MST 2 Alternative First Choice Implementation

[amaar-mc]

Compares the brain's clip voxel embedding data collected from ses-03 to the embeddings of 31 pairs of MST images (62 images) to determine which MST image pairmate the clip voxel embedding data is closer to via cosine similarity.

Because the 31 pairs of MST images are shown twice, the MST 2AFC Score is calculated twice for both showings of each set of pairmates, as reflected by subset0 and subset1, thus we have mst_2afc_subset0 and mst_2afc_subset1.

The details regarding the 2AFC process are printed in the mindeye.ipynb notebook in the format as follows (example):

  image 1: pair_46_w_pool2.jpg
  image 2: pair_46_w_pool1.jpg
  img 1 brain embedding chose: pair_46_w_pool2.jpg Correct
  img 2 brain embedding chose: pair_46_w_pool1.jpg Correct

Cosine similarity is performed on the img 1's brain embeddings against the two images, and the img 2's brain embedding data. The labeling of whether or not the decision was correct is labeled on the right.

The net MST 2AFC Score is provided after the detailed breakdown above, as in this format:

mst 2afc score: 0.7903 (79.03%)

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB