I broke up the NLP parsing into batches, size 500. Take s bit of time, approximately 30 minutes, but its fine. I've also updated some subsequent functions to use all the parsed data points.
It looks like there are about 7,000 points that meet our current criteria of "no parking" and a start and end time. That's out of 80,000 total points. Need to look into improve my "recall" aspect, getting more of the 80k points parsed. Also need to look at my "precision" numbers. Of the points I have picked, what percentage are correctly labeled? Random sampling is the technique here.
No comments:
Post a Comment