drfc-manager 0.1.0

A comprehensive Python package for managing AWS DeepRacer training workflows, model evaluation, and deployment. Features include pipeline management, custom model training, evaluation metrics, and visualization tools.

drfc_manager

A wrapper for DeepRacer for Cloud functionalities

Table of Contents

1. Objective
2. Key Advantages
3. Usage
4. Idea behind

Objective

The main purpose of this library is to provide an easy way to manage your workflow within the DeepRacer for Cloud (DRfC) environment.

This library allows users to optimize the training of their Reinforcement Learning (RL) models by managing the entire process primarily within a Jupyter Notebook. This includes configuring model parameters, creating training pipelines, and utilizing machine learning algorithms to improve training outcomes.

Key Advantages

• Easily set model configuration to training (hyperparameters, model metadata, and reward function)
• Create many model configurations for various training
• Stack multiple training configurations to be executed in a logical sequence

Future ideas :bulb:

• Set a stop criteria for model training (eg. convergence, loss behavior, iteration count, or another)
• Integrate with some hyperparameter tuning techniques from machine learning libs (e.g., Sckit Learning)

Usage

Define configuration model data

# Default values set from official documentation
model_name = 'rl-deepracer-sagemaker'
hyperparameters = Hyperparameters() 
model_metadata = ModelMetadata()

Define the reward function

def reward_function():
  reward = ...
  return float(reward)

Run pipeline

first_train = train_pipeline(model_name, hyperparameters, model_metadata, bytes_io_reward_function)
first_train()

Idea behind

This lib is being developed directly using the same ideas and implementation of the repo https://github.com/aws-deepracer-community/deepracer-for-cloud according to its description: "A quick and easy way to get up and running with a DeepRacer training environment using a cloud virtual machine or a local computer".