tag:blogger.com,1999:blog-19803222.post1179044417038138482..comments2024-03-18T01:45:45.724-06:00Comments on natural language processing blog: Debugging machine learninghalhttp://www.blogger.com/profile/02162908373916390369noreply@blogger.comBlogger1125tag:blogger.com,1999:blog-19803222.post-70465176611138327292016-08-24T19:08:28.461-06:002016-08-24T19:08:28.461-06:00I've learned to always pitch my model against ...I've learned to always pitch my model against a random and an averaging predictor. If your regressor or classifier can't beat a simple average (or worse, total random guessing), well... no need to continue before finding more signal.<br /><br />I also like using VW or Random Forest 500 to benchmark against. It can give estimates on the hardness of a problem and how well your optimization is doing vs. very standard modeling techniques.<br /><br />I really dig the data-halving trick. Have to try that out soon.Anonymousnoreply@blogger.com