{ "id": 397902, "title": "Value should be discrete(0 or 1) or it has to be probability of that even occuring(Between 0 to 1).", "url": "/competitions/tlvmc-parkinsons-freezing-gait-prediction/discussion/397902", "author_name": "Rishab", "post_date": "2023-03-27T17:35:39.606000", "votes": 4, "comment_count": 4, "views": 0, "content": "

Hey,
\nI am confused. Should the prediction contain discrete value(0 or 1) 0 meaning that event didn't occur and 1 meaning it did or the probability of that event occuring? (between 0 to 1)

", "messages": [ { "id": 2199427, "postDate": "2023-03-27T17:35:39.607Z", "content": "

Hey,
\nI am confused. Should the prediction contain discrete value(0 or 1) 0 meaning that event didn't occur and 1 meaning it did or the probability of that event occuring? (between 0 to 1)

", "rawMarkdown": "Hey, \nI am confused. Should the prediction contain discrete value(0 or 1) 0 meaning that event didn't occur and 1 meaning it did or the probability of that event occuring? (between 0 to 1)", "votes": 4 }, { "id": 2199448, "postDate": "2023-03-27T17:52:49.277Z", "content": "

Hi Rishab,

\n

The evaluation page contains the following:

\n
\n

For each Id in the test set, you must predict a confidence score for each of the three event types. In the ground truth, at most one event class has a non-zero value for each Id, but there is no restriction on the values of predicted scores. The predicted scores may, but are not required, to take the form of a probability (to be in the range 0.0 to 1.0, that is).

\n
\n

So I think you can do a binary 0/1 classification or return a probability of a data point being an event. e.g.: 0.79 probability that an event is a StartHesitation, where 1 indicates an event and 0 no event.

", "rawMarkdown": "Hi Rishab,\n\nThe [evaluation](https://www.kaggle.com/competitions/tlvmc-parkinsons-freezing-gait-prediction/overview/evaluation) page contains the following:\n\n>For each Id in the test set, you must predict a confidence score for each of the three event types. In the ground truth, at most one event class has a non-zero value for each Id, but there is no restriction on the values of predicted scores. The predicted scores may, but are not required, to take the form of a probability (to be in the range 0.0 to 1.0, that is).\n\nSo I think you can do a binary 0/1 classification or return a probability of a data point being an event. e.g.: 0.79 probability that an event is a `StartHesitation`, where 1 indicates an event and 0 no event.", "votes": 1, "replies": [ { "id": 2199451, "postDate": "2023-03-27T17:58:51.557Z", "content": "

I tried binary but i am getting a scoring error.(Basically i do predict_proba(predicting the probability of the class) and do np.argmax(predict_proba))Very weird, i am thinking of creating submitting notebook where i just submit the sample submission file and check if that is getting accepted. Till now i have used that as the standard for creating my submission file. Thoughts?

", "rawMarkdown": "I tried binary but i am getting a scoring error.(Basically i do predict_proba(predicting the probability of the class) and do np.argmax(predict_proba))Very weird, i am thinking of creating submitting notebook where i just submit the sample submission file and check if that is getting accepted. Till now i have used that as the standard for creating my submission file. Thoughts?", "votes": 1 } ] }, { "id": 2199609, "postDate": "2023-03-27T21:10:56.873Z", "rawMarkdown": "", "votes": 1, "isDeleted": true, "replies": [ { "id": 2199618, "postDate": "2023-03-27T21:22:52.213Z", "content": "

Makes sense. This is also explains why people are using regressors. Thank you so much for this, really helped.

", "rawMarkdown": "Makes sense. This is also explains why people are using regressors. Thank you so much for this, really helped.", "votes": 1 } ] } ], "comments": [ { "id": 2199448, "author_name": "Austin Hinkel", "author_url": "", "post_date": "2023-03-27T17:52:49.277000", "content": "

Hi Rishab,

\n

The evaluation page contains the following:

\n
\n

For each Id in the test set, you must predict a confidence score for each of the three event types. In the ground truth, at most one event class has a non-zero value for each Id, but there is no restriction on the values of predicted scores. The predicted scores may, but are not required, to take the form of a probability (to be in the range 0.0 to 1.0, that is).

\n
\n

So I think you can do a binary 0/1 classification or return a probability of a data point being an event. e.g.: 0.79 probability that an event is a StartHesitation, where 1 indicates an event and 0 no event.

", "votes": 1, "replies": [ { "id": 2199451, "author_name": "Rishab", "author_url": "", "post_date": "2023-03-27T17:58:51.557000", "content": "

I tried binary but i am getting a scoring error.(Basically i do predict_proba(predicting the probability of the class) and do np.argmax(predict_proba))Very weird, i am thinking of creating submitting notebook where i just submit the sample submission file and check if that is getting accepted. Till now i have used that as the standard for creating my submission file. Thoughts?

", "votes": 1, "replies": [] } ] }, { "id": 2199609, "author_name": "", "author_url": "", "post_date": "2023-03-27T21:10:56.873000", "content": "", "votes": 1, "replies": [ { "id": 2199618, "author_name": "Rishab", "author_url": "", "post_date": "2023-03-27T21:22:52.213000", "content": "

Makes sense. This is also explains why people are using regressors. Thank you so much for this, really helped.

", "votes": 1, "replies": [] } ] } ], "raw_markdown_by_id": { "2199427": "Hey, \nI am confused. Should the prediction contain discrete value(0 or 1) 0 meaning that event didn't occur and 1 meaning it did or the probability of that event occuring? (between 0 to 1)", "2199448": "Hi Rishab,\n\nThe [evaluation](https://www.kaggle.com/competitions/tlvmc-parkinsons-freezing-gait-prediction/overview/evaluation) page contains the following:\n\n>For each Id in the test set, you must predict a confidence score for each of the three event types. In the ground truth, at most one event class has a non-zero value for each Id, but there is no restriction on the values of predicted scores. The predicted scores may, but are not required, to take the form of a probability (to be in the range 0.0 to 1.0, that is).\n\nSo I think you can do a binary 0/1 classification or return a probability of a data point being an event. e.g.: 0.79 probability that an event is a `StartHesitation`, where 1 indicates an event and 0 no event.", "2199609": "" } }