Technical University of Crete - Thesis

This work is part of my thesis for the completion of my studies in Technical University of Crete.

You can download the ".soclog" files as well as a pretrained CNN from this google drive.

• For the full datasets, of size 500,000 each, you shall download zip files from the "LARGE" folder.
• For a smaller version of the datasets, of size 100,000 each, you shall download zip files from the "SMALL" folder.

Note that "DEFAULT_DROID4_500000.zip" did not fit in the 15GB of free google drive space.
Therefore "DEFAULT_DROID4_100000.zip" was added, in the "LARGE" folder as well.
It shall be uploaded proprely once the availability of more space in the cloud is secured.

Steps to execute the code:

1. Download miniconda.
2. Run conda create --name <environment_name> python=3.8 and conda activate <environment_name>.
3. Run git clone https://github.com/diamantis-rafail-papadam/SoC-InitialPlacement-NN.git and cd SoC-InitialPlacement-NN.
4. Run pip install -r requirements.txt.

You might need to download another pytorch version according to your NVIDIA CUDA driver (shown with nvidia-smi or nvcc --version).
If this is the case, check current versions and old versions.

5. Unzip the downloaded dataset inside "SoC-InitialPlacement-NN" folder.

You should see a "logs" folder created which contains all ".soclog" files. If you want to see the win ratio of each player, run "python count_wins.py" (inside the created folder where "logs" are located).
The first number you see is the win ratio for each player while the second number is the average score over all ".soclog" files.

6. Make a directory named "DATASET", this will be needed in the next step.
7. Run "python extract_log_data.py" which might take a few minutes.
8. Run "python preprocess_data.py" which will create the input for the neural network.

• Regarding the "produce_graphs.py" file, you might want to change a few things:
  • Line 16   | Feel free to change the hyperparameters.
  • Line 125 | The path for the basic pretrained model, if you have one.
  • Line 126 | The path for the cnn pretrained model, if you have one.
  • Line 180 | This is the number of epochs.
  • Line 184 | Choose a device for the basic model, according to your system availability.
  • Line 187 | Choose a device for the cnn model, according to your system availability.

9. Run "python produce_graphs.py" and enjoy the results!

As a final note, you can use the "train.py" script to pre-train either network in whatever way you like.