Lesson 2: pred_class not showing class name on Windows

Part 1 (2019)
1

When I ran

pred_class,pred_idx,outputs = learn.predict(img)
pred_class

I got the result below intead of the class name

<fastai.core.Category at 0x2d4b2879240>

The fastai version that I used is 1.0.46.
The platform is Anaconda Prompt in Windows10.

Thanks.

Jack

2 Likes
2

Hey Jack

How is your data set up?

In the mnist example the data is split into folders called ‘3’ and ‘7’ and we use ImageDataBunch.from_folder.

What are you using for bunching your data?

3

Hi @mcclomitz, thanks for your reply. I was working on the bear example in lesson2-download.ipynb. I was using the folder setup for bunching the data.

1 Like
4

I’m getting the same thing on my local windows machine.

I also then tried it in GCP, which has default Debian 9.6, with the latest fastai version and just ran a simple mnist like in the tutorial and got the same:

(<fastai.core.Category at 0x7f05502ce668>,
 tensor(0),
 tensor([9.9995e-01, 4.6885e-05]))

The classes do still show up in the Learner object:

learn.data.classes
['3', '7']

All above is running with fastai v.1.0.46. I’m not sure at which version the problem started though…

5

Turns out that you can still get the class name if you just save it and then print it out explicitly:

pred_class, pred_idx, outputs = learn.predict(img)
print(pred_class)
3
4 Likes
6

class Category(ItemBase):
“Basic class for single classification labels.”
def init(self,data,obj): self.data,self.obj = data,obj
def int(self): return int(self.data)
def str(self): return str(self.obj)

Category did overwrite str(self) implemented. Can’t figure out why print() needs to be called explicitly.

7

Thank you! it’s embarrassing how long I’ve been struggling w/this one step!!