ambimorph / neurotopics Goto Github PK

Authors: Jessica Mollick, Tim Rubin, L. Amber Wilcox-O'Hearn

Contact: [email protected], [email protected], [email protected]

Released under the GNU AFFERO GENERAL PUBLIC LICENSE, see COPYING file for details.

Neurotopics is a topic modelling tool for NeuroSynth. We will take a Neurosynth dataset, a brain atlas in NIFTI or Analyze format, and a file containing word frequency counts for all words in the documents of the dataset. For each ROI in the atlas, we will generate a distribution on a subset of the words.

• Python
• NeuroSynth

The formatting of the docwfreqs file uses a sparse matrix format which is efficient for  storing large text files (because a Word x Document matrix is highly sparse). Each row represents a document. Within each row, the document's word-counts are formatted as follows:

wordID:wordCOUNT wordID:wordCOUNT ....

So for each unique word-type in the document, there will be a pair of numbers separated by a colon, indicating: (1) wordID; the identifier for the word-type, which maps to the strings in the vocab-list file (2) wordCOUNT; the number of times the wordID occurred in the document (i.e. the number of tokens of wordID in the doc)

Suppose our vocab list was:

0 brain 1 function 2 eye 3 vision 4 pizza 5 tacos

Now suppose we had a document with the following tokens (ignoring order)

brain eye eye eye eye pizza tacos tacos tacos

In the docwfreqs file it would look like this:

0:1 2:4 4:1 5:3