7 PM – AI/ML – Darin

May 20 @ 7:00 pm - 8:00 pm

For Rexford, choose the dataset for your project (has to be multiclass, and catch up the below)

HW 1:

Choosing what you want to classify — Can be anything in real life, the only thing is you have to make sure the data exists on Kaggle.

The project has to be able to classify at least 5 different categories.

What we did last week:

Set up the labelling function
Verify Dataloader works with .show_batch() (reference old code for how to do this, it will depend on how you extract labels, whether its via path or filename)
Run training with a vision_learner.

What you have to do next week:

Use the proper imports (fast ai 2.7.19)
Use lr.find() to get the proper learning rate, then run finetuning with the discovered value (you must only call fine_tune once!)
Add the following into the vision learner to ensure lr_find works!:
path=Path(“/kaggle/working”),
model_dir=”models”,
Export the file as a .pt file onto your computer!

HW 2:

Then do the following in a new notebook called WED-7PM-PandasHW1:

Load the data
Show only the Channel and Subscribers columns
Find channels with more than 2000 subscribers
For this you can use something like this:
df[df[“Math”] > 80],but of course for this data.
Add a new column Subs_per_Video
Answer the question in Markdown, which channel is the most efficient?

You can reach me at ddjapri@ayclogic.com.

All class notes can be found here.