https://gist.github.com/chtnnh/15a77653279d50a0b90179aa83db4fca

This is the difference between model performance before and after adding words_to_watch to the feature_lists/ptwiki.py on selected articles that were misclassified by the old model

I just reviewed @Chtnnh's post at https://www.mediawiki.org/wiki/ORES/Issues/Article_quality#Summary I made some modifications to the table to make it easier to read.

I think we should change the column values to include both the prediction and the weighted_sum. That will help us track improvements.

CLASS_WEIGHTS = {"1": 1, "2": 2, "3": 3, "4": 4, "5": 5, "6": 6}

def weighted_sum(score):
  sum = 0
  for cls, proba in score['probability'].items():
    sum += CLASS_WEIGHTS[cls] * proba
  return sum

https://www.mediawiki.org/wiki/ORES/Issues/Article_quality

The table has been updated under the summary section of the misclassification reports. This task can be closed as resolved.