What is the purpose of the /models and /tmp directories?

I’m too cheap to spin up my AWS instance to verify, but I’m assuming from the lesson 1 notebook that:

/models = where models are saved/loaded from (e.g., when you run learn.save() or learn.load() )

/tmp = where your precomputed features are stored (eg., when precompute=True in ConvLearner.pretrained() )

Is this correct? If so, does that mean we must always include them in our /data directory or can we override these default settings?

Correct. There are params you can use to choose other folders. If you make them relative paths, they are relative to your data dir.

Cool. Thanks for the confirmation.

In the notebook, you have a commented out line for deleting precomputed features (!rm -rf {PATH}tmp). I assume that the default behavior is to use the precomputed features if they are there rather than precomputing them?

Btw, It’s interesting to see where you’ve made many of the recommended and repetitive practices introduced in v1 of part 1 into one liners. The framework code is a little bit difficult to read, but niceties like this are great time savers.

2 Likes

That’s right.