I'm just trying to understand the procedure for making a submission in this competition (which is quite different to anything else I've done here on kaggle):
\nThe code to access the API is in the Demo Submission Notebook.
\nMy understanding is that the predict function that we need to fill out takes one sequence at a time (i.e. one sequence_id, and hence one subject_id at a time). So, the dataframe that we need to process will look similar to that from test.csv.
\nAnd from each sequence processed (i.e. each test.csv-like dataframe) we get one gesture class prediction so the predict function returns one string.
\nIs that right?
", "messages": [ { "id": "3219236", "postDate": "06/07/2025 10:58:48", "content": "I'm just trying to understand the procedure for making a submission in this competition (which is quite different to anything else I've done here on kaggle):
\nThe code to access the API is in the Demo Submission Notebook.
\nMy understanding is that the predict function that we need to fill out takes one sequence at a time (i.e. one sequence_id, and hence one subject_id at a time). So, the dataframe that we need to process will look similar to that from test.csv.
\nAnd from each sequence processed (i.e. each test.csv-like dataframe) we get one gesture class prediction so the predict function returns one string.
\nIs that right?
", "rawMarkdown": "I'm just trying to understand the procedure for making a submission in this competition (which is quite different to anything else I've done here on kaggle):\n\nThe code to access the API is in the [Demo Submission Notebook](https://www.kaggle.com/code/sohier/cmi-2025-demo-submission). \n\nMy understanding is that the predict function that we need to fill out takes one sequence at a time (i.e. one sequence_id, and hence one subject_id at a time). So, the dataframe that we need to process will look similar to that from test.csv. \n\nAnd from each sequence processed (i.e. each test.csv-like dataframe) we get one gesture class prediction so the predict function returns one string.\n\nIs that right?", "votes": null }, { "id": "3219270", "postDate": "06/07/2025 11:53:41", "content": "You are required to implement the predict function to handle a specific type of input: a single sequence and demographics_demo. The function should output a single gesture prediction corresponding to that sequence.
", "rawMarkdown": "You are required to implement the predict function to handle a specific type of input: a single sequence and demographics_demo. The function should output a single gesture prediction corresponding to that sequence.", "votes": null }, { "id": "3219809", "postDate": "06/08/2025 10:41:38", "content": "Thanks shanzhong8.
", "rawMarkdown": "Thanks shanzhong8.", "votes": null }, { "id": "3219814", "postDate": "06/08/2025 10:50:34", "content": "I successfully managed to make some predictions! As shanzhong8 points out, the predict function takes a single sequence and the associated demographics data (which I didn't use yet). The data passed in will be like the test.csv set but just for one sequence_id. The sample test.csv provided as input has two sequence_ids, so you can mimic the input into predict() by splitting the test.csv data file into two sequences. Column names/types in the data served up by the API will be the same as test.csv.
\nHere are some tips that might help others who are struggling with the submission API:
\nEven though the documentation in the Demo Submission Notebook says that the predict function can accept polars or pandas, I had problems with using pandas dataframes as somewhere along the line my pandas got converted into a polars dataframe. So it was easier to change my data processing workflow to use polars dataframes in the first instance.
\nThe function return value should be a single string (i.e. the classified gesture). I kept getting <GatewayRuntimeErrorType.SERVER_RAISED_EXCEPTION: 3> and this was because my function was returning a np.str_ value, which, although it looks similar enough, is different enough from a base python str to cause problems.
You are required to implement the predict function to handle a specific type of input: a single sequence and demographics_demo. The function should output a single gesture prediction corresponding to that sequence.
", "votes": null, "replies": [ { "id": 3219809, "author_name": "nnjjpp", "author_url": "", "post_date": "06/08/2025 10:41:38", "content": "Thanks shanzhong8.
", "votes": null, "replies": [] } ] }, { "id": 3219814, "author_name": "nnjjpp", "author_url": "", "post_date": "06/08/2025 10:50:34", "content": "I successfully managed to make some predictions! As shanzhong8 points out, the predict function takes a single sequence and the associated demographics data (which I didn't use yet). The data passed in will be like the test.csv set but just for one sequence_id. The sample test.csv provided as input has two sequence_ids, so you can mimic the input into predict() by splitting the test.csv data file into two sequences. Column names/types in the data served up by the API will be the same as test.csv.
\nHere are some tips that might help others who are struggling with the submission API:
\nEven though the documentation in the Demo Submission Notebook says that the predict function can accept polars or pandas, I had problems with using pandas dataframes as somewhere along the line my pandas got converted into a polars dataframe. So it was easier to change my data processing workflow to use polars dataframes in the first instance.
\nThe function return value should be a single string (i.e. the classified gesture). I kept getting <GatewayRuntimeErrorType.SERVER_RAISED_EXCEPTION: 3> and this was because my function was returning a np.str_ value, which, although it looks similar enough, is different enough from a base python str to cause problems.