A lightweight PyTorch implementation of the Triangle Model (Harm and Seidenberg, 2004).
- model.py: Gradient computation.
- dataset.py: Data utilities.
- train.py: Training utilities + ODE solver.
- benchmark.py: Runs phased training procedure. Saves losses + accuracies.
- evaluation.py: Plots behavioral results.
To run the default set of experiments,
python3 benchmaking.py -ID baseline -model_config configs/baseline/model_config.json -trainer_config configs/baseline/trainer_config.json -optimizer_config configs/baseline/opt_config.json
python3 evaluation.py -ID baseline -model_config configs/baseline/model_config.json -trainer_config configs/baseline/trainer_config.json