GymDash

GymDash (originally meant to be Gymnasium Simulation Dashboard) is a simulation dashboard for keeping track of, analyzing, and interacting with simulations, reinforcement learning projects, and other machine learning projects.

Table of Contents

1. Install
2. Getting Started
   1. Controls
   2. Examples
3. Frontend
   1. Sidebar
   2. Control
   3. Analyze
   4. Configs
4. Backend
   1. Project
   2. Simulations
5. FAQ

See the Getting Started documentation.

Disclaimer

This project is still IN DEVELOPMENT. Many frontend and backend features have not been finalized or thoroughly checked and may be subject to significant change at any time.

Installation

NOTE: Full Installation requires PyTorch
Option 1 - Local Install: Download the project and install via pip locally. Download the repo with:
git clone https://github.com/ChaseDVickery/GymDash.git

Option 2 - pip: Download the project via pip using a full install or a minimal base install:

• Full Installation: Full installation includes Tensorboard, Gymnasium, Stable Baselines, and PyTorch/Torchvision support and packages. Full installation requires PyTorch installation:
  python -m pip install gymdash[full]

• Minimal Installation: Minimal installation just includes the base packages required to run the frontend and backend. The example Simulations will not properly work without a full installation:
  python -m pip install gymdash

Example Full Install:

• Windows 10
• Cuda 11.7

1. Venv Setup

# Create and activate venv
python -m venv testvenv
testvenv\Scripts\activate.bat
python -m pip install --upgrade pip

2. Install CUDA via pip

# Install CUDA via pip: https://docs.nvidia.com/cuda/cuda-installation-guide-microsoft-windows/index.html#pip-wheels
python -m pip install --upgrade setuptools pip wheel
python -m pip install nvidia-pyindex
python -m pip install nvidia-cuda-runtime-cu117

3. Install torch cuda 118

# I know there is a mismatch b/w 11.7 and 11.8, but it still worked for me.
python -m pip install torch torchvision --index-url https://download.pytorch.org/whl/cu118
# Optionally, you can replace steps 2-3 with `python -m pip install torch torchvision` if you don't need GPU support. See PyTorch installation: pytorch.org/get-started/locally/

4. Install gymdash via pip

python -m pip install gymdash[full]

Getting Started

Running Locally

Launch GymDash

# This creates '.gymdash-projects' folder in the launching folder
# and starts a frontend HTTP server and a backend API server.
python -m gymdash.start

For additional launch options, see the Launch Options.

Navigation

Navigate your web browser to your localhost port 8888: http://127.0.0.1:8888/src/gymdash/frontend/. This loads the HTML page used to interact with your GymDash project.

NAQ (Nobody Asked Questions)

Q: Is the API RESTful?
A: No. The API is not RESTful. Most importantly, the entire system is STATEFUL in that certain API calls could depend on prior API calls. This is for a couple reasons:

1. Dynamic Project: Users are interacting with a potentially dynamic project on the backend. It makes sense for API calls to represent all or part of the current state of the project.
2. Lower Bandwidth: Simulations have to potential to generate lots of logging data that should be available. To ensure we don't waste bandwidth transferring repeat data, we track which data still needs to be sent on the backend.

Q: Why is the interface so ugly?
A: That's just like, your opinion, man. Also it's because I'm a very utilitarian designer (aside from just being not good at it), and I haven't felt the need to go full-focus on the frontend visuals because there is so much else that can change before that. There are color themes, though. You just can't select them yet. Feel free to put your own spin on it, too. You should even be able to just edit the HTML, CSS, and JavaScript behind it.