HillClimber is an implementation of HillClimbing algorithm to optimize the weights of a given neural network.
This repository also contains:
- A Dash application to visualize the searching process and the search graph (partially, since the graph has more than 3 dimensions).
- A Jupyter notebook to demonstrate how to use the packages.
- Sample log and model files.
- OpenAI Gym
- PyTorch
- Plotly Dash
- Numpy
- Pandas
- SciPy
| CartPole-v1 | MountainCar-v0 | LunarLander-v2 |
|---|---|---|
 |
 |
 |
| ~10-30 seconds | ~1-10 minutes | ~10-60 minutes |
I was inspired by this video of TheComputerScientist. It is a nice channel if you are interested in AI.
Also recorder.py is taken from highway-env. I was planning to include highway environment in this project as well, but for some reason each episode took around 3-12 seconds during my trials, so I decided not to.