This project is focusing on DQN AI for the bullet hell games from Touhou 10 - Mountaion of Faith, using PyTorch.
Because it is on progress, it would be not working perfect.

Inspired by

• TH10_DataReversing
• TouhouCV
• deep-learning-bullet-hell-environment

First Trial(youtube)

• Windows 10
• Anaconda 4.5.11

git clone https://github.com/actumn/touhou10-dqn
cd touhou10-dqn
conda create -n myenv -f environment.yml
conda activate myenv
python main.py or train.py 

Like other deep q-learning examples, it takes a game screenshot from th10.exe at real time(th10/game.py) and treat it as a deep learning model input.

1. You must have touhou 10.0 - Mountaion of Faith
2. Before run train.py, There must be th10.exe on your processes
3. Before run train.py, enter 'game start' or 'practice start'