Lesson 1 learn.lr_find() gives #nan# in valid_loss

Part 1 (2019)
1

I am getting #na# when I ran learn.lr_find() from the lesson2-download notebook on my dataset. I had already made the necessary changes to make it work on the dataset I created.
This is the output of learn.lr_find().

epoch train_loss valid_loss error_rate time
0 0.199959 #na# 00:10
1 0.182260 #na# 00:09
2 0.171656 #na# 00:09
3 0.145169 #na# 00:09
4 0.149205 #na# 00:08
5 0.143180 #na# 00:09
6 0.136640 #na# 00:09
7 0.135336 #na# 00:09
8 0.122550 #na# 00:09
9 0.109931 #na# 00:08
10 0.097871 #na# 00:09
11 0.088499 #na# 00:08
12 0.083600 #na# 00:09
13 0.086121 #na# 00:10
14 0.236878 #na# 00:09

Does this affect the model? And how do I resolve this issue?

2

I’m facing the same issue in Lesson2 on COLAB :frowning_face: Any help is appreciated…

3

Hi Karanb. I’d suggest to paste your code and the specification of the system you are running on, so that it is easier to identify what the problem could be.

4

Also experiencing this issue when running lesson-2 notebook in colab.

The code is as in the default notebook.

learn.lr_find()

This is immediately after calling learn.unfreeze()

I also notice that in the printed results ‘time’ is sitting under the ‘error_rate’ heading, just like in OPs screenshot.

If I find the answer on another thread I will come back and post here.

5

Update: other forum users are explaining that valid_loss is not necessary for lr_finder, and printing as #na# is correct / appropriate.

6

Hi Team,

I am getting same the error for Lesson 2 - Download but here i am using cats,dogs,monkeys classes instead of bear types. But the issue is still the same.

image
image820×750 20.7 KB

7

lr_find does not use the validation set. So its not an issue really. Cheers!

8

Hi ,

I am getting error for ImageList:

image
image1920×1080 145 KB

Can anyone guide please?

9

@shiven_004
I’m guessing you’re using the fastaiv1 library.
ImageList doesnt have a .from_files attribute. You’re looking for .from_folder method.

Syntax: ImageList.from_folder(path)
where path is the path object of the folder where your data is present
Cheers!