[BUG] - Disabling advanced/usb_semisup_learn

[svekars]

Add Link

https://pytorch.org/tutorials/advanced/usb_semisup_learn.html

Describe the bug

In the current shape, the tutorial takes 143 minutes to build which increases the time of the whole tutorial build by 2 hours.
To enable this tutorial, we need to reduce the execution time to a reasonable number - most of the tutorials take less than 30 minutes to build. This can be achieved by reducing the number of training iterations to a reasonable number or optimizing other parts of the code.

After that, the tutorial can be enabled in the .jenkins/validate_tutorials_built.py script

There is also an image that is not rendering correctly:

Describe your environment

N/A

[atskae]

There is also an image that is not rendering correctly:

@Hhhhhhao @svekars What was the intended image in the tutorial?

[Hhhhhhao]

There is also an image that is not rendering correctly:

@Hhhhhhao @svekars What was the intended image in the tutorial?

The image should be this one: https://github.com/microsoft/Semi-supervised-learning/blob/main/figures/code.png
I don't know why it is not rendered.

To reduce the training iterations, we can reduce this number:

tutorials/advanced_source/usb_semisup_learn.py

Line 95 in 082c8b1

'num_train_iter': 4000,

[atskae]

The image should be this one: https://github.com/microsoft/Semi-supervised-learning/blob/main/figures/code.png

Would it be ok if I link the image to that URL directly in the notebook?

To reduce the training iterations, we can reduce this number:

I will try reducing that value and see what is reasonable.

[Hhhhhhao]

The image should be this one: https://github.com/microsoft/Semi-supervised-learning/blob/main/figures/code.png

Would it be ok if I link the image to that URL directly in the notebook?

To reduce the training iterations, we can reduce this number:

I will try reducing that value and see what is reasonable.

Yeah. Using the link directly would be fine.
I guess changing the iteration numbers to 1000 would be fine?

[atskae]

I guess changing the iteration numbers to 1000 would be fine?

I tried 500 earlier and it ran for 3 minutes with an accuracy of 87% for both FreeMatch and SoftMatch. Trying 1000 now.

[Hhhhhhao]

I guess changing the iteration numbers to 1000 would be fine?

I tried 500 earlier and it ran for 3 minutes with an accuracy of 87% for both FreeMatch and SoftMatch. Trying 1000 now.

if that's the case I think 500 would be good.

[atskae]

Interesting, I just ran with 1000 iterations for FreeMatch (ran for 8 minutes) and the accuracy went slightly down to 86%.