequence.pad_sequences(X_train, maxlen=max_review_length)\n",
        "      X_test = sequence.pad_sequences(X_test, maxlen=max_review_length)\n",
        "    elif dataset_percantage == 75:\n",
        "      X_train = sequence.pad_sequences(X_train[0:18750], maxlen=max_review_length)\n",
        "      X_test = sequence.pad_sequences(X_test[0:18750], maxlen=max_review_length)\n",
        "      y_train = y_train[0:18750]\n",
        "      y_test = y_test[0:18750]\n",
        "      print(\"75\")\n",
        "    elif dataset_percantage == 50:\n",
        "      X_train = sequence.pad_sequences(X_train[0:12500], maxlen=max_review_length)\n",
        "      X_test = sequence.pad_sequences(X_test[0:12500], maxlen=max_review_length)\n",
        "      y_train = y_train[0:12500]\n",
        "      y_test = y_test[0:12500]\n",
        "    elif dataset_percantage == 25:\n",
        "      X_train = sequence.pad_sequences(X_train[0:6250], maxlen=max_review_length)\n",
        "      X_test = sequence.pad_sequences(X_test[0:6250], maxlen=max_review_length)\n",
        "      y_train = y_train[0:6250]\n",
        "      y_test = y_test[0:6250]\n",
        "    elif dataset_percantage == 10:\n",
        "      X_train = sequence.pad_sequences(X_train[0:2500], maxlen=max_review_length)\n",
        "      X_test = sequence.pad_sequences(X_test[0:2500], maxlen=max_review_length)\n",
        "      y_train = y_train[0:2500]\n",
        "      y_test = y_test[0:2500]\n",
        "    embedding_vecor_length = 32\n",
        "    model = Sequential()\n",
        "    model.add(layers.Embedding(top_words, embedding_vecor_length, input_length=max_review_length))\n",
        "    #model.add(Flatten())\n",
        "    if mod=='simplernn':\n",
        "        model.add(SimpleRNN(100))\n",
        "    elif mod=='focused':\n",
        "        model.add(SimpleFocusedRNN(units=N,\n",
        "                                  name='focus-1',\n",
        "                                  kernel_initializer='he_normal',\n",
        "                                  dropout=dropout,\n",
        "                                  recurrent_dropout=recurrent_dropout,\n",
        "                                  init_sigma_current=init_sigma_current,\n",
        "                                  init_sigma_prev=init_sigma_prev,\n",
        "                                   initer=initer,\n",
        "                                   distribution=distribution))\n",
        "               \n",
        "            \n",
        "    model.add(layers.Dense(1,name='dense-3',activation='sigmoid'))\n",
        "    \n",
        "\n",
        "    if optimizer_s == 'SGDwithLR' and sgd_settings != None:\n",
        "        decay_epochs =[3,5,8,11]      \n",
        "        opt = SGDwithLR(lr=sgd_settings[0], momentum=sgd_settings[1],decay=sgd_settings[2],clips=sgd_settings[3],decay_epochs=decay_epochs,verbose=1)#, decay=None)\n",
        "    elif optimizer_s == 'AdamwithClip':\n",
        "        opt=AdamwithClip()\n",
        "    elif optimizer_s=='RMSpropwithClip':\n",
        "        opt = RMSpropwithClip(lr=0.001, rho=0.9, epsilon=None, decay=0.0,clips=clip_dict)\n",
        "    elif optimizer_s == 'adam':\n",
        "        opt='adam'\n",
        "    else:\n",
        "        opt= SGD(lr=0.01, momentum=0.9)#, decay=None)\n",
        "    print(\"opt= \",opt)\n",
        "    model.compile(loss='binary_crossentropy', optimizer=opt, metrics=['accuracy'])\n",
        "    stat_func_name = ['max: ', 'mean: ', 'min: ', 'var: ', 'std: ']\n",
        "    stat_func_list = [np.max, np.mean, np.min, np.var, np.std]\n",
        "    callbacks = []\n",
        "    pr_0 = PrintLayerVariableStats(\"focus-1\",\"kernel:0\",stat_func_list,stat_func_name,0)\n",
        "    pr_1 = PrintLayerVariableStats(\"focus-1\",\"Sigma_current:0\",stat_func_list,stat_func_name,1)\n",
        "    pr_2 = PrintLayerVariableStats(\"focus-1\",\"Mu_current:0\",stat_func_list,stat_func_name,2)\n",
        "    pr_3 = PrintLayerVariableStats(\"focus-1\",\"recurrent_kernel:0\",stat_func_list,stat_func_name,3)\n",
        "    callbacks+=[pr_0,pr_1,pr_2,pr_3]\n",
        "\n",
        "\n",
        "    print(model.summary())\n",
        "    history = model.fit(X_train, y_train, epochs=num_epochs, batch_size=64,verbose=1,callbacks=callbacks)\n",
        "    # Final evaluation of the model\n",
        "    scores = model.evaluate(X_test, y_test, verbose=0)\n",
        "    print(\"Accuracy: %.2f%%\" % (scores[1]*100))\n",
        "    return history"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "onJ_U8ebbEjI",
        "colab_type": "code",
        "outputId": "67ec520b-2f46-4bb4-99f6-e56f878e4c59",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        }
      },
      "source": [
        "dataset_percantage=100  #use dataset %10,%25,%50,%75\n",
        "K.clear_session()\n",
        "mod='focused'\n",
        "N=100\n",
        "all_lr=0.015\n",
        "sigma_lr=0.015\n",
        "mu_lr=0.015\n",
        "mom=0.9\n",
        "decay_dict_lr=0.1\n",
        "numpy.random.seed(7)\n",
        "# load the dataset but only keep the top n words, zero the rest\n",
        "sgd_settings=lr_settings(all_lr=all_lr,sigma_lr=sigma_lr,mu_lr=mu_lr,mom=mom,decay_dict_lr=decay_dict_lr)\n",
        "print(sgd_settings[0])\n",
        "history_glorot_uniform = build_model(N,mod,optimizer_s='AdamwithClip',num_epochs=15,sgd_settings=None,dataset_percantage=dataset_percantage,initer='Glorot',distribution='uniform')#glorot uniform\n",
        "\n",
        "pd.DataFrame(history_glorot_uniform.history).plot(figsize=(8, 5))\n",
        "plt.grid(True)\n",
        "plt.gca().set_ylim(0, 1)\n",
        "plt.show()"
      ],
      "execution_count": 6,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "{'all': 0.015, 'focus-1/Sigma_current:0': 0.015, 'focus-1/Mu_current:0': 0.015, 'focus-1/kernel:0': 0.015, 'focus-1/Sigma_prev:0': 0.015, 'focus-1/Mu_prev:0': 0.015, 'focus-1/recurrent_kernel:0': 0.015, 'dense-3/Weights:0': 0.015}\n",
            "100\n",
            "input_shape (None, 32)\n",
            "self.input_dim=  32\n",
            "initilizer=  Glorot\n",
            "distribution uniform\n",
            "u_current Tensor(\"focus-1/transpose_2:0\", shape=(32, 100), dtype=float32)\n",
            "u_previous Tensor(\"focus-1/transpose_3:0\", shape=(100, 100), dtype=float32)\n",
            "u_current Tensor(\"focus-1/while/transpose:0\", shape=(32, 100), dtype=float32)\n",
            "u_previous Tensor(\"focus-1/while/transpose_1:0\", shape=(100, 100), dtype=float32)\n",
            "opt=  <keras_utils.AdamwithClip object at 0x7fecf85cae10>\n",
            "Model: \"sequential_1\"\n",
            "_________________________________________________________________\n",
            "Layer (type)                 Output Shape              Param #   \n",
            "=================================================================\n",
            "embedding_1 (Embedding)      (None, 500, 32)           160000    \n",
            "_________________________________________________________________\n",
            "focus-1 (SimpleFocusedRNN)   (None, 100)               13700     \n",
            "_________________________________________________________________\n",
            "dense-3 (Dense)              (None, 1)                 101       \n",
            "=================================================================\n",
            "Total params: 173,801\n",
            "Trainable params: 173,801\n",
            "Non-trainable params: 0\n",
            "_________________________________________________________________\n",
            "None\n",
            "\n",
            "Stats for kernel:0  ['max: 0.8', 'mean: 0.49999997', 'min: 0.2', 'var: 0.030606063', 'std: 0.17494589']\n",
            "Stats for Sigma_current:0  ['max: 0.1', 'mean: 0.10000003', 'min: 0.1', 'var: 8.881784e-16', 'std: 2.9802322e-08']\n",
            "Stats for Mu_current:0  ['max: 0.8', 'mean: 0.49999997', 'min: 0.2', 'var: 0.030606063', 'std: 0.17494589']\n",
            "Stats for recurrent_kernel:0  ['max: 0.1', 'mean: 0.10000003', 'min: 0.1', 'var: 8.881784e-16', 'std: 2.9802322e-08']\n",
            "Epoch 1/15\n",
            "25000/25000 [==============================] - 275s 11ms/step - loss: 0.5779 - acc: 0.6880\n",
            "\n",
            "Stats for kernel:0  ['max: 0.85936433', 'mean: 0.48834103', 'min: 0.10794642', 'var: 0.036198307', 'std: 0.19025853']\n",
            "Stats for Sigma_current:0  ['max: 0.26576295', 'mean: 0.1257033', 'min: 0.022133376', 'var: 0.0024083105', 'std: 0.04907454']\n",
            "Stats for Mu_current:0  ['max: 0.90054995', 'mean: 0.5029219', 'min: 0.11980081', 'var: 0.041398864', 'std: 0.20346712']\n",
            "Stats for recurrent_kernel:0  ['max: 0.18608893', 'mean: 0.11351242', 'min: 0.041742224', 'var: 0.0008907323', 'std: 0.029845139']\n",
            "Epoch 2/15\n",
            "25000/25000 [==============================] - 273s 11ms/step - loss: 0.4503 - acc: 0.7913\n",
            "\n",
            "Stats for kernel:0  ['max: 0.8991124', 'mean: 0.47884247', 'min: 0.093638726', 'var: 0.039116718', 'std: 0.19777946']\n",
            "Stats for Sigma_current:0  ['max: 0.3028377', 'mean: 0.14137334', 'min: 0.0010261795', 'var: 0.0049526677', 'std: 0.07037519']\n",
            "Stats for Mu_current:0  ['max: 0.90990114', 'mean: 0.49455968', 'min: 0.109790824', 'var: 0.041120313', 'std: 0.20278144']\n",
            "Stats for recurrent_kernel:0  ['max: 0.18939123', 'mean: 0.10829816', 'min: 0.0061888676', 'var: 0.0013358003', 'std: 0.036548603']\n",
            "Epoch 3/15\n",
            "25000/25000 [==============================] - 273s 11ms/step - loss: 0.4141 - acc: 0.8174\n",
            "\n",
            "Stats for kernel:0  ['max: 0.90759414', 'mean: 0.47885048', 'min: 0.08123039', 'var: 0.040847305', 'std: 0.20210716']\n",
            "Stats for Sigma_current:0  ['max: 0.3101727', 'mean: 0.14802842', 'min: 0.00080277043', 'var: 0.0060844375', 'std: 0.0780028']\n",
            "Stats for Mu_current:0  ['max: 0.98425263', 'mean: 0.49441493', 'min: 0.09147095', 'var: 0.043671474', 'std: 0.20897721']\n",
            "Stats for recurrent_kernel:0  ['max: 0.19588478', 'mean: 0.105962604', 'min: 0.006186162', 'var: 0.0014293977', 'std: 0.037807375']\n",
            "Epoch 4/15\n",
            "25000/25000 [==============================] - 273s 11ms/step - loss: 0.3828 - acc: 0.8322\n",
            "\n",
            "Stats for kernel:0  ['max: 0.9384576', 'mean: 0.4786893', 'min: 0.0702677', 'var: 0.043299217', 'std: 0.20808464']\n",
            "Stats for Sigma_current:0  ['max: 0.3463787', 'mean: 0.156527', 'min: 0.00080277043', 'var: 0.0075788796', 'std: 0.08705676']\n",
            "Stats for Mu_current:0  ['max: 0.9666275', 'mean: 0.49724495', 'min: 0.083527796', 'var: 0.04168051', 'std: 0.20415805']\n",
            "Stats for recurrent_kernel:0  ['max: 0.19793937', 'mean: 0.10039649', 'min: -0.0012047429', 'var: 0.0016624645', 'std: 0.040773332']\n",
            "Epoch 5/15\n",
            "25000/25000 [==============================] - 272s 11ms/step - loss: 0.3680 - acc: 0.8408\n",
            "\n",
            "Stats for kernel:0  ['max: 0.9414637', 'mean: 0.48308986', 'min: 0.09748789', 'var: 0.045213677', 'std: 0.21263508']\n",
            "Stats for Sigma_current:0  ['max: 0.3605886', 'mean: 0.15905039', 'min: 0.00080277043', 'var: 0.008415738', 'std: 0.09173733']\n",
            "Stats for Mu_current:0  ['max: 0.99155563', 'mean: 0.49859208', 'min: 0.07404894', 'var: 0.044094615', 'std: 0.20998718']\n",
            "Stats for recurrent_kernel:0  ['max: 0.22117198', 'mean: 0.10381139', 'min: -0.0012047429', 'var: 0.002146738', 'std: 0.046332907']\n",
            "Epoch 6/15\n",
            "25000/25000 [==============================] - 273s 11ms/step - loss: 0.3665 - acc: 0.8399\n",
            "\n",
            "Stats for kernel:0  ['max: 0.944962', 'mean: 0.48165077', 'min: 0.060569763', 'var: 0.048172254', 'std: 0.21948178']\n",
            "Stats for Sigma_current:0  ['max: 0.37278044', 'mean: 0.16050811', 'min: -0.004305011', 'var: 0.009679254', 'std: 0.0983832']\n",
            "Stats for Mu_current:0  ['max: 1.0199765', 'mean: 0.49751782', 'min: 0.0735543', 'var: 0.04545275', 'std: 0.2131965']\n",
            "Stats for recurrent_kernel:0  ['max: 0.22362697', 'mean: 0.105800435', 'min: -0.021554852', 'var: 0.0026881096', 'std: 0.051846985']\n",
            "Epoch 7/15\n",
            "25000/25000 [==============================] - 272s 11ms/step - loss: 0.3199 - acc: 0.8681\n",
            "\n",
            "Stats for kernel:0  ['max: 0.98159856', 'mean: 0.4834195', 'min: 0.07355912', 'var: 0.05055383', 'std: 0.22484179']\n",
            "Stats for Sigma_current:0  ['max: 0.38395882', 'mean: 0.16492283', 'min: -0.004305011', 'var: 0.010059937', 'std: 0.10029924']\n",
            "Stats for Mu_current:0  ['max: 1.000151', 'mean: 0.5022124', 'min: 0.07615349', 'var: 0.043245763', 'std: 0.20795615']\n",
            "Stats for recurrent_kernel:0  ['max: 0.2342415', 'mean: 0.10574726', 'min: -0.021554852', 'var: 0.0028467583', 'std: 0.053355023']\n",
            "Epoch 8/15\n",
            "25000/25000 [==============================] - 280s 11ms/step - loss: 0.2936 - acc: 0.8792\n",
            "\n",
            "Stats for kernel:0  ['max: 0.9986464', 'mean: 0.48619598', 'min: 0.07345886', 'var: 0.051669084', 'std: 0.22730835']\n",
            "Stats for Sigma_current:0  ['max: 0.41710478', 'mean: 0.16436456', 'min: -0.004305011', 'var: 0.010371831', 'std: 0.10184219']\n",
            "Stats for Mu_current:0  ['max: 1.0005145', 'mean: 0.498861', 'min: 0.06200833', 'var: 0.04571209', 'std: 0.21380386']\n",
            "Stats for recurrent_kernel:0  ['max: 0.22295655', 'mean: 0.10542481', 'min: -0.021554852', 'var: 0.0027428279', 'std: 0.052372016']\n",
            "Epoch 9/15\n",
            "25000/25000 [==============================] - 279s 11ms/step - loss: 0.3251 - acc: 0.8609\n",
            "\n",
            "Stats for kernel:0  ['max: 1.0535724', 'mean: 0.48257312', 'min: 0.06238254', 'var: 0.050481595', 'std: 0.2246811']\n",
            "Stats for Sigma_current:0  ['max: 0.4427741', 'mean: 0.16811293', 'min: -0.004305011', 'var: 0.011764756', 'std: 0.10846546']\n",
            "Stats for Mu_current:0  ['max: 0.96343994', 'mean: 0.48834842', 'min: 0.03451605', 'var: 0.04749799', 'std: 0.21794035']\n",
            "Stats for recurrent_kernel:0  ['max: 0.2681524', 'mean: 0.11213988', 'min: -0.021554852', 'var: 0.0037294028', 'std: 0.061068837']\n",
            "Epoch 10/15\n",
            "25000/25000 [==============================] - 277s 11ms/step - loss: 0.4019 - acc: 0.8166\n",
            "\n",
            "Stats for kernel:0  ['max: 1.0235848', 'mean: 0.48686916', 'min: 0.066722296', 'var: 0.050008964', 'std: 0.22362684']\n",
            "Stats for Sigma_current:0  ['max: 0.46264753', 'mean: 0.17514092', 'min: -0.004305011', 'var: 0.013201522', 'std: 0.11489788']\n",
            "Stats for Mu_current:0  ['max: 0.93308437', 'mean: 0.48662966', 'min: 0.034263585', 'var: 0.04805481', 'std: 0.21921408']\n",
            "Stats for recurrent_kernel:0  ['max: 0.29869896', 'mean: 0.11891676', 'min: -0.021554852', 'var: 0.0043104384', 'std: 0.06565393']\n",
            "Epoch 11/15\n",
            "25000/25000 [==============================] - 276s 11ms/step - loss: 0.3402 - acc: 0.8536\n",
            "\n",
            "Stats for kernel:0  ['max: 1.0218872', 'mean: 0.48889327', 'min: 0.0626812', 'var: 0.050508805', 'std: 0.22474164']\n",
            "Stats for Sigma_current:0  ['max: 0.46309638', 'mean: 0.17890619', 'min: -0.004305011', 'var: 0.013625093', 'std: 0.11672657']\n",
            "Stats for Mu_current:0  ['max: 0.93298537', 'mean: 0.48723117', 'min: 0.034760218', 'var: 0.047724925', 'std: 0.21846035']\n",
            "Stats for recurrent_kernel:0  ['max: 0.29875073', 'mean: 0.118017', 'min: -0.021554852', 'var: 0.0044074324', 'std: 0.066388495']\n",
            "Epoch 12/15\n",
            "25000/25000 [==============================] - 277s 11ms/step - loss: 0.3419 - acc: 0.8508\n",
            "\n",
            "Stats for kernel:0  ['max: 1.0108092', 'mean: 0.48959094', 'min: 0.05914052', 'var: 0.049196992', 'std: 0.22180395']\n",
            "Stats for Sigma_current:0  ['max: 0.4903969', 'mean: 0.1821674', 'min: -0.004305011', 'var: 0.01438816', 'std: 0.11995066']\n",
            "Stats for Mu_current:0  ['max: 0.93288463', 'mean: 0.49041858', 'min: 0.037276465', 'var: 0.048125092', 'std: 0.21937431']\n",
            "Stats for recurrent_kernel:0  ['max: 0.29874355', 'mean: 0.11452988', 'min: -0.021554852', 'var: 0.0047602896', 'std: 0.06899485']\n",
            "Epoch 13/15\n",
            "25000/25000 [==============================] - 277s 11ms/step - loss: 0.3027 - acc: 0.8720\n",
            "\n",
            "Stats for kernel:0  ['max: 1.0171885', 'mean: 0.49320492', 'min: 0.06479117', 'var: 0.048881575', 'std: 0.22109178']\n",
            "Stats for Sigma_current:0  ['max: 0.49050003', 'mean: 0.18553215', 'min: -0.004305011', 'var: 0.015086797', 'std: 0.12282833']\n",
            "Stats for Mu_current:0  ['max: 0.93267083', 'mean: 0.4906312', 'min: 0.030701272', 'var: 0.048461407', 'std: 0.22013952']\n",
            "Stats for recurrent_kernel:0  ['max: 0.29877034', 'mean: 0.11509782', 'min: -0.021554852', 'var: 0.004814674', 'std: 0.06938785']\n",
            "Epoch 14/15\n",
            "25000/25000 [==============================] - 279s 11ms/step - loss: 0.3795 - acc: 0.8238\n",
            "\n",
            "Stats for kernel:0  ['max: 1.0180528', 'mean: 0.49090084', 'min: 0.053741716', 'var: 0.049230356', 'std: 0.22187914']\n",
            "Stats for Sigma_current:0  ['max: 0.50233537', 'mean: 0.1883446', 'min: -0.004305011', 'var: 0.016031887', 'std: 0.12661709']\n",
            "Stats for Mu_current:0  ['max: 0.9320525', 'mean: 0.4881631', 'min: 0.05346716', 'var: 0.047586992', 'std: 0.21814443']\n",
            "Stats for recurrent_kernel:0  ['max: 0.29835153', 'mean: 0.11202952', 'min: -0.021554852', 'var: 0.00499382', 'std: 0.07066697']\n",
            "Epoch 15/15\n",
            "25000/25000 [==============================] - 279s 11ms/step - loss: 0.3327 - acc: 0.8529\n",
            "\n",
            "Stats for kernel:0  ['max: 1.020686', 'mean: 0.49259284', 'min: 0.055791587', 'var: 0.04836682', 'std: 0.21992457']\n",
            "Stats for Sigma_current:0  ['max: 0.5066099', 'mean: 0.1939862', 'min: -0.004305011', 'var: 0.016754873', 'std: 0.12944062']\n",
            "Stats for Mu_current:0  ['max: 0.9311799', 'mean: 0.49019298', 'min: 0.04761992', 'var: 0.046916977', 'std: 0.21660326']\n",
            "Stats for recurrent_kernel:0  ['max: 0.2985315', 'mean: 0.11261297', 'min: -0.021554852', 'var: 0.0052559213', 'std: 0.07249773']\n",
            "Accuracy: 73.94%\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAekAAAEzCAYAAAAVa/veAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nMlQpMJZnVQAAAABJRU5ErkJggg==\n",
            "text/plain": [
              "<Figure size 576x360 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "u-Hat5eRbkhw",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "outputId": "ab8f0859-1851-4f3e-feaf-edbee6d069d5"
      },
      "source": [
        "dataset_percantage=100  #use dataset %10,%25,%50,%75\n",
        "K.clear_session()\n",
        "mod='focused'\n",
        "N=100\n",
        "all_lr=0.015\n",
        "sigma_lr=0.015\n",
        "mu_lr=0.015\n",
        "mom=0.9\n",
        "decay_dict_lr=0.1\n",
        "numpy.random.seed(7)\n",
        "# load the dataset but only keep the top n words, zero the rest\n",
        "sgd_settings=lr_settings(all_lr=all_lr,sigma_lr=sigma_lr,mu_lr=mu_lr,mom=mom,decay_dict_lr=decay_dict_lr)\n",
        "print(sgd_settings[0])\n",
        "history_he_uniform = build_model(N,mod,optimizer_s='AdamwithClip',num_epochs=15,sgd_settings=None,dataset_percantage=dataset_percantage,initer='He',distribution='uniform')#he uniform\n",
        "\n",
        "pd.DataFrame(history_he_uniform.history).plot(figsize=(8, 5))\n",
        "plt.grid(True)\n",
        "plt.gca().set_ylim(0, 1)\n",
        "plt.show()"
      ],
      "execution_count": 5,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "{'all': 0.015, 'focus-1/Sigma_current:0': 0.015, 'focus-1/Mu_current:0': 0.015, 'focus-1/kernel:0': 0.015, 'focus-1/Sigma_prev:0': 0.015, 'focus-1/Mu_prev:0': 0.015, 'focus-1/recurrent_kernel:0': 0.015, 'dense-3/Weights:0': 0.015}\n",
            "100\n",
            "input_shape (None, 32)\n",
            "self.input_dim=  32\n",
            "initilizer=  He\n",
            "distribution uniform\n",
            "u_current Tensor(\"focus-1/transpose_2:0\", shape=(32, 100), dtype=float32)\n",
            "u_previous Tensor(\"focus-1/transpose_3:0\", shape=(100, 100), dtype=float32)\n",
            "u_current Tensor(\"focus-1/while/transpose:0\", shape=(32, 100), dtype=float32)\n",
            "u_previous Tensor(\"focus-1/while/transpose_1:0\", shape=(100, 100), dtype=float32)\n",
            "opt=  <keras_utils.AdamwithClip object at 0x7fecf89412e8>\n",
            "Model: \"sequential_1\"\n",
            "_________________________________________________________________\n",
            "Layer (type)                 Output Shape              Param #   \n",
            "=================================================================\n",
            "embedding_1 (Embedding)      (None, 500, 32)           160000    \n",
            "_________________________________________________________________\n",
            "focus-1 (SimpleFocusedRNN)   (None, 100)               13700     \n",
            "_________________________________________________________________\n",
            "dense-3 (Dense)              (None, 1)                 101       \n",
            "=================================================================\n",
            "Total params: 173,801\n",
            "Trainable params: 173,801\n",
            "Non-trainable params: 0\n",
            "_________________________________________________________________\n",
            "None\n",
            "\n",
            "Stats for kernel:0  ['max: 0.8', 'mean: 0.49999997', 'min: 0.2', 'var: 0.030606063', 'std: 0.17494589']\n",
            "Stats for Sigma_current:0  ['max: 0.1', 'mean: 0.10000003', 'min: 0.1', 'var: 8.881784e-16', 'std: 2.9802322e-08']\n",
            "Stats for Mu_current:0  ['max: 0.8', 'mean: 0.49999997', 'min: 0.2', 'var: 0.030606063', 'std: 0.17494589']\n",
            "Stats for recurrent_kernel:0  ['max: 0.1', 'mean: 0.10000003', 'min: 0.1', 'var: 8.881784e-16', 'std: 2.9802322e-08']\n",
            "Epoch 1/15\n",
            "25000/25000 [==============================] - 275s 11ms/step - loss: 0.5877 - acc: 0.6813\n",
            "\n",
            "Stats for kernel:0  ['max: 0.8438556', 'mean: 0.49486908', 'min: 0.08246725', 'var: 0.03556036', 'std: 0.18857454']\n",
            "Stats for Sigma_current:0  ['max: 0.2334998', 'mean: 0.113284454', 'min: 0.0035543032', 'var: 0.0021073716', 'std: 0.045906115']\n",
            "Stats for Mu_current:0  ['max: 0.88119906', 'mean: 0.50424707', 'min: 0.08510638', 'var: 0.039569233', 'std: 0.19892016']\n",
            "Stats for recurrent_kernel:0  ['max: 0.18086737', 'mean: 0.10612996', 'min: 0.015822956', 'var: 0.00092264696', 'std: 0.030375104']\n",
            "Epoch 2/15\n",
            "25000/25000 [==============================] - 275s 11ms/step - loss: 0.4520 - acc: 0.7912\n",
            "\n",
            "Stats for kernel:0  ['max: 0.9087723', 'mean: 0.4971727', 'min: 0.017921379', 'var: 0.04324528', 'std: 0.20795499']\n",
            "Stats for Sigma_current:0  ['max: 0.29686216', 'mean: 0.12451372', 'min: 0.0017514634', 'var: 0.003813833', 'std: 0.06175624']\n",
            "Stats for Mu_current:0  ['max: 0.8999579', 'mean: 0.50477296', 'min: 0.04654378', 'var: 0.040214483', 'std: 0.20053549']\n",
            "Stats for recurrent_kernel:0  ['max: 0.18248548', 'mean: 0.10249527', 'min: 0.006662628', 'var: 0.0013658634', 'std: 0.036957588']\n",
            "Epoch 3/15\n",
            "25000/25000 [==============================] - 275s 11ms/step - loss: 0.4902 - acc: 0.7672\n",
            "\n",
            "Stats for kernel:0  ['max: 0.9215371', 'mean: 0.49240172', 'min: 0.035018895', 'var: 0.046667315', 'std: 0.21602619']\n",
            "Stats for Sigma_current:0  ['max: 0.3305998', 'mean: 0.12926032', 'min: 0.0017514634', 'var: 0.0049975063', 'std: 0.070693046']\n",
            "Stats for Mu_current:0  ['max: 0.91582865', 'mean: 0.50566983', 'min: 0.099609286', 'var: 0.043846335', 'std: 0.20939517']\n",
            "Stats for recurrent_kernel:0  ['max: 0.2099665', 'mean: 0.10550685', 'min: 0.006653425', 'var: 0.0019512757', 'std: 0.044173248']\n",
            "Epoch 4/15\n",
            "25000/25000 [==============================] - 275s 11ms/step - loss: 0.4418 - acc: 0.7937\n",
            "\n",
            "Stats for kernel:0  ['max: 0.93776405', 'mean: 0.49407294', 'min: 0.03097854', 'var: 0.047032174', 'std: 0.21686903']\n",
            "Stats for Sigma_current:0  ['max: 0.35288972', 'mean: 0.13224763', 'min: 0.0017514634', 'var: 0.00546159', 'std: 0.07390257']\n",
            "Stats for Mu_current:0  ['max: 0.91858995', 'mean: 0.5045837', 'min: 0.076306395', 'var: 0.044441525', 'std: 0.21081159']\n",
            "Stats for recurrent_kernel:0  ['max: 0.20882282', 'mean: 0.10444433', 'min: 0.006653425', 'var: 0.0019702392', 'std: 0.044387378']\n",
            "Epoch 5/15\n",
            "25000/25000 [==============================] - 276s 11ms/step - loss: 0.4305 - acc: 0.8008\n",
            "\n",
            "Stats for kernel:0  ['max: 0.92989', 'mean: 0.49580818', 'min: 0.020935021', 'var: 0.046698246', 'std: 0.21609777']\n",
            "Stats for Sigma_current:0  ['max: 0.3797127', 'mean: 0.13374227', 'min: 0.0017514634', 'var: 0.0058274455', 'std: 0.07633771']\n",
            "Stats for Mu_current:0  ['max: 0.9276004', 'mean: 0.5022761', 'min: 0.052975018', 'var: 0.046311185', 'std: 0.21520033']\n",
            "Stats for recurrent_kernel:0  ['max: 0.21274956', 'mean: 0.10663297', 'min: 0.006653425', 'var: 0.002243525', 'std: 0.047365863']\n",
            "Epoch 6/15\n",
            "25000/25000 [==============================] - 275s 11ms/step - loss: 0.3828 - acc: 0.8296\n",
            "\n",
            "Stats for kernel:0  ['max: 0.9684182', 'mean: 0.49511564', 'min: 0.026919499', 'var: 0.04705186', 'std: 0.2169144']\n",
            "Stats for Sigma_current:0  ['max: 0.3785186', 'mean: 0.13358423', 'min: 0.0017514634', 'var: 0.0057696495', 'std: 0.07595821']\n",
            "Stats for Mu_current:0  ['max: 0.92064124', 'mean: 0.50130826', 'min: 0.047412124', 'var: 0.04565515', 'std: 0.21367066']\n",
            "Stats for recurrent_kernel:0  ['max: 0.2128729', 'mean: 0.10502655', 'min: 0.0031365585', 'var: 0.0023152132', 'std: 0.04811666']\n",
            "Epoch 7/15\n",
            "25000/25000 [==============================] - 274s 11ms/step - loss: 0.3419 - acc: 0.8534\n",
            "\n",
            "Stats for kernel:0  ['max: 0.96613955', 'mean: 0.49352267', 'min: 0.020044016', 'var: 0.047135316', 'std: 0.21710669']\n",
            "Stats for Sigma_current:0  ['max: 0.38822955', 'mean: 0.1341', 'min: -0.000121421166', 'var: 0.0058794827', 'std: 0.076677784']\n",
            "Stats for Mu_current:0  ['max: 0.92083883', 'mean: 0.502809', 'min: 0.04255547', 'var: 0.045614783', 'std: 0.21357618']\n",
            "Stats for recurrent_kernel:0  ['max: 0.21206817', 'mean: 0.104778044', 'min: 0.0031365585', 'var: 0.0022722518', 'std: 0.04766814']\n",
            "Epoch 8/15\n",
            "25000/25000 [==============================] - 275s 11ms/step - loss: 0.3255 - acc: 0.8631\n",
            "\n",
            "Stats for kernel:0  ['max: 0.9778936', 'mean: 0.49507752', 'min: 0.022671754', 'var: 0.048121855', 'std: 0.21936694']\n",
            "Stats for Sigma_current:0  ['max: 0.39626098', 'mean: 0.13327664', 'min: -0.000121422076', 'var: 0.0059918696', 'std: 0.07740717']\n",
            "Stats for Mu_current:0  ['max: 0.93662', 'mean: 0.50297064', 'min: 0.027276821', 'var: 0.04624549', 'std: 0.21504764']\n",
            "Stats for recurrent_kernel:0  ['max: 0.20960651', 'mean: 0.10352902', 'min: 0.0031365585', 'var: 0.0022571364', 'std: 0.04750933']\n",
            "Epoch 9/15\n",
            "25000/25000 [==============================] - 276s 11ms/step - loss: 0.3331 - acc: 0.8563\n",
            "\n",
            "Stats for kernel:0  ['max: 0.98308223', 'mean: 0.4980019', 'min: 0.030207936', 'var: 0.047912415', 'std: 0.21888904']\n",
            "Stats for Sigma_current:0  ['max: 0.41607577', 'mean: 0.13361153', 'min: -0.000121422076', 'var: 0.006361442', 'std: 0.07975865']\n",
            "Stats for Mu_current:0  ['max: 0.9372294', 'mean: 0.5050275', 'min: 0.035747174', 'var: 0.04672774', 'std: 0.216166']\n",
            "Stats for recurrent_kernel:0  ['max: 0.20936242', 'mean: 0.102676876', 'min: 0.0031365585', 'var: 0.0021840893', 'std: 0.04673424']\n",
            "Epoch 10/15\n",
            "25000/25000 [==============================] - 276s 11ms/step - loss: 0.3241 - acc: 0.8624\n",
            "\n",
            "Stats for kernel:0  ['max: 0.97957486', 'mean: 0.49943912', 'min: -0.005720744', 'var: 0.04989943', 'std: 0.2233818']\n",
            "Stats for Sigma_current:0  ['max: 0.44310817', 'mean: 0.13500652', 'min: -0.020967113', 'var: 0.00688704', 'std: 0.082988195']\n",
            "Stats for Mu_current:0  ['max: 0.9483162', 'mean: 0.5043885', 'min: 0.03860851', 'var: 0.04648107', 'std: 0.2155947']\n",
            "Stats for recurrent_kernel:0  ['max: 0.20990925', 'mean: 0.1030313', 'min: 0.0031365585', 'var: 0.002183064', 'std: 0.04672327']\n",
            "Epoch 11/15\n",
            "25000/25000 [==============================] - 276s 11ms/step - loss: 0.2961 - acc: 0.8768\n",
            "\n",
            "Stats for kernel:0  ['max: 0.9766648', 'mean: 0.49885765', 'min: 0.002237285', 'var: 0.051332813', 'std: 0.22656746']\n",
            "Stats for Sigma_current:0  ['max: 0.46001136', 'mean: 0.13675177', 'min: -0.020967139', 'var: 0.0072761984', 'std: 0.08530063']\n",
            "Stats for Mu_current:0  ['max: 0.94725394', 'mean: 0.50434184', 'min: 0.027544688', 'var: 0.045451064', 'std: 0.21319255']\n",
            "Stats for recurrent_kernel:0  ['max: 0.20990637', 'mean: 0.102919176', 'min: 0.0031365585', 'var: 0.0022386909', 'std: 0.047314804']\n",
            "Epoch 12/15\n",
            "25000/25000 [==============================] - 276s 11ms/step - loss: 0.2799 - acc: 0.8850\n",
            "\n",
            "Stats for kernel:0  ['max: 0.977277', 'mean: 0.49610737', 'min: -0.024271782', 'var: 0.051542353', 'std: 0.22702941']\n",
            "Stats for Sigma_current:0  ['max: 0.44277227', 'mean: 0.13705894', 'min: -0.020967139', 'var: 0.0075424', 'std: 0.08684699']\n",
            "Stats for Mu_current:0  ['max: 0.934422', 'mean: 0.5049741', 'min: 0.0256201', 'var: 0.044125676', 'std: 0.21006112']\n",
            "Stats for recurrent_kernel:0  ['max: 0.2137871', 'mean: 0.10325695', 'min: 0.0031365585', 'var: 0.0023711498', 'std: 0.048694454']\n",
            "Epoch 13/15\n",
            "25000/25000 [==============================] - 282s 11ms/step - loss: 0.2705 - acc: 0.8890\n",
            "\n",
            "Stats for kernel:0  ['max: 0.98200357', 'mean: 0.49644563', 'min: -0.017252373', 'var: 0.050738867', 'std: 0.2252529']\n",
            "Stats for Sigma_current:0  ['max: 0.44122398', 'mean: 0.13718766', 'min: -0.020967139', 'var: 0.0075469436', 'std: 0.086873144']\n",
            "Stats for Mu_current:0  ['max: 0.95474607', 'mean: 0.50589776', 'min: 0.0076423544', 'var: 0.046095587', 'std: 0.21469884']\n",
            "Stats for recurrent_kernel:0  ['max: 0.21981838', 'mean: 0.102649644', 'min: 0.0031365585', 'var: 0.0023740842', 'std: 0.048724573']\n",
            "Epoch 14/15\n",
            "25000/25000 [==============================] - 279s 11ms/step - loss: 0.2497 - acc: 0.8990\n",
            "\n",
            "Stats for kernel:0  ['max: 0.97726846', 'mean: 0.49622124', 'min: -0.010974156', 'var: 0.05125627', 'std: 0.22639847']\n",
            "Stats for Sigma_current:0  ['max: 0.45991278', 'mean: 0.13786116', 'min: -0.020967139', 'var: 0.007789808', 'std: 0.08825989']\n",
            "Stats for Mu_current:0  ['max: 0.9612434', 'mean: 0.50583357', 'min: -0.0011488886', 'var: 0.04615387', 'std: 0.21483451']\n",
            "Stats for recurrent_kernel:0  ['max: 0.21533969', 'mean: 0.10108567', 'min: 0.0031365585', 'var: 0.002388822', 'std: 0.048875574']\n",
            "Epoch 15/15\n",
            "25000/25000 [==============================] - 276s 11ms/step - loss: 0.2501 - acc: 0.8985\n",
            "\n",
            "Stats for kernel:0  ['max: 0.9932712', 'mean: 0.49521744', 'min: -0.010867196', 'var: 0.051744036', 'std: 0.22747315']\n",
            "Stats for Sigma_current:0  ['max: 0.4722238', 'mean: 0.13919571', 'min: -0.020967139', 'var: 0.008149237', 'std: 0.09027313']\n",
            "Stats for Mu_current:0  ['max: 0.9672213', 'mean: 0.50418156', 'min: -0.01913051', 'var: 0.048051454', 'std: 0.21920642']\n",
            "Stats for recurrent_kernel:0  ['max: 0.23330264', 'mean: 0.10074044', 'min: 0.0031365585', 'var: 0.0025245876', 'std: 0.050245274']\n",
            "Accuracy: 75.81%\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAekAAAEzCAYAAAAVa/veAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nSUVORK5CYII=\n",
            "text/plain": [
              "<Figure size 576x360 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "7Srd4SnHboaY",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "outputId": "72189617-279e-4ca4-fb8f-54a242386d4e"
      },
      "source": [
        "dataset_percantage=100  #use dataset %10,%25,%50,%75\n",
        "K.clear_session()\n",
        "mod='focused'\n",
        "N=100\n",
        "all_lr=0.015\n",
        "sigma_lr=0.015\n",
        "mu_lr=0.015\n",
        "mom=0.9\n",
        "decay_dict_lr=0.1\n",
        "numpy.random.seed(7)\n",
        "# load the dataset but only keep the top n words, zero the rest\n",
        "sgd_settings=lr_settings(all_lr=all_lr,sigma_lr=sigma_lr,mu_lr=mu_lr,mom=mom,decay_dict_lr=decay_dict_lr)\n",
        "print(sgd_settings[0])\n",
        "history_he_normal = build_model(N,mod,optimizer_s='AdamwithClip',num_epochs=15,sgd_settings=None,dataset_percantage=dataset_percantage,initer='He',distribution='normal')#he normal\n",
        "pd.DataFrame(history_he_normal.history).plot(figsize=(8, 5))\n",
        "plt.grid(True)\n",
        "plt.gca().set_ylim(0, 1)\n",
        "plt.show()"
      ],
      "execution_count": 4,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:107: The name tf.reset_default_graph is deprecated. Please use tf.compat.v1.reset_default_graph instead.\n",
            "\n",
            "WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:111: The name tf.placeholder_with_default is deprecated. Please use tf.compat.v1.placeholder_with_default instead.\n",
            "\n",
            "WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:66: The name tf.get_default_graph is deprecated. Please use tf.compat.v1.get_default_graph instead.\n",
            "\n",
            "{'all': 0.015, 'focus-1/Sigma_current:0': 0.015, 'focus-1/Mu_current:0': 0.015, 'focus-1/kernel:0': 0.015, 'focus-1/Sigma_prev:0': 0.015, 'focus-1/Mu_prev:0': 0.015, 'focus-1/recurrent_kernel:0': 0.015, 'dense-3/Weights:0': 0.015}\n",
            "Downloading data from https://s3.amazonaws.com/text-datasets/imdb.npz\n",
            "17465344/17464789 [==============================] - 1s 0us/step\n",
            "100\n",
            "WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:541: The name tf.placeholder is deprecated. Please use tf.compat.v1.placeholder instead.\n",
            "\n",
            "WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:4432: The name tf.random_uniform is deprecated. Please use tf.random.uniform instead.\n",
            "\n",
            "input_shape (None, 32)\n",
            "self.input_dim=  32\n",
            "initilizer=  He\n",
            "distribution normal\n",
            "WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:190: The name tf.get_default_session is deprecated. Please use tf.compat.v1.get_default_session instead.\n",
            "\n",
            "WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:197: The name tf.ConfigProto is deprecated. Please use tf.compat.v1.ConfigProto instead.\n",
            "\n",
            "WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:203: The name tf.Session is deprecated. Please use tf.compat.v1.Session instead.\n",
            "\n",
            "WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:207: The name tf.global_variables is deprecated. Please use tf.compat.v1.global_variables instead.\n",
            "\n",
            "WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:216: The name tf.is_variable_initialized is deprecated. Please use tf.compat.v1.is_variable_initialized instead.\n",
            "\n",
            "WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:223: The name tf.variables_initializer is deprecated. Please use tf.compat.v1.variables_initializer instead.\n",
            "\n",
            "u_current Tensor(\"focus-1/transpose_2:0\", shape=(32, 100), dtype=float32)\n",
            "u_previous Tensor(\"focus-1/transpose_3:0\", shape=(100, 100), dtype=float32)\n",
            "WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:3733: calling dropout (from tensorflow.python.ops.nn_ops) with keep_prob is deprecated and will be removed in a future version.\n",
            "Instructions for updating:\n",
            "Please use `rate` instead of `keep_prob`. Rate should be set to `rate = 1 - keep_prob`.\n",
            "u_current Tensor(\"focus-1/while/transpose:0\", shape=(32, 100), dtype=float32)\n",
            "u_previous Tensor(\"focus-1/while/transpose_1:0\", shape=(100, 100), dtype=float32)\n",
            "opt=  <keras_utils.AdamwithClip object at 0x7feceee13c88>\n",
            "WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/keras/optimizers.py:793: The name tf.train.Optimizer is deprecated. Please use tf.compat.v1.train.Optimizer instead.\n",
            "\n",
            "WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:3657: The name tf.log is deprecated. Please use tf.math.log instead.\n",
            "\n",
            "WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/tensorflow_core/python/ops/nn_impl.py:183: where (from tensorflow.python.ops.array_ops) is deprecated and will be removed in a future version.\n",
            "Instructions for updating:\n",
            "Use tf.where in 2.0, which has the same broadcast rule as np.where\n",
            "Model: \"sequential_1\"\n",
            "_________________________________________________________________\n",
            "Layer (type)                 Output Shape              Param #   \n",
            "=================================================================\n",
            "embedding_1 (Embedding)      (None, 500, 32)           160000    \n",
            "_________________________________________________________________\n",
            "focus-1 (SimpleFocusedRNN)   (None, 100)               13700     \n",
            "_________________________________________________________________\n",
            "dense-3 (Dense)              (None, 1)                 101       \n",
            "=================================================================\n",
            "Total params: 173,801\n",
            "Trainable params: 173,801\n",
            "Non-trainable params: 0\n",
            "_________________________________________________________________\n",
            "None\n",
            "WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:1033: The name tf.assign_add is deprecated. Please use tf.compat.v1.assign_add instead.\n",
            "\n",
            "WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:1020: The name tf.assign is deprecated. Please use tf.compat.v1.assign instead.\n",
            "\n",
            "\n",
            "Stats for kernel:0  ['max: 0.8', 'mean: 0.49999997', 'min: 0.2', 'var: 0.030606063', 'std: 0.17494589']\n",
            "Stats for Sigma_current:0  ['max: 0.1', 'mean: 0.10000003', 'min: 0.1', 'var: 8.881784e-16', 'std: 2.9802322e-08']\n",
            "Stats for Mu_current:0  ['max: 0.8', 'mean: 0.49999997', 'min: 0.2', 'var: 0.030606063', 'std: 0.17494589']\n",
            "Stats for recurrent_kernel:0  ['max: 0.1', 'mean: 0.10000003', 'min: 0.1', 'var: 8.881784e-16', 'std: 2.9802322e-08']\n",
            "Epoch 1/15\n",
            "25000/25000 [==============================] - 291s 12ms/step - loss: 0.5688 - acc: 0.6917\n",
            "\n",
            "Stats for kernel:0  ['max: 0.84115976', 'mean: 0.49634457', 'min: 0.11973767', 'var: 0.038962487', 'std: 0.19738917']\n",
            "Stats for Sigma_current:0  ['max: 0.22486627', 'mean: 0.1091297', 'min: 0.0032639462', 'var: 0.0019267931', 'std: 0.043895252']\n",
            "Stats for Mu_current:0  ['max: 0.83992577', 'mean: 0.4970259', 'min: 0.16040859', 'var: 0.03662652', 'std: 0.19138056']\n",
            "Stats for recurrent_kernel:0  ['max: 0.18852101', 'mean: 0.10208238', 'min: 0.05433199', 'var: 0.0006999727', 'std: 0.026456997']\n",
            "Epoch 2/15\n",
            "25000/25000 [==============================] - 292s 12ms/step - loss: 0.4412 - acc: 0.8048\n",
            "\n",
            "Stats for kernel:0  ['max: 0.8617567', 'mean: 0.49878865', 'min: 0.034795374', 'var: 0.04482024', 'std: 0.21170792']\n",
            "Stats for Sigma_current:0  ['max: 0.29740554', 'mean: 0.111368306', 'min: -0.004908886', 'var: 0.0040765577', 'std: 0.06384793']\n",
            "Stats for Mu_current:0  ['max: 0.85789084', 'mean: 0.4959989', 'min: 0.10506842', 'var: 0.037843056', 'std: 0.19453292']\n",
            "Stats for recurrent_kernel:0  ['max: 0.20037712', 'mean: 0.10213228', 'min: -0.013907369', 'var: 0.0013901953', 'std: 0.037285324']\n",
            "Epoch 3/15\n",
            "25000/25000 [==============================] - 290s 12ms/step - loss: 0.4423 - acc: 0.7925\n",
            "\n",
            "Stats for kernel:0  ['max: 0.92564374', 'mean: 0.5000183', 'min: 0.053244222', 'var: 0.046789363', 'std: 0.21630849']\n",
            "Stats for Sigma_current:0  ['max: 0.31867468', 'mean: 0.11861017', 'min: -0.004908886', 'var: 0.0052539897', 'std: 0.07248441']\n",
            "Stats for Mu_current:0  ['max: 0.880088', 'mean: 0.4961006', 'min: 0.017231818', 'var: 0.04268144', 'std: 0.20659487']\n",
            "Stats for recurrent_kernel:0  ['max: 0.2098337', 'mean: 0.101472944', 'min: -0.02180077', 'var: 0.0018111777', 'std: 0.042557932']\n",
            "Epoch 4/15\n",
            "25000/25000 [==============================] - 286s 11ms/step - loss: 0.4438 - acc: 0.7952\n",
            "\n",
            "Stats for kernel:0  ['max: 0.95929945', 'mean: 0.5056312', 'min: 0.011580548', 'var: 0.051098842', 'std: 0.22605053']\n",
            "Stats for Sigma_current:0  ['max: 0.33999836', 'mean: 0.12376032', 'min: -0.004908886', 'var: 0.0066867005', 'std: 0.081772245']\n",
            "Stats for Mu_current:0  ['max: 0.88461614', 'mean: 0.48783183', 'min: 0.024875835', 'var: 0.04474447', 'std: 0.21152888']\n",
            "Stats for recurrent_kernel:0  ['max: 0.20433077', 'mean: 0.10358091', 'min: -0.021801326', 'var: 0.0023114993', 'std: 0.048078053']\n",
            "Epoch 5/15\n",
            "25000/25000 [==============================] - 287s 11ms/step - loss: 0.4180 - acc: 0.8080\n",
            "\n",
            "Stats for kernel:0  ['max: 0.9487046', 'mean: 0.5044799', 'min: 0.023550205', 'var: 0.052210797', 'std: 0.22849682']\n",
            "Stats for Sigma_current:0  ['max: 0.37138095', 'mean: 0.1259129', 'min: -0.004908886', 'var: 0.008243061', 'std: 0.0907913']\n",
            "Stats for Mu_current:0  ['max: 0.85945404', 'mean: 0.48543906', 'min: 0.024088532', 'var: 0.044634625', 'std: 0.21126908']\n",
            "Stats for recurrent_kernel:0  ['max: 0.21106917', 'mean: 0.10362139', 'min: -0.021801326', 'var: 0.0025451735', 'std: 0.050449714']\n",
            "Epoch 6/15\n",
            "25000/25000 [==============================] - 289s 12ms/step - loss: 0.3643 - acc: 0.8402\n",
            "\n",
            "Stats for kernel:0  ['max: 0.9449245', 'mean: 0.5010477', 'min: 0.016257856', 'var: 0.052903924', 'std: 0.23000853']\n",
            "Stats for Sigma_current:0  ['max: 0.37783232', 'mean: 0.13148098', 'min: -0.004908886', 'var: 0.008818561', 'std: 0.09390719']\n",
            "Stats for Mu_current:0  ['max: 0.88223165', 'mean: 0.49205205', 'min: 0.024799062', 'var: 0.045284204', 'std: 0.21280086']\n",
            "Stats for recurrent_kernel:0  ['max: 0.21003675', 'mean: 0.10394312', 'min: -0.021801326', 'var: 0.002551458', 'std: 0.05051196']\n",
            "Epoch 7/15\n",
            "25000/25000 [==============================] - 289s 12ms/step - loss: 0.3311 - acc: 0.8626\n",
            "\n",
            "Stats for kernel:0  ['max: 0.9467387', 'mean: 0.50018334', 'min: 0.023001418', 'var: 0.053708535', 'std: 0.23175102']\n",
            "Stats for Sigma_current:0  ['max: 0.40094516', 'mean: 0.13606194', 'min: -0.004908886', 'var: 0.009958921', 'std: 0.099794395']\n",
            "Stats for Mu_current:0  ['max: 0.8978232', 'mean: 0.48786464', 'min: 0.022006864', 'var: 0.04410432', 'std: 0.21001029']\n",
            "Stats for recurrent_kernel:0  ['max: 0.2345096', 'mean: 0.10637414', 'min: -0.021801326', 'var: 0.00282131', 'std: 0.053116005']\n",
            "Epoch 8/15\n",
            "25000/25000 [==============================] - 289s 12ms/step - loss: 0.3003 - acc: 0.8759\n",
            "\n",
            "Stats for kernel:0  ['max: 1.0073931', 'mean: 0.5018354', 'min: 0.004224158', 'var: 0.055328522', 'std: 0.23522015']\n",
            "Stats for Sigma_current:0  ['max: 0.41502097', 'mean: 0.13460465', 'min: -0.004908886', 'var: 0.010504147', 'std: 0.10248974']\n",
            "Stats for Mu_current:0  ['max: 0.8987854', 'mean: 0.4896338', 'min: 0.0015847642', 'var: 0.044373084', 'std: 0.21064919']\n",
            "Stats for recurrent_kernel:0  ['max: 0.26080352', 'mean: 0.10686879', 'min: -0.021801326', 'var: 0.0029685425', 'std: 0.054484334']\n",
            "Epoch 9/15\n",
            "25000/25000 [==============================] - 289s 12ms/step - loss: 0.2834 - acc: 0.8860\n",
            "\n",
            "Stats for kernel:0  ['max: 0.9905583', 'mean: 0.5069699', 'min: -0.00061585306', 'var: 0.0558782', 'std: 0.2363857']\n",
            "Stats for Sigma_current:0  ['max: 0.41609812', 'mean: 0.13728774', 'min: -0.004908886', 'var: 0.011249696', 'std: 0.10606458']\n",
            "Stats for Mu_current:0  ['max: 0.9309181', 'mean: 0.49514875', 'min: -0.026215095', 'var: 0.04463736', 'std: 0.21127555']\n",
            "Stats for recurrent_kernel:0  ['max: 0.31222308', 'mean: 0.110267915', 'min: -0.021801326', 'var: 0.0035568762', 'std: 0.05963955']\n",
            "Epoch 10/15\n",
            "25000/25000 [==============================] - 290s 12ms/step - loss: 0.3104 - acc: 0.8710\n",
            "\n",
            "Stats for kernel:0  ['max: 0.9803638', 'mean: 0.5059992', 'min: -0.013631255', 'var: 0.05680215', 'std: 0.23833202']\n",
            "Stats for Sigma_current:0  ['max: 0.44195357', 'mean: 0.1425747', 'min: -0.004908886', 'var: 0.012568073', 'std: 0.11210742']\n",
            "Stats for Mu_current:0  ['max: 0.9333593', 'mean: 0.49477473', 'min: -0.029694784', 'var: 0.047765322', 'std: 0.21855278']\n",
            "Stats for recurrent_kernel:0  ['max: 0.27320272', 'mean: 0.11284702', 'min: -0.021801326', 'var: 0.003512889', 'std: 0.05926963']\n",
            "Epoch 11/15\n",
            "25000/25000 [==============================] - 277s 11ms/step - loss: 0.2945 - acc: 0.8784\n",
            "\n",
            "Stats for kernel:0  ['max: 0.98831314', 'mean: 0.50672513', 'min: -0.0030639744', 'var: 0.05630022', 'std: 0.23727667']\n",
            "Stats for Sigma_current:0  ['max: 0.47437188', 'mean: 0.14435795', 'min: -0.004908886', 'var: 0.012912666', 'std: 0.113633916']\n",
            "Stats for Mu_current:0  ['max: 0.9368672', 'mean: 0.5011066', 'min: -0.031442154', 'var: 0.04558034', 'std: 0.21349552']\n",
            "Stats for recurrent_kernel:0  ['max: 0.27347136', 'mean: 0.11236055', 'min: -0.021801326', 'var: 0.0037248302', 'std: 0.061031386']\n",
            "Epoch 12/15\n",
            "25000/25000 [==============================] - 272s 11ms/step - loss: 0.3238 - acc: 0.8639\n",
            "\n",
            "Stats for kernel:0  ['max: 0.99561906', 'mean: 0.504957', 'min: -0.018645886', 'var: 0.057147168', 'std: 0.23905474']\n",
            "Stats for Sigma_current:0  ['max: 0.49208733', 'mean: 0.14731772', 'min: -0.004908886', 'var: 0.0140559', 'std: 0.11855758']\n",
            "Stats for Mu_current:0  ['max: 0.9160967', 'mean: 0.49686718', 'min: -0.031065715', 'var: 0.047404643', 'std: 0.21772607']\n",
            "Stats for recurrent_kernel:0  ['max: 0.274081', 'mean: 0.11397127', 'min: -0.021801326', 'var: 0.0038744363', 'std: 0.06224497']\n",
            "Epoch 13/15\n",
            "25000/25000 [==============================] - 272s 11ms/step - loss: 0.2769 - acc: 0.8844\n",
            "\n",
            "Stats for kernel:0  ['max: 1.0019537', 'mean: 0.5095344', 'min: -0.0302964', 'var: 0.05513159', 'std: 0.23480117']\n",
            "Stats for Sigma_current:0  ['max: 0.52548015', 'mean: 0.14713554', 'min: -0.004908886', 'var: 0.014583383', 'std: 0.12076168']\n",
            "Stats for Mu_current:0  ['max: 0.95610905', 'mean: 0.5012768', 'min: -0.030636098', 'var: 0.049917534', 'std: 0.22342232']\n",
            "Stats for recurrent_kernel:0  ['max: 0.30501005', 'mean: 0.115549155', 'min: -0.021801326', 'var: 0.0042308057', 'std: 0.06504464']\n",
            "Epoch 14/15\n",
            "25000/25000 [==============================] - 273s 11ms/step - loss: 0.2707 - acc: 0.8854\n",
            "\n",
            "Stats for kernel:0  ['max: 0.9999124', 'mean: 0.50579935', 'min: -0.019821482', 'var: 0.055809993', 'std: 0.23624139']\n",
            "Stats for Sigma_current:0  ['max: 0.5455096', 'mean: 0.14921565', 'min: -0.004908886', 'var: 0.015230842', 'std: 0.123413295']\n",
            "Stats for Mu_current:0  ['max: 1.0261395', 'mean: 0.50101084', 'min: -0.030591596', 'var: 0.05090475', 'std: 0.2256208']\n",
            "Stats for recurrent_kernel:0  ['max: 0.30199066', 'mean: 0.115038194', 'min: -0.021801326', 'var: 0.004389538', 'std: 0.06625359']\n",
            "Epoch 15/15\n",
            "25000/25000 [==============================] - 272s 11ms/step - loss: 0.2399 - acc: 0.9033\n",
            "\n",
            "Stats for kernel:0  ['max: 1.0144588', 'mean: 0.509264', 'min: -0.013364955', 'var: 0.056159772', 'std: 0.23698053']\n",
            "Stats for Sigma_current:0  ['max: 0.53213525', 'mean: 0.14699395', 'min: -0.004908886', 'var: 0.015114566', 'std: 0.122941315']\n",
            "Stats for Mu_current:0  ['max: 1.032362', 'mean: 0.49979228', 'min: -0.030567912', 'var: 0.05242424', 'std: 0.2289634']\n",
            "Stats for recurrent_kernel:0  ['max: 0.31268245', 'mean: 0.114367545', 'min: -0.021801326', 'var: 0.004302565', 'std: 0.06559394']\n",
            "Accuracy: 75.58%\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAekAAAEzCAYAAAAVa/veAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nIz5HPcy0UDMAAAAASUVORK5CYII=\n",
            "text/plain": [
              "<Figure size 576x360 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    }
  ]
}