That's right. It took TEN DAYS for me to make meaningful progress on the notebook we glimpsed at the end of our last entry. Ten days. It's a whole new world, really. The Warriors won the championship. Amazon bought Whole Foods. An attractive new dancer joined our crew.
I would love to say that this was because the notebook was simply too much to process in a couple of days, and I've been spending the time since our last entry learning to draw the metaphorical owl:
The truth is, of course, that I'd let myself get far too distracted building sleeping pods from trampolines, wearing rompers to safari parties, attending experimental dance shows, and generally playing tour guide in the city.
As I found out once I'd finally made it through the notebook, making predictions on our test data is (surprise) embarrassingly simple:
One line of code? What was all the rest of it then? Looking over the notebook, most of it involves creating a process for sorting images from start to finish, ways to evaluate the accuracy of our model on the validation data, and formatting our output for submission to Kaggle.
I learned a lot though! I learned about globs, I learned about np.random, np.where, np.argsort... Okay, so a lot of NumPy stuff that I'd learned months ago and promptly forgotten.
I replicated the process in a new notebook, and uploaded my results to Kaggle.
Kaggle gives me a score of 0.09929, which is... Good? Bad?
The "Jump to your position on the leaderboard" button does't work, so I compare my score against the entire public leaderboard, which puts me in 430th place, just ahead of step time and Matthew Phillips.
The Dogs vs. Cats competition has seen 1,314 submissions...
... So I'm in the top 33%!