A Biologically Constrained Model of Semantic Memory Search

This is a provisory repository for the paper submission: "A Biologically Constrained Model of Semantic Memory Search".

These instructions explain how to install the necessary software, run the model and reproduce paper results. Before doing this, it is helpful to read the paper. Also, since this is a provisory repository some parts of it might be unstable.

1. Install all Python package requirements with pip install -r requirements.txt (this might take some time). To run simulations faster (used here) install pyopencl and nengo_ocl (with pip). Installing them might require some manual intervention on some machines.

2. Install this project as package: python setup.py develop

3. Fetch FAN data from the web by running python fetch_external.py in the scripts folder. This will get Free Norms from the University of South Florida webiste, which hosts the data. Ngram data is also needed, here we provide bi-gram matrices. They need to be manually downloaded into the association_data directory.

4. In the same directory, run python categorize_animals.py. This step will create pickled files in animal_data that contain dictionary with animal to category mappings.

5. In the cogsci17-semflu directory run python create_database.py to create multiple pickled databases of association data and association matrices (FAN, Beagle and Ngrams) that are loaded by the model.

Run the model

To run the model, it should suffice to run the script run_models.py in the ./cogsci17_semflu/models/ directory. Model simulations will be generated in the sub-directory data.

Reproducing data

The results-plot notebook in the notebook directory can be used to reproduce Fig2 from the paper.