search

sophieball / toxicity-detector

MIT License
0 stars 0 forks source link

Sample 10K CLs from this year and train prompt types #72

Closed sophieball closed 4 years ago

sophieball commented 4 years ago

71 contains the code to train an unsupervised classifier using 10K code reviews

CaptainEmerson commented 4 years ago

@sophieball Just to clarify, I should run on the train_prompt_types target, with a CSV schema that's the same as data/pr_body_comments.csv?

CaptainEmerson commented 4 years ago

Each row is a one comment.

We don't have:

CaptainEmerson commented 4 years ago

@sophieball I ran this and uploaded the results to Drive, but I'm not sure what I'm looking at. I can also show you a sample of the data I fed in.

sophieball commented 4 years ago

8/14 doesn't seem like the output from prompt types. They look like old output from fighting words and politeness

CaptainEmerson commented 4 years ago

Success!