Incorrect Labelled Data

Hi Sir,

We had couple of doubts from the lecture Cleaning Up Incorrectly. Can you please help to clarify sir?

Statement 1: In the below pic, if we did sum up 8% + 43% + 61% + 6%, the total should come to 100% but not coming in this case . What could be the reason behind ?

Statement 2: But you don’t trust your dev set anymore to be correctly telling you whether this classifier is actually better than this because your 0.6% of these mistakes are due to incorrect labels.

My intuition about statement 2: Assume before fix incorrect label, classifier A 2% error better than classifier B 4% error evaluating against the dev set. But after fixing incorrect label in the dev set, now we will get classifier B 1% error than classifier A 2% error. Is it due to the reason we should not trust dev set ?

Hi, regarding statement 1: Since a picture can be both Blurry and contain a Great Cat it will not sum to 100%.