Question regarding test prediction submission csv file.

[ccleveland33]

Great work on this repo!

My submissions are no better than 0.5 AUC despite the training & validation going smoothly (~0.86 to 0.88). This seems to imply that the id and prediction are not matching up. However, when I check submission.csv the ids are in the correct order which indicated the predictions are being scambled in some way.

Is there a difference in tensflow/keras flatten or .predict() that is version dependent?
Or is something else going on?

Thanks!

[ccleveland33]

I suggest changing:
test_ds = test_gen.get_dataset(shuffle = False, repeat = False, target = False)

To the following:
test_ds = test_gen.get_dataset(shuffle=False, buffer_size=None, ordered=True, repeat=False, target=False)