feat: additional data loggers and enhancements, reducing gateway course limit from 25 to 10 #62
Conversation
feat: logs missing "M" course grades oddities feat: add cohort and cohort term breakdown + additional logging statement prints for inference pipeline too fix: capping gateway features at 10 instead of 25 fix: <0.1%% has a double percent sign - changed to "<0.1 percent" for sake of ease (couldn't fix it to just print 0.1%, idk why)
nm3224
changed the title
nm3224
changed the title
…is triggered currently, the logs were not being updated for the same model run ID but a new inference run. this should fix this (changed mode = "a" (append) to mode = "w" (write). Need to test still.
| except Exception: | ||
| pass | ||
|
|
||
| # --- IMPORTANT PART 2: open in write mode so we overwrite each run --- |
There was a problem hiding this comment.
What are we overwriting here? When we re-run inference, shouldn't that just produce a new run ID subfolder?
There was a problem hiding this comment.
when we re-run inference if it's the same training model (so if we didn't need to retrain) the logs for some reason aren't getting updated. I still need to test this to ensure it works but we need the logs to be updated to the most recent inference run.
There was a problem hiding this comment.
this makes me think if under the inference folder, we should have additional sub folders for old & new inference runs
There was a problem hiding this comment.
I see, so then we are only keeping the latest inference run then? I think I remember discussing this. Ideally, we want to have all historical inference run logs as well.
But yeah, for now, I understand we need to prioritize capturing the latest and ensuring that is done correctly.
There was a problem hiding this comment.
agreed, we do need to keep old runs
changes & context
Small tweaks/improvements
questions