[Algo] Added sac codebase

examples/config/sac.yaml

@@ -0,0 +1,46 @@
+default:
+    - override hydra/output: local
+    - override hydra/launcher: local
+
+# Logger
+exp_name: ${task}_sac_${visual_transform}
+visual_transform: r3m
+record_interval: 1
+device: "cuda:0"
+device_collection: "cuda:1"
+wandb_entity: "RLHive"
+wandb_mode: "offline"
+
+# Environment
+task: visual_franka_slide_random-v3
+#frame_skip: 1
+reward_scaling: 5.0
+init_env_steps: 1000
+seed: 42
+eval_traj: 25
+eval_interval: 1000
+num_envs: 8
+
+# Collector
+env_per_collector: 1
+max_frames_per_traj: -1
+total_frames: 1000000
+init_random_frames: 25000
+frames_per_batch: 1000
+
+# Replay Buffer
+prb: 0
+buffer_size: 100000
+buffer_scratch_dir: /tmp/
+
+# Optimization
+gamma: 0.99
+batch_size: 256
+lr: 3.0e-4
+weight_decay: 0.0
+target_update_polyak: 0.995
+utd_ratio: 1
+
+hydra:
+    job:
+        name: sac_${task}_${seed}

examples/install/install_rlhive.sh

@@ -0,0 +1,41 @@
+#!/bin/zsh
+
+# Instructions to install a fresh anaconda environment with RLHive
+
+set -e
+
+conda_path=$(conda info | grep -i 'base environment' | awk '{ print $4 }')
+source $conda_path/etc/profile.d/conda.sh
+
+here=$(pwd)
+module_path=$HOME/modules/
+
+module purge
+module load cuda/11.6
+
+conda env remove -n rlhive -y
+
+conda create -n rlhive -y python=3.8
+
+conda activate rlhive
+
+python3 -mpip install --pre torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/nightly/cu116
+
+mkdir $module_path
+cd $module_path
+git clone -c submodule.mj_envs/sims/neuromuscular_sim.update=none --branch v0.4dev --recursive https://github.com/vikashplus/mj_envs.git mj_envs
+cd mj_envs
+python3 -mpip install .  # one can also install it locally with the -e flag
+cd $here
+
+python3 -mpip install git+https://github.com/pytorch-labs/tensordict  # or stable or nightly with pip install tensordict(-nightly)
+python3 -mpip install git+https://github.com/pytorch/rl.git  # or stable or nightly with pip install torchrl(-nightly)
+
+# this
+# python3 -mpip install git+https://github.com/facebookresearch/rlhive.git  # or stable or nightly with pip install torchrl(-nightly)
+# or this
+cd ../..
+pip install -e .
+cd $here
+
+pip install wandb tqdm hydra-core moviepy