Lesson 3 In-Class Discussion

Part 1 (2018)
1

This thread contains the in-class discussion from Lesson 3. The Wiki links have been moved to this new thread. Please ask any questions about lesson 3 in the new wiki thread.

23 Likes
2
3

The live stream has not started yet right? Or is it just me

4

Not for an hour.

1 Like
5

Had a doubt regarding creating a submission file for Kaggle. The competition page says :

Submissions are evaluated on **Multi Class Log Loss** between the predicted probability and the observed target.

logloss=−1N∑i=1N∑j=1Myi,jlog(pi,j)

Just wondering when did we apply this formula in our model? And how could we change our code if they required some other log loss other than multi class?

Thanks!

1 Like
6

when will the stream start?

7

The loss criteria gets chosen by virtue of the data in fastai as assigned to self.crit below:

fastai/conv_learner.py

class ConvLearner(Learner):
    def __init__(self, data, models, precompute=False, **kwargs):
        self.precompute = False
        super().__init__(data, models, **kwargs)
        self.crit = F.binary_cross_entropy if data.is_multi else F.nll_loss
        if data.is_reg: self.crit = F.l1_loss

binary_cross_entropy, nll_loss and l1_loss can be looked up here: http://pytorch.org/docs/master/nn.html

–

5 Likes
8


Looks like live stream has not started, please let me know if there is another link to live stream

9

It’s due in 15 mins. (6:30 PM PST, 8 AM IST)

1 Like
10

Not started yet.

11

The live stream is still offline?

12

Yes, It’s.

13

Its offline for me still

14

Correct url: https://www.youtube.com/watch?v=_VpaKaMyjqI

20 Likes
15

Thanks, Ankit!

16

Thank you

17

It has been updated in the wiki above.

18

what is the environment file?

19

@jeremy Could we get the rest of the links you presented in the beginning please?

20

@jeremy Can you please share all the blog links here in the wiki section that you just showed in the class?

2 Likes