In local_dataset_utilities.py (e.g., Unit 8.2), the line of code:

np.bincount(df["label"].values)

should be changed into:

print(np.bincount(df["label"].values))

Hi,

The pdf link is not correct for Part 2: The Different Sequence Modeling Tasks

Thank you

https://lightning.ai/pages/courses/deep-learning-fundamentals/

where can I find unit6 videos?

Hello Rasbt,
Thank you for your awesome course.
I ran into above problem while running mlp_cli.py for section 6.5 . Is that parameter no longer available in lightning 2.0.2 ?

https://github.com/Lightning-AI/dl-fundamentals/blob/68bc6ca7716c315301723d41a66cfc34aa5d20dc/unit06-dl-tips/6.5-hparam-opt/6.5-part2-cli-simple/mlp_cli.py#LL18C36-L18C36

In multi-response questions, incorrectly not-marked options are always accepted as correct ones.
E.g., Quiz: 8.4 From RNNs to the Transformer Architecture (Part 1).

This may lead to a false understanding and should be corrected.

During testing, the loss is computed in test_step, but it is not used:

One can control loss computation as follows:

    def _shared_step(self, batch, compute_loss=True):
        features, true_labels = batch
        logits = self(features)

        loss = F.cross_entropy(logits, true_labels) if compute_loss else None
        predicted_labels = torch.argmax(logits, dim=1)
        return loss, true_labels, predicted_labels

    def test_step(self, batch, batch_idx):
        _, true_labels, predicted_labels = self._shared_step(batch, compute_loss=False)

Although, for teaching, it's likely simpler to keep _shared_step as is.

The slides seem wrong. The slides are inconsistent with the API document about the Callback class.
In the slides:
from pytorch_lightning.callbacks import Callback
Class MyCustomCallback(Callback):
...
In the "Callback API" document:
Class MyCustomCallback(lightning.Callback):
...
I defined MyCustomCallback class for exercise 2 and it give me runtime error at trainer = L.Trainer(...
When I used lightning.Callback, it works.

Discussed in #69

I noticed that some sections have only one video, but there is still a button labeled "Watch video 2" on those pages, even though there is no video 2.
See units: 1.3, 1.5, 2.1, 2.3, 2.5, 2.6, 2.7, 3.2, 3.4, 4.6, 5.1, 5.8, 6.4, 10.3, 10.4.

Here's a picture for unit 1.3:

Hi,

Im using the following custom data to reproduce this notebook.:

• test.csv
• train.csv
• val.csv

It is a multiclass label. However, I'm getting 0 loss and 0 accuracy. Any suggestion?
Thanks!

On section 1.6 about implementing a perceptron the URL for Pandas documentation has a typo. Instead of being, https://pandas.pydata.or it should have linked to https://pandas.pydata.org/. This is a very minor typo. And thank you for the great job with the course.

Hello, thanks for your course. I appreciate your work. Unfortunately, I ran into two issues navigating the course.

"Next Video" doesn't work.

When you are on the last video of a Unit subsection, there is a "Next Video" button, which is "slick-disabled" in the button class, but in fact, there is no difference from the "Previous Video" button in behavior. It feels like the button isn't working.

Expected result:
The button either leads to the next video or is missing because the video section is over.

Actual result:
The button is present but doesn't indicate about it can no longer be used.

Quiz number 2 without video 2.

It was kind of a surprise to see two quizzes in the first video. It triggered me to click "Next Video" so I determined that each subsection is represented with multiple videos like that:

Expected result:
Clean information about nesting within subsections.

Actual result:

Hi, I am currently pursuing this course and found it really great. The exercises at the end of each units/sub-units is a great addition in order to check the understanding. However I found out that the equations in the exercises are still in their LaTeX format.

Here's an example:

The expected notation should be $x_{3}^{[5]}$

Thank you for the aswesome courses. I wanted to express my appreciation for the engaging quizzes and exercises provided for each unit. They have been instrumental in keeping me motivated and actively learning throughout the course.

However, while working on Quiz 4.1, I came across a UI error that I would like to bring to your attention. Here are the specifies of the error:

Quiz 4.1, Dealing with More than Two Classes: Softmax Regression (PART 4) quiz option buttons don't work

When attempting to click my anwers for Quiz 4.1 (PART 4), I noticed that the "Radio button" options are unresponsive due to typo coding in UI. Therefore, clicking on the button does not trigger any action. I have attached a screenshot of the unresponsive "Radio buttion" option page for your reference.

Expected result:

• Ability to click on audio buttons
• Concise text answer options for questions

Actual result:

• The button does not trigger any action
• Typo text included in text answer options e.g. 1,-0,-0,-0,-0 -1" name="question-1" value=" 1,-0,-0,-0,-0 -1"> 1,-0,-0,-0,-0 -1"> 1, 0, 0, 0, 0

P.S. During the process of identifying and investigating a UI issue, I unintentionally discovered a method to easily bypass the correct answers in a quiz, enabling me to easily find out the answers. It may be necessary to implement measures to prevent such an approach.

The committed version of part2_distilbert-finetune-last-layers.ipynb contains error:

I believe the path should be something else than "csv".

On my Windows 10, with Python 3.11 and:

torch       : 2.1.2+cu121
transformers: 4.37.2
datasets    : 2.17.0
lightning   : 2.1.3

The code in many files of sections 8.2 and 8.7:

download_dataset()

df = load_dataset_into_to_dataframe()
partition_dataset(df)

result in:

     19 duration = time.time() - start_time
     20 progress_size = int(count * block_size)
---> 21 speed = progress_size / (1024.0**2 * duration)
     22 percent = count * block_size * 100.0 / total_size
     24 sys.stdout.write(
     25     f"\r{int(percent)}% | {progress_size / (1024.**2):.2f} MB "
     26     f"| {speed:.2f} MB/s | {duration:.2f} sec elapsed"
     27 )

ZeroDivisionError: float division by zero

I did not check other units yet.

The problem is that the computed difference in time is 0, so this creates the error.

The answer is incorrect. I think I did correct, but the feedback said it is "incorrect."

Discussed in #17