Readme.md

Environment Setup

Dependent Software

The environment needs the following software to be installed:

Pycharm Community Edition

Used as it creates a container for each project, allowing you to select the Python and dependency versions for your project, without conflicting with any global settings.

Python 3.6 for Windows 64-bit

No need to add this to path

Git for Windows

For version control support

CUDA Toolkit 9.0

Used by Tensorflow for GPU support

cuDNN 7 *

The file cudnn64_7.dll is needed by Tensorflow for GPU support. The download should be extracted and placed somewhere on drive. The path to the DLL will must be manually added to the System Environment PATH variable.

* Registration to NVIDIA required

Process

It is recommended to use a machine with a compatabile NVIDIA graphics card that can be used by TensorFlow. The UCLan Games Lab has PCs with suitable NVIDIA GeForce GTX 1070 cards.

Not using a graphics card is significantly slower and not recommended.

1. Open Pycharm and checkout code from Github https://github.com/scooby359/RL-Simple.git

2. When prompted by PyCharm, set the environment to Python 3.6, or manually go to the project options and set.

3. From the PyCharm terminal, verify Pip is available with pip -V

4. Install packages for skimage (scikit-image), matplotlib by using the prompts in PyCharm.

5. Install TensorFlow 1.10 with pip install tensorflow-gpu==1.10. If not using a machine with a suitable graphics card, install tensorflow==1.10 instead.

6. Install Numpy with pip install numpy==1.17.4 Tensorflow will require version <= 1.14.5, but this isn't compatible with the rest of the project so needs to be manually changed.

7. Install OpenAI Gym with pip install gym

8. The project can then be run from PyCharm by right clicking on the Python file and selecting Run.

9. After completion, result graphs will be displayed. The graphs, raw CSV values and the TensorFlow model are saved in ./result/[TIMESTAMP]/.