I see. So basically low confidence score indicates not 3 and since we are classifying only 3s and 7s then it’s probably a 7. TY giggs!!
Careful, do not confuse a prediction and the confidence!
The prediction is just a value between 0 and 1, where any number in [0, 0.5) is 7 and any number in (0.5, 1] is 3.
What determines the confidence is how close the prediction is to each target.
0.9 is > 0.5 so it’s a 3, and it’s very close to 1, so it’s a high confidence of a 3.
0.2 is < 0.5 so it’s a 7, and it’s fairly close to 0, so it’s a fair confidence of a 7.
0.6 is > 0.5 so it’s a 3, but it’s pretty far from 1, so it’s a low confidence of a 3.
0.4 is < 0.5 so it’s a 7, but it’s pretty far from 0, so it’s a low confidence of a 7.
1 Like
Check out the NLP lesson in the course. That has something similar.
1 Like
“new stuff” is a bit vague to provide specific advice on.
You should get most of what you need from… GitHub - bencoman/paperspace-setup: Setup a paperspace instance for fastai
1 Like
You’ve encoded the predicted-seven incorrectly. The meaning of prds is the “probability that it is a 3”, so 0.9 is high probability of a 3 and high probability of a 7 is 0.1.
Consider closely this paraphasing from 04_mnist_basics.ipynb:
- high confidence of a 3 ==> (
0.9) - slight confidence of a 7 ==> (
0.4) - good confidence of a 7 ==> (
0.2)
trgts = tensor([1,0,1])
prds = tensor([0.9, 0.4, 0.2])
For determining the 0.2, ignore the “although incorrect”. That statement refers only to the 0.2 not matching the “1” in the third element of trgts.
So basically low confidence score indicates not 3
Not exactly. Low confidence is values near 0.5, which indicates the model is “not sure” if its a 3 or 7.
High confidence is the two far ends, 0.9 or 0.1, being a high confidence of 3 or 7 respectively.
2 Likes
I’m not sure where to find the video you’ve watched, but a heads up just in case: you’re not currently inside of your WSL environment, powershell is still on your regular windows environment.
Here’s a useful stack overflow answer to your error (can be found by just searching the error string: “the term ‘git’ is not recognized as the name of a cmdlet…”)
This is happening either cause you’ve not installed git, or you haven’t set the path. You can install git here:
https://git-scm.com/book/en/v2/Getting-Started-Installing-Git
And there are instructions to setting the path in the stack overflow answer above.
1 Like
Lesson 4. Just looked it up now.
1 Like
I’m following the Fast AI program in this page: https://course.fast.ai/
On lesson 4 (Practical Deep Learning for Coders - 4: Natural Language (NLP)) we have a video which shows a code based on Chapter 10 of the book (https://www.amazon.com/Deep-Learning-Coders-fastai-PyTorch/dp/1492045527).
Chapter 10’s code is located here: fastbook/10_nlp.ipynb at master · fastai/fastbook · GitHub
I’m trying to find a way of downloading the above ipynb code.
Can you help me with some guidance?
thanks Dylan, slowly figuring out git and wsl environment, slowly but surely ill get there, all brand new!
thanks again ill check out the stack link!
You can click on “Raw”, the Right-Click > Save As…
Make sure the extension ends up “.ipynb” and not “.txt”…
Ahhh, I think I just got it. I may have had a similar question when I started about “How to get a blank workspace on paperspace”. You can get an empty the workspace from the terminal doing…
$ rm -r /notebooks
but actually in the short term, the advice was to not do that, and just create a folder “MyStuff”.
This is makes it easier to template off fastai notebooks.
1 Like
It worked. TY!
1 Like
No idea if I’m late to the party but please help me.
I’m unable to run the vision learner using any architecture(resnet 18 / 34).
To be honest I don’t even know what most of this means, I only tried to replicate what was there in the official lesson 1 Notebook on kaggle.
Uploading: Screenshot from 2022-09-25 22-26-37.png…
Hey, it’s super hard to find the error for your problem with the amount of information you gave. In order to do so you should at least show how you created your DataLoaders and Learner, better even share your whole notebook 
.
Generally: I think I encounterd that kind of error message with loss/metric issues, so the loss/metric function your learner is using doesnt fit the data you have or problem you want to solve.
How does the validation set function , i mean we
valid_pct=0.2
by default in fast.ai but where do i know how is this used ? it seems like a black box to me.
Can some one please explain as to how this works and where do i get to view how the validation set is used in the model?
Thanks for the suggestions @benkarr , I don’t know how to share the notebook directly thus I’ve uploaded all the screenshots if that could be of any help.