Dogs vs Cats in Crestle

Hi, I’ve decided to use Crestle for the first “homework” (aka setting up an environment and getting the cats vs dogs identifier working). I had a few questions on doing it, though.

  1. Even though Crestle comes loaded with the github repo, I was wondering how to create/edit files in the directory. Do you use the terminal and mkdir to do so?

  2. What files/folders are needed to run the program? I know that you need some training data for it, but what files/folders are needed and where in the repo are they? Also, where’s the jupyter notebook file for the first lesson? I’ve found several folders that contain some sort of “lesson 1”.

  3. Is anything else needed to create and run the program besides what I’ve asked for above?


You could navigate to this folder hierarchy and open lesson1.ipynb to start running it.

That’s one major question answered, thanks!

You shouldn’t need to edit any files, other than running the Jupyter notebook as shown by @beecoder . But if you do, you can click any file in the Jupyter notebook file list to edit it.

Thanks professor! I’m hoping to copy the data files to a new directory and either copy or retype the code in a new jupyter file to help me understand the major components of the program.

I couldn’t find the directory you listed on Crestle. This is the closes thing I could find to that:

The root directory contains this:

Is there something I missed?

That looks perfect

Might be a good reading

The most elaborate model that I used for this competition was a 10-fold CV 269-layer deep ResNet. It took about 15 hours to train each fold on my machine, so that translates into about 6 days training.

Looking back, what would you do differently now?
I would try harder and start earlier to look into training really deep neural networks from scratch. I’ve been able to train a ResNet 50 from scratch, and it outperformed the fine tuned pertained model.

1 Like

That’s fascinating and surprising!

Warning warning! Those are notes from Bojan Tunguz interview.