ml-lab / neurips-18-prosthetics-challenge

NeurIPS 2018: AI for Prosthetics Challenge – 3rd place solution

Attention

git clone --recursive [email protected]:scitator/neurips-18-prosthetics-challenge.git

How2run

1. System requirements – redis and Anaconda

   sudo apt install redis-server

2. Python requirements

   conda create -n opensim-rl -c kidzik opensim python=3.6.1
   activate opensim-rl
   source activate opensim-rl
   conda install -c conda-forge lapack git
   conda install pytorch torchvision -c pytorch
   pip install git+https://github.com/stanfordnmbl/osim-rl.git
   pip install git+https://github.com/pytorch/tnt.git@master
   pip install pyaml tensorboardX jpeg4py albumentations    

3. Mujoco example

   For installation issues follow official guide.

   redis-server --port 12000
   export GPUS=""
   CUDA_VISIBLE_DEVICES="$GPUS" PYTHONPATH=. \
       python rl/offpolicy/scripts/run_trainer.py \
       --config=experiments/mujoco/ecritic_quantile.yml
   
   CUDA_VISIBLE_DEVICES="" PYTHONPATH=. \
       python rl/offpolicy/scripts/run_samplers.py \
       --config=experiments/mujoco/ecritic_quantile.yml
   
   CUDA_VISIBLE_DEVICE="" tensorboard \
      --logdir=./experiments/logs

4. L2R example – ensemble training

   1. Redis

      redis-server --port 13131

   2. FC

      export GPUS=""
      CUDA_VISIBLE_DEVICES="$GPUS" PYTHONPATH=. \
          python catalyst/rl/offpolicy/scripts/run_trainer.py \
          --config=experiments/prosthetics/ecritic_quantile_fc.yml
      
      CUDA_VISIBLE_DEVICES="" PYTHONPATH=. \
          python catalyst/rl/offpolicy/scripts/run_samplers.py \
          --config=experiments/prosthetics/ecritic_quantile_fc.yml

   3. LAMA

      export GPUS=""
      CUDA_VISIBLE_DEVICES="$GPUS" PYTHONPATH=. \
          python catalyst/rl/offpolicy/scripts/run_trainer.py \
          --config=experiments/prosthetics/ecritic_quantile_lama.yml
      
      CUDA_VISIBLE_DEVICES="" PYTHONPATH=. \
          python catalyst/rl/offpolicy/scripts/run_samplers.py \
          --config=experiments/prosthetics/ecritic_quantile_lama.yml

   4. Monitoring

      CUDA_VISIBLE_DEVICE="" tensorboard \
         --logdir=./experiments/logs

5. L2R submit test

   bash ./submit/run.sh

Additional links

Medium post

Video