Clone the repository and switch to one of the branches named easy, medium and hard. Fill in the gaps which are marked with TODO. Execute script dqn/src/main.py. All scripts can include TODOs. After finishing development, you can download pretrained weights from the Nextcloud-folder and see the agent play. You switch between training and playing process by changing the train-Flag. The slides can be downloaded as pdf from the Nextcloud (filename: "Build your own Hero.pdf").
Develop an AI which is able to play Atari games. Reinforement Learning with Deep Q Networks is used to train an autonomous agent.
- VirtualBox
- At least 15 Gb free disk space
- Follow install instructions of VirtualBox
- Downloaded .vdi-file from nextcloud
- Click New-Button
- Type an appropriate name for the image
- Choose Type Linux and Version Ubuntu (64 Bit)
- If you are not able to choose 64 bit:
- Ensure Hyper-V is disabled
- Ensure Hardware virtualization is enabled in the BIOS
- Choose Use an existing virtual hard drive file
- Click on the folder button and browse for the .vdi file which you downloaded in the previous step
- Click Create
- Double-Click on image created in previous step
- If error occur: make sure Hyper-V is disabled
- Login as modalg with password 12345678
- Click on Ubuntu-Button and search for eclipse
- Execute eclipse
- Click ok when workspace dialog appears
- Open test folder in modalg181-Project
- Double click on any of the files
- Execute file with click on run button (run as Python Run)
- Open explorer
- Navigate to folder Projekte
- Right click in explorer and select Open in terminal option
- Type git clone https://github.com/mati3230/MetaMarathon.git in terminal
- If closed, start eclipse
- Right click in PyDev Package Explorer
- Select New/Project... option
- Select PyDev/PyDev Project and click Next
- Type project name MetaMarathon
- Select grammar version 3.0-3.5
- Select interpreter /home/modalg/miniconda3/envs/modalg/bin/python3.5
- Press Finish
- Open explorer
- Navigate to folder Projekte
- Right click in explorer and select Open in terminal option
- Type git clone https://github.com/mati3230/MetaMarathon.git in terminal
- Open explorer
- Navigate to folder Projekte
- Right click in explorer and select Open in terminal option
- Type git pull origin master in terminal
- Open explorer
- Navigate to folder Projekte
- Right click in explorer and select Open in terminal option
- There are the the branches easy, medium and hard. Type git checkout easy in terminal to switch to easy-branch. Type git checkout medium in terminal to switch to medium-branch
- If closed, start eclipse
- Left click on the project MetaMarathon you created before in step Clone this repository and setup eclipse and press F5 to refresh the project. Now you should see a dqn folder.
- You can download pretrained weights of the model here
- Place the files to /dqn/src/experiments/Breakout-v0/checkpoints