#deep-past-initiative-machine-translation

1 messages · Page 1 of 1 (latest)

spice crag
#

Is there any scope for adding Sacrebleu to Kaggle's runtime environment for notebooks? It would be nice to be able to calculate a quick score estimate before submission.
https://pypi.org/project/sacrebleu/

hasty cape
spice crag
#

The rules of the competition are that we can’t have Internet connected notebooks doesn’t that stop this from being a permitted approach?

hasty cape
#

you can clone the repo into your notebook env (as in, set it as a 'input data source' and install from there)

grave burrow
#

@hasty cape - wondering about the use of tools like Claude Code to help generate code. The rules for this competition say to not share competition data with any entities not participating in the competition. Does that preclude me from using Claude Code in a local jupyter notebook? Thank you in advance, from a Kaggle newbie!

quaint reef
#

Hey all, how long are people’s successful submissions taking to run?

hasty cape
# grave burrow <@534659493439733761> - wondering about the use of tools like Claude Code to hel...

In 'normal usage' you are not really sending any competition data to anthropic, except maybe one or two examples, in which case I would assume shouldn't be problematic. you can make sure to not share any input prompts/data for training as per https://code.claude.com/docs/en/data-usage#data-policies (I hope you are not trying to get claude code to read the whole dataset and generate some code to train a model... ) disclaimer, I'm not one of the competition organisers, I'm just another participant..., so if you are really worried, you might need to ask on the kaggle discussion forum

ruby root
grave burrow
quaint reef
ebon pecan
#

I need help submitting my successful run. I have successfully completed "Save & Run All" multiple times, but my notebook does not appear as a selectable option when I click "Submit Prediction" on the Submissions tab.

My notebook runs offline (Internet OFF) and produces a submission.csv in the output folder. Has anyone else encountered this issue or is there a specific step I'm missing to make a version "eligible" for scoring?

ebon pecan
ebon pecan
#

I found that I had used up a 'successful notebook' slot by submitting a new notebook before uploading the actual code. And now my successful submissions has been running for 1+ hours.
Here's why the spinner is still going:
The Scale: While the "Commit" only processed 4 records, the real competition test set likely has thousands of rows (8,000+ indicated in the challenge details).
The Math: If your notebook takes ~3 seconds per record (based on your logs), then:
1,000 records = 50 minutes.
2,000 records = 1 hour 40 minutes.
The Hidden Run: Kaggle does not show the logs for the active Submission run to keep the test data private. You will only see the "Success" and the final score once it has translated every single record in the hidden file.

modest edge
#

It's in session options -- accelerator -- choose whichever you want, probably GPU P100

violet fern
#

Hi guys from where the dataset is to be downloaded

pure zinc
#

Hello all, I need 1-2 team mate for 2 competition. Interested individuals should have deep interest and expertise in modern deep learning architecture specially NLP and transformers. Must have access to GPU machine or cloud. Interested individuals fee free to dm me or mention me here.

sick thistle
#

Hello, everyone. This is my first competition. Have a wonderful day

dawn terrace
#

hey everyone, this is my first competition

dry crypt
#

Hey is anyone onlin ei need some idea on something

#

my question is regarding the analysis of the training dataset in itself

dawn terrace
dry crypt
#

so as of now to implement a model for this hackhathon i am working on this given train dataset but it is very big and manually working to make inferences, get domain knowledge, and build up paramets to input whne training the model seems like a hefty approach, is there any alternative you are taking up? like maybe to infer data on the train dataset better or using an entirely different source of data to train the model on for testing?

#

i have curated a manual dictionary and rules that i have inferred uptil now but working with 1561 rows and going through each row manually, translating it word by word manually, in this day and age with the speed we have, this seems like a very slow approach

#

any inputs?

#

@dawn terrace

dawn terrace
#

you can maybe pass your dictionary as context

glossy bluff
#

I'm using LightGBM+CatBoost+XGBoost regression+RecSys for this task

plain hinge
wooden glen
#

Hii guys can anybody help . I am new in kaggle competition and unable to make submission . I am getting this red flag

Cannot submit
Your Notebook cannot use internet access in this competition. Please disable internet in the Notebook editor and save a new version.

fleet river
#

disable internet acces in your session options before submit

#

and if you have to install packages use the package manager or offline wheels

dry crypt
#

im running a submission file and i have tried submitting it 5 times on different days with multiple tweaks but it is running into an error that says notebook threw an exception and there is no specific error output to work on , any help?

sharp nebula
#

hey guys any one want to team up for this i got 21.7 score in it on first try without any extra data now i have collected a data worth of 35000 rows which took me 1 and half week
if any one have good knowledge of dl and had made model regrading this problem lets team up and score high rank together
i will be able to work on it from next month cuz of exam right now

sharp nebula
glacial nebula
wanton tree
#

Looser

upbeat oxide
#

check my bio 😁

honest lotus
#

check my bio 😁