Object Oriented Programming Results

[topherbuckley]

Curious if anyone has already implemented this somewhere as I cannot find anything. I'm getting quite frustrated by all the dependence on pathlib and assumed directory structures when dealing with this library. Parsing a directory for files and directories that may or may not exist is quite error prone and frustrating in my opinion. Has anyone looked at wrapping the results into some class? i.e. taking all the outputs from train.py:

confusion_matrix.png
events.out.tfevents.*
F1_curve.png
hyp.yaml
labels_correlogram.jpg
labels.jpg
opt.yaml
P_curve.png
PR_curve.png
R_curve.png
results.csv
results.html
results.png
val_batch0_pred.jpg
val_batch1_pred.jpg
weights/*

and makes them accessible from calls like yolov5Results.hyp (e.g. a parsed yaml dict) where YoloV5Results defines some class that stores these things in memory. As it stands now the train.run only returns opts. One can use opt.save_dir to find the results path, but then to interact with any of the weights, csv, or yaml contents, one must:

1. Write some pathlib code to find a statically defined string e.g. "results.csv" for a file you hope doesn't change its name in any future updates.
2. Load it from disk rather than using it while in memory.

Ideally I'd like to see all of this available in whatever is returned from train.run, but as a workaround I was thinking to just write a wrapper class that does steps 1-2 above and saves them to some standard .