https://github.com/willclarktech/ts-rl

Reinforcement learning using TensorFlow.js
https://github.com/willclarktech/ts-rl

Last synced: 17 days ago
JSON representation

Reinforcement learning using TensorFlow.js

• Host: GitHub
• URL: https://github.com/willclarktech/ts-rl
• Owner: willclarktech
• License: mit
• Created: 2020-06-22T16:58:07.000Z (over 4 years ago)
• Default Branch: main
• Last Pushed: 2024-06-17T22:29:45.000Z (8 months ago)
• Last Synced: 2024-06-18T23:12:10.705Z (8 months ago)
• Language: TypeScript
• Size: 1.45 MB
• Stars: 0
• Watchers: 3
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# ts-rl

Reinforcement learning using TensorFlow.js

## Installation

Install Node modules:

```shell

npm install

```

For Python integration you will also need:

- Python v3

- Jupyter Notebook

- pip modules (TODO: document these)

## Starting the results server

```shell

npm run serve

```

Navigate to [http://localhost:8000]() to see charts of training losses. The chart data will be refreshed every few seconds for live feedback during training.

## Training models

```shell

# npm start [algorithm] [environment]

# eg:

npm start dqn blackjack

```

## Loading models into Python

Convert saved model using `npm run convert  `, eg:

```shell

npm run convert ./models/DQN-CartPole-q-network/model.json ./models/DQN-CartPole-q-network-converted

```

Start Jupyter Notebook:

```shell

npm run notebook

```

See demo model loading/evaluation in [`https://github.com/willclarktech/ts-rl/blob/master/notebooks/model-load-demo.ipynb`]().