DQN + HER
This repository contains the implementation of DQN + HER. The implementation is tested on the toy problem presented in the paper. Here is a blog post about HER.
The hyperparameters used in this repo are the same as the paper..
: 0.001
: 0.98
- Q-Network is an MLP with 256 hidden units
- Buffer holds up to
transitions
How to train?
python train.py --help
usage: train.py [-h] [-v] [-s S] [-i I] [-e E] [-c C] [-o O]
HER Bit Flipping
optional arguments:
-h, --help show this help message and exit
-v Verbose flag
-s S Size of bit string
-i I Num epochs
-e E Num episodes
-c C Num cycles
-o O Optimization steps
Inference
TODO
Results
TODO