Understanding the effect of varying amounts of replay per step in DQN

I acknowledge that this work is part of the credited course group project on Reinforcement learning I at the University of Alberta.

Group Member:

a) Animesh Kumar Paul animeshk@ualberta.ca

b) Videh Raj Nema nema@ualberta.ca

Usage

Run codes one of the following ways:

Option 1: Execute all runs at once

python 3_run_mc_colab.py --exp experiment_name --algo dqn --replay_frequency 2 --learning_rate 0.1 --console_output 0 --use_gpu 0 --is_mac 0 --run_start 1

Option 2: Execute each run seperately manually

start = 1

max_runs = start + 1

python 3_run_mc_colab.py --exp experiment_name --algo dqn --replay_frequency 2 --learning_rate 0.1 --console_output 0 --use_gpu 0 --is_mac 0 --max_runs max_runs --run_start start

Option 3: Execute each run seperately automatically (Sequentially execute each run) -You need to set the hyper-paramters in 2_easy_run_sequentially.py file.

python 2_easy_run_sequentially.py

Option 4: Execute each run seperately automatically (Parallelly execute each run)-You need to set the hyper-paramters in 2_easy_run_parallel_backgroun.py file.

python 2_easy_run_parallel_background.sh

It contains also two jupyter notebooks for plotting purposes.