commonlit-evaluate-student-summaries

[vernal nova]

Anyone competing in Commonlit? 2 months to go!

[stiff turret]

Hi @vernal nova , I am. I'm kind of beginner to competitive data science. Hoping to learn a lot through this competition no matter the outcome😌 Any suggestions for a newbie like me?

[vernal nova]

That's great to hear! Welcome to competing 🙂 Hmm if you're just getting started, I'd recommend copying 1 or 2 of the most popular notebooks and read through their documentation and try to understand their output. Then, I'd tweak some of the methods and weightings myself to gain an understanding of how these affect the output.

[pallid bone]

Thanks for the @vernal nova , I'm also a newbie, will give give it a try

[ornate oracle]

I’m also on this. Have you made any submissions yet?

[wanton musk]

I'm also taking part in this competition, submitted a few models, but I'm struggling in doing better, and still at the bottom of the leaderboard. I'm still using old school machine learning techniques by trying to extract text features. I'm considering now using transformers 🤷‍♂️

[orchid harness]

Can we use BERT in this competition? Is it against the terms of this competition or not?

[hard wren]

you can use BERT

I have a very strange bug where the submission file has only 4 rows, do anyone knows how to fix?

https://www.kaggle.com/competitions/commonlit-evaluate-student-summaries/discussion/431446

[jolly zinc]

I am facing the "Submission csv not found" error after submission. Can anybody help in how to resolve the issue?

[faint sequoia]

Assuming "Submission csv not found" was not a typo on your part: file names usually don't have spaces in them.

[jolly zinc]

Yes they don't, it was a typo. "Submission.csv not found" this was the issue. Apologies from my end.

[alpine briar]

This is not a bug. In code competitions, the real test set is hidden and once you submit your notebook -- it runs on the actual test set which is much larger.

[alpine briar]

Try to run your notebook on a sample of the train as a proxy for the real test set and see what output it generates in the Kaggle env. That will help with debug

[jolly zinc]

In the Output folder of the Kaggle environment it is storing the "submission.csv" file. But after submitting, the error keeps popping up. This is the first time I am participating in a competition, so I don't have a lot of knowledge in the Kaggle Environment.

[alpine briar]

It probably errors out at some point when it runs the actual test set and so no submission.csv is generated. Try to run the same code on train like it is test and see what happens

[jolly zinc]

It saved the submission file in the interactive Kaggle notebook without any error.
Provided, this time I tested on the whole train data, just to be sure if this is happening due to size of the data or not.

[tawny spear]

Hi guys, do you have any ideas how to imporve performance of training? At the moment my 1 epoch is around 3 min, it's long, and model consist only of backbone + linear

[lyric plaza]

hi every one in competiton can we submit mutliple submissions files with different names or with same name

[jolly zinc]

Only Submission.csv will get evaluated.

[faint sequoia]

I think again you are inaccurate with names. It is "submission.csv" rather than "Submission.csv" as that single capitalized letter makes a difference.

[jolly zinc]

Yes, again apologies on my part.

[uneven reef]

Hello Everyone,

I have started this competition. I wanna learn about Roberta and everything basics in NLP that reqiures to learn Roberta. Can you provide me with resources or roadmap on where to start and what to learn? I couldn't find any good tutorials on Roberta. Can you suggest me resources on where to start and good platforms I can use to learn?

[gleaming kindle]

Hi @uneven reef , Just search for BERT variants , you can find several practical resources, such as:
https://github.com/facebookresearch/fairseq/tree/main/examples/roberta

or original paper:
https://arxiv.org/pdf/1907.11692.pdf

[uneven reef]

Thank you

[uneven reef]

Anyone wanna work together for this competition?

[pale blade]

use deberta v3

[round sundial]

Hi, I am Riya Saxena, 4th year UnderGrad @IITRoorkee. I am looking for teammates to participate with. Anyone in? (I have good experience in NLP.)

[lyric plaza]

CAN I JOIN

IAM ALSO A 4TH YEAR UNGERGRAD AT IIT DHANBAD

[dire onyx]

interested

[high heron]

Is this competition very similar to LLM Science Exam ?

🤔

[uneven reef]

Can we use RNN for this competition?

[undone dagger]

Hi, I am new to kaggle competitions and have some questions. Are we allowed to use a pretrained model like llama2 in this competition? How do I decide if a model is allowed based on the license? Can I train a model and upload to huggingface and use it? Thanks

[ember mortar]

Hi there,
Yes, you are allowed to use a pretrained model like llama2 in most Kaggle competitions. However, you should always check the competition rules carefully to make sure. The rules will usually specify which models are allowed and which are not.
To decide if a model is allowed based on the license, you need to read the license carefully. Most licenses will allow you to use the model for competitions, but there may be some restrictions. For example, some licenses may require you to give credit to the original authors of the model.
Yes, you can train a model and upload it to huggingface and use it in a Kaggle competition. However, you should make sure that the model is not violating any licenses. You should also make sure that the model is not plagiarized.

[lyric plaza]

self.model_name = "username/debertav3large"
self.model_dir = "/path/to/local/directory"

can anyone help me in getting username of debertav3large and its repository id

[tawny spear]

microsoft/deberta-v3-large

[ornate oracle]

Did anyone use BERT for this competition?

[bold cobalt]

deberta?

[simple field]

I think many people use variants/improvements of bert, like distilbert and roberta

because native Bert performs worse than the new variants iirc

[simple field]

Hey everyone, I recently made a discussion post on how to get started for beginners! You can check it out here: https://www.kaggle.com/competitions/commonlit-evaluate-student-summaries/discussion/436739

Let me know if you have any feedback or if I should add/change anything to the post!

[ornate oracle]

Thanks

[vocal moon]

Zoom is end:)
Video: https://youtu.be/L6OQmXk1Am4

Presentation: https://docs.google.com/presentation/d/101mmtjEvIwKKP2klTwJZ_yIGydXyorQkKn19SrMoZnE/edit?usp=sharing

Please @everyone come to Zoom: ⌚️ Thurday 7 September, 19.00 (CET time e.g. Paris time)
"Hands on Hugging Face Transformers and Kaggle CommonLit challenge"

[wise lynx]

hi folks, I recently posted a question here about the deberta v3 model,
https://www.kaggle.com/code/tsunotsuno/debertav3-baseline-content-and-wording-models/comments

please feel free to drop by and give it a look

[long drift]

too late to join?

[north hare]

I got 0.22 for content and 0.46 wording mse is this good or I need to improve ?

[hushed gull]

I have just completed implementing MobileBERT.

There are still some modifications to be made in preprocessing, and the result is not that great, but I believe putting this model on mobiles will be crucial for the real-life application of this service.

There are still many areas that has poor or no internet connections, and the children in such area shouldn't be neglected.
Please feel free to take a look!
https://www.kaggle.com/code/jasonheesanglee/updated-mobilebert-implementation

[prisma arch]

this looks really good, just make sure it is properly cross-validated and does not over fit