Pocket-titan / rogers_mcclelland_pytorch

🖥🧠PyTorch model for semantic cognition

PyTorch models of semantic cognition

Based on the paper "Semantic Cognition: A Parallel Distributed Processing Approach" from 2003 by Rogers and McClelland, whose model is itself an adaptation of an earlier one by Rumelhart (1990).

Inspired by the PyTorch implementation from https://github.com/jeffreyallenbrooks/rogers-and-mcclelland.

Running the model

The necessary packages change over time, but be sure to

pip/conda install torch torchvision tensorboard pandas scipy numpy seaborn scikit-learn matplotlib

To train the model and generate log data, run

python feedforward.py

To view the logs in TensorBoard (it's nice to keep this running in a seperate process), run

tensorboard --logdir logs/fit

and go to localhost:6006 in your browser. One run will show up in TensorBoard as two sets of data; one containing the scalar metrics (in ${run_name}/train) and another containing the images (in ${run_name}/).

To clear the logs (the folder tends to fill up fast), run

python clear_logs.py