Neural-linear reinforcement learning

Requirements

1. Run pip install -r requirements.txt from the project directory to install the requirements.
2. Follow the steps in here to install OpenAI Baselines

Documentation

1. The starting point of the code is main.py in the root directory
2. Code is organised into two packages utils and models.
   1. models package consists of all the code files related to feature extractor models and the linear SARSA models
   2. utils package consists of all the code files that are common utilities for the model training/evaluation

Feature Extractor Models Implemented: dqn_pretrained, dqn_glorot_normal, dqn_glorot_uniform, dqn_pooling, resnet50_pretrained, resnet50_random_init, resnet101_pretrained, resnet101_random_init, resnet152_pretrained, resnet152_random_init

Run instructions

1. To train the model, run

   python main.py train -s <save_dir> -e <feature_extractor_model_name> -w <weights_directory|Optional> 

   This will train the model and save the plots and weights to the specified save directory

2. To evaluate the model on the test data, run

   python main.py evaluate -s <save_dir> -m <model_weights_path> -e <feature_extractor_model_name> -w <pretrained_weights_path|Optional>   

   This will evaluate the model and save the gameplay video to save directory and prints the rewards