![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8553208%2F08c166dcae68517b91ddcbe9beae21d0%2Foutput.png?generation=1682603465751859&alt=media\")
So I'm fairly new to machine learning, but very interested in this challenge. The question might sound easy for some advanced users, but I was wondering how we can handle data sections where no labels were given over a larger period of time? Below, I was plotting the sensor data and the labels over a time span with WALKING=0, TURN=0 and STARTHESITATION=0. Am I right, that in these sections no freezing of gait events occurred and should we consider establishing four outputs for the model to detect this \"non-events\"?
\n
So I'm fairly new to machine learning, but very interested in this challenge. The question might sound easy for some advanced users, but I was wondering how we can handle data sections where no labels were given over a larger period of time? Below, I was plotting the sensor data and the labels over a time span with WALKING=0, TURN=0 and STARTHESITATION=0. Am I right, that in these sections no freezing of gait events occurred and should we consider establishing four outputs for the model to detect this \"non-events\"?
\n
Hey, thank you very much for the detailed answer, following this explanation and reading some papers recently I wondered how we can handle the data distribution for training a multiclass model with the \"non-events\" attached. Isn't there a balance problem, when the \"non-events\" are more frequently present in the possible training data?
", "rawMarkdown": "Hey, thank you very much for the detailed answer, following this explanation and reading some papers recently I wondered how we can handle the data distribution for training a multiclass model with the \"non-events\" attached. Isn't there a balance problem, when the \"non-events\" are more frequently present in the possible training data?", "votes": 1, "replies": [ { "id": 2238577, "postDate": "2023-04-28T16:32:11.820Z", "rawMarkdown": "", "isDeleted": true }, { "id": 2238616, "postDate": "2023-04-28T17:09:38.463Z", "rawMarkdown": "", "isDeleted": true }, { "id": 2238618, "postDate": "2023-04-28T17:10:08.453Z", "content": "I'll take a look into the sklearn method, the next couple of days I'm going to try to preprocess the dataset to standardize and balance the data by downsampling the \"non-events\" I guess, so I can start training on the tdcsfog dataset. Thank you for your help, wish you all the best :)
", "rawMarkdown": "I'll take a look into the sklearn method, the next couple of days I'm going to try to preprocess the dataset to standardize and balance the data by downsampling the \"non-events\" I guess, so I can start training on the tdcsfog dataset. Thank you for your help, wish you all the best :)" } ] } ] } ], "comments": [ { "id": 2237740, "author_name": "", "author_url": "", "post_date": "2023-04-27T23:39:30.917000", "content": "", "votes": 0, "replies": [ { "id": 2238379, "author_name": "Lukas Marschall", "author_url": "", "post_date": "2023-04-28T13:20:35.387000", "content": "Hey, thank you very much for the detailed answer, following this explanation and reading some papers recently I wondered how we can handle the data distribution for training a multiclass model with the \"non-events\" attached. Isn't there a balance problem, when the \"non-events\" are more frequently present in the possible training data?
", "votes": 1, "replies": [ { "id": 2238577, "author_name": "", "author_url": "", "post_date": "2023-04-28T16:32:11.820000", "content": "", "votes": 0, "replies": [] }, { "id": 2238616, "author_name": "", "author_url": "", "post_date": "2023-04-28T17:09:38.463000", "content": "", "votes": 0, "replies": [] }, { "id": 2238618, "author_name": "Lukas Marschall", "author_url": "", "post_date": "2023-04-28T17:10:08.453000", "content": "I'll take a look into the sklearn method, the next couple of days I'm going to try to preprocess the dataset to standardize and balance the data by downsampling the \"non-events\" I guess, so I can start training on the tdcsfog dataset. Thank you for your help, wish you all the best :)
", "votes": 0, "replies": [] } ] } ] } ], "raw_markdown_by_id": { "2237264": "So I'm fairly new to machine learning, but very interested in this challenge. The question might sound easy for some advanced users, but I was wondering how we can handle data sections where no labels were given over a larger period of time? Below, I was plotting the sensor data and the labels over a time span with WALKING=0, TURN=0 and STARTHESITATION=0. Am I right, that in these sections no freezing of gait events occurred and should we consider establishing four outputs for the model to detect this \"non-events\"?\n\n", "2237740": "" } }