DeepMind Lab

Installation

Installation DeepMind Lab can be time consuming. If you are on a Linux system, we provide a prebuilt wheel.

• Either pip install deepmind_lab-1.0-py3-none-any.whl
• Or alternatively, DMLab can be compiled from source by following the instructions on the DMLab Github.
• pip install dm_env
• To train on DMLab-30 you will need brady_konkle_oliva2008 dataset.
• To significantly speed up training on DMLab-30 consider downloading our dataset of pre-generated environment layouts (see paper for details). Command lines for running experiments with these datasets are provided in the sections below.

Running Experiments

Run DMLab experiments with the scripts in sf_examples.dmlab.

Example of training in the DMLab watermaze environment for 1B environment steps

python -m sf_examples.dmlab.train_dmlab --env=dmlab_watermaze --train_for_env_steps=1000000000 --gamma=0.99 --use_rnn=True --num_workers=90 --num_envs_per_worker=12 --num_epochs=1 --rollout=32 --recurrence=32 --batch_size=2048 --benchmark=False --max_grad_norm=0.0 --dmlab_renderer=software --decorrelate_experience_max_seconds=120 --encoder_conv_architecture=resnet_impala --encoder_conv_mlp_layers=512 --nonlinearity=relu --rnn_type=lstm --dmlab_extended_action_set=True --num_policies=1 --experiment=dmlab_watermaze_resnet_w90_v12 --set_workers_cpu_affinity=True --max_policy_lag=35  --dmlab30_dataset=~/datasets/brady_konkle_oliva2008 --dmlab_use_level_cache=True --dmlab_level_cache_path=/home/user/dmlab_cache

DMLab-30 run on a 36-core server with 4 GPUs using Population-Based Training with 4 agents:

python -m sf_examples.dmlab.train_dmlab --env=dmlab_30 --train_for_env_steps=10000000000 --gamma=0.99 --use_rnn=True --num_workers=90 --num_envs_per_worker=12 --num_epochs=1 --rollout=32 --recurrence=32 --batch_size=2048 --benchmark=False --max_grad_norm=0.0 --dmlab_renderer=software --decorrelate_experience_max_seconds=120 --encoder_conv_architecture=resnet_impala --encoder_conv_mlp_layers=512 --nonlinearity=relu --rnn_type=lstm --dmlab_extended_action_set=True --num_policies=4 --pbt_replace_reward_gap=0.05 --pbt_replace_reward_gap_absolute=5.0 --pbt_period_env_steps=10000000 --pbt_start_mutation=100000000 --with_pbt=True --experiment=dmlab_30_resnet_4pbt_w90_v12 --dmlab_one_task_per_worker=True --set_workers_cpu_affinity=True --max_policy_lag=35 --pbt_target_objective=dmlab_target_objective --dmlab30_dataset=~/datasets/brady_konkle_oliva2008 --dmlab_use_level_cache=True --dmlab_level_cache_path=/home/user/dmlab_cache

Models

DMLab Command Line Parameter	DMLab Environment name	Model Checkpooints
dmlab_30	DMLab-30	🤗 Hub DMLab30 checkpoints