I’m done cleaning up my data use cleaner = ImageClassifierCleaner(learn)
. And then the book says “Once we’ve cleaned up our data, we can retrain our model.” so where do I start retraining, do I need to recreate the Dataloaders, or do I just need to run "learn = vision_learner(dls, resnet101, metrics=accuracy)
learn.fine_tune(7) "?
My understanding is that the cleaner
steps edits or removes image file paths from the path
folder, so once cleaning is complete, you create a new DataLoaders
with that (now cleaned and updated) path
and then train the model. Following the steps in 02_production.ipynb
:
# load the cleaner and select images to delete/move
cleaner = ImageClassifierCleaner(learn)
cleaner
# delete or move images as selected in the `cleaner`
for idx in cleaner.delete(): cleaner.fns[idx].unlink()
for idx,cat in cleaner.change(): shutil.move(str(cleaner.fns[idx]), path/cat)
# create a new `DataLoaders` object using the updated `path` and train the model
dls = bears.dataloaders(path)
learn = vision_learner(dls, resnet18, metrics=error_rate)
learn.fine_tune(4)
1 Like