run_ml1_new.sh

#!/bin/bash
   # Script to reproduce results

 Foldername="0723_offline_meta_rl_dasdas"
 mkdir out_logs/${Foldername} &> /dev/null
 declare -a tasks=( "cpearl-ml1" )
 declare -a algos=( "ensemble" )
 ##
 declare -a seeds=( "1" "2" )
 declare -a datadirs=( "hammer-v2")
 declare -a is_sparses=( "0" )
 declare -a use_bracs=( "1")
 declare -a use_information_bottlenecks=( "0" )
 declare -a is_zlosses=( "1" )
 declare -a is_onlineadapt_threses=( "0" )
 declare -a is_onlineadapt_maxes=( "1" )
 declare -a num_exp_traj_evals=( "5" )
 declare -a allow_backward_zs=( "0" )
 declare -a is_true_sparses=( "0" )
 declare -a r_threses=( "-1000000" )
 n=6
 gpunum=8
 for task in "${tasks[@]}"
 do
 for algo in "${algos[@]}"
 do
 for seed in "${seeds[@]}"
 do
 for datadir in "${datadirs[@]}"
 do
 for is_sparse in "${is_sparses[@]}"
 do
 for use_brac in "${use_bracs[@]}"
 do
 for use_information_bottleneck in "${use_information_bottlenecks[@]}"
 do
 for is_zloss in "${is_zlosses[@]}"
 do
 for is_onlineadapt_thres in "${is_onlineadapt_threses[@]}"
 do
 for is_onlineadapt_max in "${is_onlineadapt_maxes[@]}"
 do
 for num_exp_traj_eval in "${num_exp_traj_evals[@]}"
 do
 for allow_backward_z in "${allow_backward_zs[@]}"
 do
 for is_true_sparse in "${is_true_sparses[@]}"
 do
 for r_thres in "${r_threses[@]}"
 do
 OMP_NUM_THREADS=16 KMP_AFFINITY="compact,granularity\=fine" nohup python launch_experiment_${algo}.py \
 ./configs/${task}.json \
 ./data/${datadir} \
 --gpu=${n} \
 --is_sparse_reward=${is_sparse} \
 --use_brac=${use_brac} \
 --use_information_bottleneck=${use_information_bottleneck} \
 --is_zloss=${is_zloss} \
 --is_onlineadapt_thres=${is_onlineadapt_thres} \
 --is_onlineadapt_max=${is_onlineadapt_max} \
 --num_exp_traj_eval=${num_exp_traj_eval} \
 --allow_backward_z=${allow_backward_z} \
 --is_true_sparse_rewards=${is_true_sparse} \
 --r_thres=${r_thres} \
 >& out_logs/${Foldername}/${task}_${algo}_${datadir}_${is_sparse}_${use_brac}_${use_information_bottleneck}_${is_zloss}_${is_onlineadapt_thres}_${is_onlineadapt_max}_${num_exp_traj_eval}_${allow_backward_z}_${is_true_sparse}_${r_thres}_${seed}_${n}.txt &
 echo "task: ${task}, algo: ${algo}, datadir: ${datadir}, is_sparse: ${is_sparse}, use_brac: ${use_brac}"
 echo "     use_information_bottleneck: ${use_information_bottleneck}, is_zloss: ${is_zloss}"
 echo "     is_onlineadapt_thres: ${is_onlineadapt_thres}, is_onlineadapt_max: ${is_onlineadapt_max}"
 echo "     num_exp_traj_eval: ${num_exp_traj_eval}, allow_backward_z: ${allow_backward_z}"
 echo "     is_true_sparse: ${is_true_sparse}, r_thres: ${r_thres}, seed: ${seed}, GPU: $n"
 n=$[($n+1) % ${gpunum}]
 sleep 10
 done
 done
 done
 done
 done
 done
 done
 done
 done
 done
 done
 done
 done
 done

# data collection: python data_collection_ml1.py ./configs/ml1.json
# To change to other environments, modify "env_name" in ./configs/ml1.json.

# Training: bash run_ml1.sh.
# To change to other environments, modify "datadirs" in line 10, as well as "env_name" in ./configs/cpeal-ml1.json.

# Plot training curves: See plot_new_ml1.py.