Remote Associates Test

Neural network simulation of the Remote Associates Test as presented in the paper:

Kajić, I., & Wennekers, T. (2015). Neural network model of semantic processing in the remote associates test. In Workshop on Cognitive Computation: Integrating Neural and Symbolic Approaches, 29th Annual Conference on Neural Information Processing Systems (NIPS 2015)

There is an algorithmic and a neural network simulation of the test: the algorithmic version is faster to run because it does not simulate neurons but search in a graph. The simulation of a model takes much longer, with minor differences in results.

The following manual explains how to reproduce figures and numbers from the paper.

0. Get this code

Get a copy of this repository by:

git clone git@github.com:ikajic/remote_associates_test.git

In the cloned directory, install the package (for easier handling of imports in scripts):

python setup.py develop

1. Requirements

First, ensure all tools and packages are installed. Most packages are available with pip, so from the cloned repository you can either do:

pip install -r requirements.txt

to get all of them (and a bit more) at once, or install them one by one:

• Python 2.7
• NumPy
• SciPy
• matplotlib
• pandas

2. Fetching and processing the data

You need to manually download the University of Florida Free Association Norms to ./data/raw/freeassociations/data. There is a README file in that directory explaining what needs to be contained there.

After downloading the sheets, run the script process_data.py in ./data/raw/freeassociations to generate the association matrix. It will store the matrix in the place where the rest of the scripts know how to find it. This script assumes the norms will be downloaded by right-clicking "Save as..". This will download a html file for every sheet and those contain html tags at the beginning and at the end. The process_data.py script automatically handles that by removing the first and the last three rows (it does not check whether those are html tags, it just removes them).

If that went well, you should be seeing a file called free_associations_vocabulary in ./data/processed, this is a Python pickle file that stores the association matrix and the word vocabularies.

3. Running simulations

To run the algorithmic version of the model (fast), go to the algorithm directory and run run_simulation.py, otherwise do the same thing for the model. The code for model is not optimized, so it is very slow and eats up a lot of memory.