Add walker

readme.md

@@ -0,0 +1,5 @@
+# Setup
+
+Install `conda env create --name c_mbrl --file environment.yml`
+
+# Running experiments

walker_500k.sh

@@ -0,0 +1,75 @@
+#!/bin/bash
+
+#SBATCH --array=14-239
+#SBATCH --partition alldlc_gpu-rtx2080
+#SBATCH --job-name CMbRL_odin
+#SBATCH --output logs/slurm/%x-%A-%a-HelloCluster.out
+#SBATCH --error logs/slurm/%x-%A-%a-HelloCluster.err
+#SBATCH --mem 32GB
+#SBATCH --ntasks=1
+#SBATCH --cpus-per-task=32
+#SBATCH --ntasks-per-core=1
+#SBATCH --gres=gpu:1
+
+
+echo "Workingdir: $PWD";
+echo "Started at $(date)";
+
+source ~/miniconda3/bin/activate # Adjust to your path of Miniconda installation
+conda activate c_mbrl
+
+start=`date +%s`
+
+tasks=("carl_dmc_walker")
+seeds=("0" "42" "1337" "13" "71" "1994" "1997" "908" "2102" "3")
+schemes=("enc_obs_dec_obs_default" "enc_img_dec_img_default" "enc_obs_dec_obs" "enc_img_dec_img" "enc_obs_ctx_dec_obs_ctx" "enc_img_ctx_dec_img_ctx" "enc_obs_dec_obs_pgm_ctx" "enc_img_dec_img_pgm_ctx")
+contexts=("single_0" "single_1" "double_box")
+
+n_tasks=${#tasks[@]}
+n_seeds=${#seeds[@]}
+n_schemes=${#schemes[@]}
+n_contexts=${#contexts[@]}
+
+task_index=$((${SLURM_ARRAY_TASK_ID} / (n_seeds * n_schemes * n_contexts) % n_tasks))
+seed_index=$((${SLURM_ARRAY_TASK_ID} / (n_schemes * n_contexts) % n_seeds))
+scheme_index=$((${SLURM_ARRAY_TASK_ID} / n_contexts % n_schemes))
+context_index=$((${SLURM_ARRAY_TASK_ID} % n_contexts))
+
+task=${tasks[$task_index]}
+seed=${seeds[$seed_index]}
+scheme=${schemes[$scheme_index]}
+context=${contexts[$context_index]}
+
+group_name="${task}_${context}_${scheme}_500k"
+
+if [ "$scheme" == "enc_obs_dec_obs_default" ]; then
+    scheme="enc_obs_dec_obs"
+    context="default"
+elif [ "$scheme" == "enc_img_dec_img_default" ]; then
+    scheme="enc_img_dec_img"
+    context="default"
+fi
+
+# start training only if logdir does not exist
+# if training dir doesn't exist copy "${task}_${context}_${scheme}_normalized" to the new logdir
+
+if [ ! -d logs/$group_name/ ]; then
+    mkdir -p logs/$group_name
+fi
+
+if [ ! -d logs/$group_name/$seed ]; then
+    python -m contextual_mbrl.dreamer.train --configs carl $scheme --task $task --env.carl.context $context --seed $seed --logdir logs/$group_name/$seed --wandb.group $group_name --jax.policy_devices 0 --jax.train_devices 0 --run.steps 500000 --wandb.project ''
+fi
+
+# if eval.jsonl doesn't exist or doesn't have 84 lines
+if [ ! -f logs/$group_name/$seed/eval.jsonl ] || [ $(wc -l < logs/$group_name/$seed/eval.jsonl) -ne 84 ]; then
+    rm logs/$group_name/$seed/eval.jsonl
+    python -m contextual_mbrl.dreamer.eval --logdir logs/$group_name/$seed
+fi
+
+
+end=`date +%s`
+runtime=$((end-start))
+
+echo Job execution complete.
+echo Runtime: $runtime