Open AObuchow opened 2 years ago
For now I just set a random seed for the validation set - if the results are poor with this seed (really bad/unlucky seed) I'll experiment with another seed.
Fixed in https://github.com/aditidagar/ocular_disease_recog/commit/43eac80595f258cdca86c88963493cbc4b05a5ba
Reopened because using a random seed doesn't seem to have caused the same images to be picked for the validation set. Need to look into how random.sample() works.
To objectively compare the improvements we make to the model, we need to ensure we're always using the same images in the validation set.
Ideally, we also should find a validation set that our model can perform well against (though this is kinda biased).