prediction flexibility - Githubissues

USGS-R / river-dl

Deep learning model for predicting environmental variables on river systems

Creative Commons Zero v1.0 Universal

21 stars 15 forks source link

prediction flexibility #102

Closed jsadler2 closed 3 years ago

jsadler2 commented 3 years ago

Make more flexibility in the prediction.

The predictions from this code allow you to only keep the second half of the predictions. The data prep step now makes the train and val sets have a half overlap, so that we can use only the second half of the predictions, but with the offset the predictions will be for the entire time period.
has a function now to make predictions with a given model for any arbitrary time period
I also break out what was one file postproc_utils.py into two other files. So now there are postproc_utils.py, predict.py, and evaluate.py

closes #101

jsadler2 commented 3 years ago

@aappling-usgs - if you have time in the next couple days to review, that'd be great. Otherwise I can either ask someone else or come back to it and review myself.

A lot of the "diff" in this PR is restructuring (copy-paste). I've pulled out the main additions to the code below.

aappling-usgs commented 3 years ago

Hi Jeff, I'm swamped. Do you think this would be a good fit for Janet to review? Or anybody else in DS you want to tap.

jsadler2 commented 3 years ago

No problem. Yeah. I'll see if Janet can. Thanks.

jsadler2 commented 3 years ago

@jzwart - do you mind taking a look at this when you get a chance? I can give you more of an overview in a quick call if that'd be helpful.

jsadler2 commented 3 years ago

@jzwart - thanks for the detailed review and suggestions. I think I addressed all the comments. Any other suggestions/comments?

jzwart commented 3 years ago

Looks good, Jeff. The diagram helped a lot for understanding the sequence ordering when making predictions. go ahead an merge