Skip to content
Deep learning model for recognizing puzzle patterns in The Witness.
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.

Finding The Witness Puzzle Patterns

This repo is example code and data for training a CNN to identify patterns from the video game "The Witness", using python3 and tensorflow 1.13.

Witness deep learning

To train the model:

# Optional: create a virtualenv
pyenv virtualenv 3.6.4 witness-3.6
pyenv local witness-3.6
echo witness-3.6 > .python-version

# Install requirements.
pip install -U -r requirements.txt

# Install tensorflow GPU support if needed
pip install tensorflow-gpu

# Link to W&B -- this will save your run results.
wandb init

# Process data from /data/all into a training set in /data/train and validation set in /data/valid.

# Train your model, and save the best one into the /model folder!

To visualize output:

# Run a prediction on all entries in your validation set.

# Generate a visualization of every layer in the model.

To submit to W&B:

Weights & Biases is running a public benchmark, to which anyone can share their model improvements. To submit your results, follow these steps.

  • Create a W&B account
  • Follow the above instructions to run a run in your personal project.
  • You can submit a run to the benchmark from the runs table:

Submit a run

You can’t perform that action at this time.