Skip to content

Evaluating story_scores and window_scores incorrectly? #10

Description

@eitanturok

Hi!

I'm trying to re-recreate the results of your paper and was wondering if you can explain why story_scores and window_scores are computed identically here:

# get raw score and normalized score for each window
window_scores[(reference, mname)] = metric.score(ref = ref_windows, pred = pred_windows)
window_zscores[(reference, mname)] = (window_scores[(reference, mname)] - window_null_scores.mean(0)) / window_null_scores.std(0)

# get raw score and normalized score for the entire story
story_scores[(reference, mname)] = metric.score(ref = ref_windows, pred = pred_windows)
story_zscores[(reference, mname)] = (story_scores[(reference, mname)].mean() - story_null_scores.mean()) / story_null_scores.std()

My understanding is that story_scores and window_scores should compute different things:

  • story_scores - a single score for the whole story, all words at once
  • window_scores - a list of scores, one score for each window of text.
    However, these values are the same and I wasn't sure if this was a mistake.

Also, does Table 1 of the paper report the story_scores or window_scores?

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions