what is y_range for in collab_learner? I didn’t understand well.
For the collab filtering example, why do we use MSE instead of RMSE (used for one of the benchmarks)? What is the intuition for using one over the other?
It’s how you tell your model to spit values between bli and bla.
1 Like
Here RMSE is the metric that is used in academic benchmarks.
2 Likes
limiting the range of prediction to be in this range…standard sigmoid is range of values between 0 and 1 ,so this is i think custom one between 0.5 and 5
1 Like
It’s a trick he added for better results. So, the y values will always be in this range only. Neural net will predict in this range only.
2 Likes
Quote of the day:
29 Likes
I believe it has been given by several Kaggle gandmasters 
4 Likes
Using the funtion on collab gives error
ratings = pd.read_csv(path/‘u.data’, delimiter=’\t’, encoding=‘latin-1’, header=None, names=[user,item,‘rating’,‘timestamp’])
FileNotFoundError: File b’file with complete path on local mac’ does not exist
Any suggestions ?
1 Like
OK thanks @sgugger
You mean, just doesn’t show them? And why does a skip_end=15 not show the last 3?
so we should us MSE instead of RMSE then?
1 Like
Just plot losses[:-skip_end], not all losses. The divergence can completely screw your graph, that’s why we have this option. But sometimes, the divergence hasn’t happened with default skip_end, and you want to see it to decide your learning rate, which is why Jeremy picked a lower value than the default.
5 Likes
Oh, thank you.
what does learn.bias actually do?
The loss function should probably be MSE as it works better usually, but having RMSE as a metric will tell us how we compare to academic benchmarks.
1 Like
Jeremy’s explanation of bias in terms of the example of anime seems to mean that it measures qualities of the movie not accounted for by the latent factors. Does this imply or require that the model will have already discovered an “is anime” latent factor?
2 Likes
Yes, in some way at least.
1 Like
It shows the bias vector for the item or user. If you want to know what a bias does in general, it has been answered above.
got it, thanks!