a2c-acktr-vizdoom

A2C and ACKTR implementations for ViZDoom

Packages to install

pytorch scipy sdl2 vizdoom

Training the agent

python main.py --algo a2c --num-processes 16 --config-path scenario/health_gathering.cfg --num-frames 10000000 --log-dir $LOG_DIR --save-dir $SAVE_DIR --no-vis

Running the agent

MODEL_FILE_NAME: name of the model without the .pt extenstion (assumes the model is in ./trained_models)

python enjoy.py --config-path scenario/health_gathering.cfg --env-name MODEL_FILE_NAME