rank_zero_only moved to lightning.fabric.utilities.rank_zero

[stroebel]

https://github.com/intro-stat-learning/ISLP/blob/bd0af3474fb4b6306c749a194b350199cd332a49/ISLP/torch/lightning.py#L17C51-L17C51

The fix for this is possibly to pin pytorch_lightning to <2.0 or to change the import to from pytorch_lightning.utilities.rank_zero import rank_zero_only.

I can confirm that the import at that location works. I haven't tested it against the lab for Ch10 yet though.

There is a similar issue for the seed_everything import for ch10. That can be handled by installing the entire lightning package and changing the import to from lightning_fabric.utilities.seed import seed_everything. Pinning pytorch_lightning might be the better course of action in this case.

[jonathan-taylor]

I think these imports were the earlier locations for these libraries. This import is fine for pytorch_lightning==2.0.6

What version of ISLP/ISLP_labs are you trying to run this on and encountering errors? Do you have a traceback?

In terms of pinning libraries, this is what tags on ISLP_labs will do, though I see these have somehow become unpinned on v2: https://github.com/intro-stat-learning/ISLP_labs/blob/v2/requirements.txt.

This is the frozen version for v2 https://github.com/intro-stat-learning/ISLP_labs/blob/v2_devel/requirements.txt -- when we push a new tag I will ensure that the requirements are frozen once again.

[stroebel]

I managed to some of the way through the Ch10 lab using the changes that I recommended above. I hit a roadblock again with lightning requiring Accuracy to be passed a positional argument.

It looks like I am a few patch versions behind for ISLP, I will update and see if that fixes my issue.

TypeError                                 Traceback (most recent call last)
[/Volumes/untitled/tx/learning/deep_learning.py](https://file+.vscode-resource.vscode-cdn.net/Volumes/untitled/tx/learning/deep_learning.py) in line 2
      [167](file:///Volumes/untitled/tx/learning/deep_learning.py?line=166) # %%
----> [168](file:///Volumes/untitled/tx/learning/deep_learning.py?line=167) mnist_module = SimpleModule.classification(mnist_model)
      [169](file:///Volumes/untitled/tx/learning/deep_learning.py?line=168) mnist_logger = CSVLogger("logs", name="MNIST")

File [~/mambaforge/envs/learning/lib/python3.10/site-packages/ISLP/torch/lightning.py:216](https://file+.vscode-resource.vscode-cdn.net/Volumes/untitled/tx/learning/~/mambaforge/envs/learning/lib/python3.10/site-packages/ISLP/torch/lightning.py:216), in SimpleModule.classification(model, metrics, device, **kwargs)
    [214](file:///Users/kolmogrov/mambaforge/envs/learning/lib/python3.10/site-packages/ISLP/torch/lightning.py?line=213) loss = nn.CrossEntropyLoss()
    [215](file:///Users/kolmogrov/mambaforge/envs/learning/lib/python3.10/site-packages/ISLP/torch/lightning.py?line=214) if 'accuracy' not in metrics:
--> [216](file:///Users/kolmogrov/mambaforge/envs/learning/lib/python3.10/site-packages/ISLP/torch/lightning.py?line=215)     metrics['accuracy'] = Accuracy()
    [217](file:///Users/kolmogrov/mambaforge/envs/learning/lib/python3.10/site-packages/ISLP/torch/lightning.py?line=216) if device is not None:
    [218](file:///Users/kolmogrov/mambaforge/envs/learning/lib/python3.10/site-packages/ISLP/torch/lightning.py?line=217)     for key, metric in metrics:

TypeError: Accuracy.__new__() missing 1 required positional argument: 'task'

ps. Thank you for your book. I have worked through the rest of the labs and they are great.

[jonathan-taylor]

This is another difference introduced since the PDF was built. You seem to have newer versions of torch libraries than ISLP

Try a fresh conda environment with these requirements:

https://github.com/intro-stat-learning/ISLP_labs/blob/v2.1/requirements.txt

Alternatively, pip install -U ISLP should give you a version that rubs with the newer torch libs

Get Outlook for Androidhttps://aka.ms/AAb9ysg

---

From: Jaco Stroebel @.> Sent: Monday, August 21, 2023 2:59:20 AM To: intro-stat-learning/ISLP @.> Cc: Jonathan Taylor @.>; Comment @.> Subject: Re: [intro-stat-learning/ISLP] rank_zero_only moved to lightning.fabric.utilities.rank_zero (Issue #2)

I managed to some of the way through the Ch10 lab using the changes that I recommended above. I hit a roadblock again with lightning requiring Accuracy to be passed a positional argument.

It looks like I am a few patch versions behind for ISLP, I will update and see if that fixes my issue.

[image]https://user-images.githubusercontent.com/5149902/261988209-de44a1f6-a0df-4e72-b77f-b8f6f3bdf4e5.png [image] https://user-images.githubusercontent.com/5149902/261988463-19f5626d-a1ba-4276-ada3-4b8a97aef9bd.png

TypeError Traceback (most recent call last) /Volumes/untitled/tx/learning/deep_learning.py in line 2 167 # %% ----> 168 mnist_module = SimpleModule.classification(mnist_model) 169 mnist_logger = CSVLogger("logs", name="MNIST")

File ~/mambaforge/envs/learning/lib/python3.10/site-packages/ISLP/torch/lightning.py:216, in SimpleModule.classification(model, metrics, device, **kwargs) 214 loss = nn.CrossEntropyLoss() 215 if 'accuracy' not in metrics: --> 216 metrics['accuracy'] = Accuracy() 217 if device is not None: 218 for key, metric in metrics:

TypeError: Accuracy.new() missing 1 required positional argument: 'task'

ps. Thank you for your book. I have worked through the rest of the labs and they are great.

— Reply to this email directly, view it on GitHubhttps://github.com/intro-stat-learning/ISLP/issues/2#issuecomment-1686024438, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AACTM2YPR6YGULH2OT4GAW3XWMWPRANCNFSM6AAAAAA3FFHLFA. You are receiving this because you commented.Message ID: @.***>

[jonathan-taylor]

BTW, there is also a forum at statlearning.com that has some of these issues.

I think issues on GitHub is easier but some may pop up there as well.

Get Outlook for Androidhttps://aka.ms/AAb9ysg

---

From: Jonathan E. Taylor @.> Sent: Monday, August 21, 2023 11:31:22 AM To: intro-stat-learning/ISLP @.>; intro-stat-learning/ISLP @.> Cc: Jonathan Taylor @.>; Comment @.***> Subject: Re: [intro-stat-learning/ISLP] rank_zero_only moved to lightning.fabric.utilities.rank_zero (Issue #2)

This is another difference introduced since the PDF was built. You seem to have newer versions of torch libraries than ISLP

Try a fresh conda environment with these requirements:

https://github.com/intro-stat-learning/ISLP_labs/blob/v2.1/requirements.txt

Alternatively, pip install -U ISLP should give you a version that rubs with the newer torch libs

Get Outlook for Androidhttps://aka.ms/AAb9ysg

---

From: Jaco Stroebel @.> Sent: Monday, August 21, 2023 2:59:20 AM To: intro-stat-learning/ISLP @.> Cc: Jonathan Taylor @.>; Comment @.> Subject: Re: [intro-stat-learning/ISLP] rank_zero_only moved to lightning.fabric.utilities.rank_zero (Issue #2)

I managed to some of the way through the Ch10 lab using the changes that I recommended above. I hit a roadblock again with lightning requiring Accuracy to be passed a positional argument.

It looks like I am a few patch versions behind for ISLP, I will update and see if that fixes my issue.

[image]https://user-images.githubusercontent.com/5149902/261988209-de44a1f6-a0df-4e72-b77f-b8f6f3bdf4e5.png [image] https://user-images.githubusercontent.com/5149902/261988463-19f5626d-a1ba-4276-ada3-4b8a97aef9bd.png

TypeError Traceback (most recent call last) /Volumes/untitled/tx/learning/deep_learning.py in line 2 167 # %% ----> 168 mnist_module = SimpleModule.classification(mnist_model) 169 mnist_logger = CSVLogger("logs", name="MNIST")

File ~/mambaforge/envs/learning/lib/python3.10/site-packages/ISLP/torch/lightning.py:216, in SimpleModule.classification(model, metrics, device, **kwargs) 214 loss = nn.CrossEntropyLoss() 215 if 'accuracy' not in metrics: --> 216 metrics['accuracy'] = Accuracy() 217 if device is not None: 218 for key, metric in metrics:

TypeError: Accuracy.new() missing 1 required positional argument: 'task'

ps. Thank you for your book. I have worked through the rest of the labs and they are great.

— Reply to this email directly, view it on GitHubhttps://github.com/intro-stat-learning/ISLP/issues/2#issuecomment-1686024438, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AACTM2YPR6YGULH2OT4GAW3XWMWPRANCNFSM6AAAAAA3FFHLFA. You are receiving this because you commented.Message ID: @.***>