Handling exceptions about geo-deep-learning

[epeterson12]

Expected behaviors should be agreed upon before we change our code to handle exceptions. This is a summary of how I think the program throws an exception during certain actions. Warnings should notify the user that the action couldn't be completed, such as an image not being able to be used for various reasons, then continue running with the rest of the data. Failures should notify the user and terminate the program.

images_to_samples.py

Risky Behavior	Expected Response	Function	Notes
Load config.yaml	Fail	main()
Load S3 bucket	Fail	main()
Get prep.csv from bucket	Fail	main()
Get prep.csv from local	Fail	main()
Get images from bucket	Warn	main()
Get images from local	Warn	main()
Create temp h5py files	Fail	main()
Get shp file from bucket	Fail	main()
Get shp file from local	Fail	main()
Assert band number	Warn	main()
Create shp raster image	Warn	main()
Validate num classes	Warn	main()
Read images and mask	Warn	main()
Create shp raster image	Warn	main()
Create masked raster	Warn	main()
Create h5py files	Fail	main()
Delete temp label tif	Warn	main()
Delete temp hdf5 files	Warn	main()
Add samples to the bucket	Fail	main()
Add labels to bucket	Warn	main()
Delete files downloaded from S3 bucket	Warn	main()

train_model.py

Risky Behavior	Expected Response	Function	Notes
Load config.yaml	Fail	main()
Load S3 bucket	Fail	main()
Load samples from bucket	Fail	main()
Create dataloaders	Fail	main()
Save intermediate checkpoint	Fail	main()
Save final Checkpoint	Fail	main()
Write model to bucket	Fail	main()
Check trn sample count	Fail	verify_sample_count()
Check val sample count	Fail	verify_sample_count()
Metrics logger	Warn	train()
Metrics logger	Warn	validation()

image_classification.py

Risky Behavior	Expected Response	Function	Notes
Load config.yaml	Fail	main()
Load S3 bucket	Fail	main()
Load model from bucket	Fail	main()
Load images to classify from bucket	Warn	main()
Assert Band number	Warn	main()
Read image as array	Warn	classification()
Create new raster from base	Warn	classification()

[epeterson12]

After discussion with ymoisan, we should set a limit to the number of warnings that we give. Once the threshold has been exceeded, the program should throw an exception and terminate. We would do this in case of systematic user error, for example trying to load images only in pdf format or having a different band number or number of classes in the config.yaml compared to the images in use.

[mpelchat04]

Relevant exceptions are being implemented ad hoc.