The test file series only contain 3 individuals, each have a time series of only 12.5 minutes (150 steps). This makes it a bit weird to predict both onset and wakeup, as the rules state that a sleep period must be at least 30 minutes…
\nSo, is there only 1 event (onset or wakeup, but not both) per individual?
", "messages": [ { "id": "2449645", "postDate": "09/21/2023 10:27:23", "content": "The test file series only contain 3 individuals, each have a time series of only 12.5 minutes (150 steps). This makes it a bit weird to predict both onset and wakeup, as the rules state that a sleep period must be at least 30 minutes…
\nSo, is there only 1 event (onset or wakeup, but not both) per individual?
", "rawMarkdown": "The test file series only contain 3 individuals, each have a time series of only 12.5 minutes (150 steps). This makes it a bit weird to predict both onset and wakeup, as the rules state that a sleep period must be at least 30 minutes...\n\nSo, is there only 1 event (onset or wakeup, but not both) per individual?", "votes": null }, { "id": "2449647", "postDate": "09/21/2023 10:30:11", "content": "The test_Series.parquet files is only a dummy file to quickly check if your code produces any errors before submission.
\nAfter submitting, this file will get exchanged with the true hidden test set (~200 individuals with roughly the same length as in the training set) and your code will be rerun against this new file for the final submission.
As in other competitions, the test data file is only given as an example of the file format and not useful for model processing. One simple option is to read in the train data as the test data, especially in this competition where the \"ground truth\" is not in the train or test series data files.
\nFor example, something like this:
\n<... your code ...>\n\n\n\ntest_rows = <your code here>\n\n test_rows > :\n REAL_TEST = \n ()\n:\n \n REAL_TEST = \n ()\n \n test_file_name = train_file_name\n <redo test parquet setup code using the modified test filename>\n\n<... your code ...>\nThen further in the code (at the end near submission) an if not(REAL_TEST): block can be run to compare the \"test\" results with the known train events to see how the model is doing.
Thanks for the clarification.
", "rawMarkdown": "Thanks for the clarification.", "votes": null }, { "id": "2452293", "postDate": "09/23/2023 07:51:28", "content": "Thanks for the clarification, and the suggestion.
", "rawMarkdown": "Thanks for the clarification, and the suggestion.", "votes": null }, { "id": "2465717", "postDate": "10/03/2023 08:21:40", "content": "Hi, I'm a rookie here and have no experience with competitions and and how usual test_files look like. Do you have a hunch whether the real test data is longer? If not I have to neglect the broader window around each step when making the prediction.
\nThanks!
", "rawMarkdown": "Hi, I'm a rookie here and have no experience with competitions and and how usual test_files look like. Do you have a hunch whether the real test data is longer? If not I have to neglect the broader window around each step when making the prediction.\n\nThanks!", "votes": null }, { "id": "2466962", "postDate": "10/04/2023 07:43:13", "content": "You can also check for
\nos.getenv('KAGGLE_IS_COMPETITION_RERUN')
This, is True for the submission run and False for the commit. No need to even access and check the test data.
", "rawMarkdown": "You can also check for \n`os.getenv('KAGGLE_IS_COMPETITION_RERUN')`\n\nThis, is True for the submission run and False for the commit. No need to even access and check the test data.", "votes": null } ], "comments": [ { "id": 2449647, "author_name": "patrickaiforfun", "author_url": "", "post_date": "09/21/2023 10:30:11", "content": "The test_Series.parquet files is only a dummy file to quickly check if your code produces any errors before submission.
\nAfter submitting, this file will get exchanged with the true hidden test set (~200 individuals with roughly the same length as in the training set) and your code will be rerun against this new file for the final submission.
Thanks for the clarification.
", "votes": null, "replies": [] } ] }, { "id": 2449929, "author_name": "dan3dewey", "author_url": "", "post_date": "09/21/2023 14:21:47", "content": "As in other competitions, the test data file is only given as an example of the file format and not useful for model processing. One simple option is to read in the train data as the test data, especially in this competition where the \"ground truth\" is not in the train or test series data files.
\nFor example, something like this:
\n<... your code ...>\n\n\n\ntest_rows = <your code here>\n\n test_rows > :\n REAL_TEST = \n ()\n:\n \n REAL_TEST = \n ()\n \n test_file_name = train_file_name\n <redo test parquet setup code using the modified test filename>\n\n<... your code ...>\nThen further in the code (at the end near submission) an if not(REAL_TEST): block can be run to compare the \"test\" results with the known train events to see how the model is doing.
Thanks for the clarification, and the suggestion.
", "votes": null, "replies": [] }, { "id": 2466962, "author_name": "ilu000", "author_url": "", "post_date": "10/04/2023 07:43:13", "content": "You can also check for
\nos.getenv('KAGGLE_IS_COMPETITION_RERUN')
This, is True for the submission run and False for the commit. No need to even access and check the test data.
", "votes": null, "replies": [] } ] }, { "id": 2465717, "author_name": "urbanandras", "author_url": "", "post_date": "10/03/2023 08:21:40", "content": "Hi, I'm a rookie here and have no experience with competitions and and how usual test_files look like. Do you have a hunch whether the real test data is longer? If not I have to neglect the broader window around each step when making the prediction.
\nThanks!
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "2449645": "The test file series only contain 3 individuals, each have a time series of only 12.5 minutes (150 steps). This makes it a bit weird to predict both onset and wakeup, as the rules state that a sleep period must be at least 30 minutes...\n\nSo, is there only 1 event (onset or wakeup, but not both) per individual?", "2449647": "The test_Series.parquet files is only a dummy file to quickly check if your code produces any errors before submission.\nAfter submitting, this file will get exchanged with the true hidden test set (~200 individuals with roughly the same length as in the training set) and your code will be rerun against this new file for the final submission.", "2449929": "As in other competitions, the test data file is only given as an example of the file format and not useful for model processing. One simple option is to read in the train data as the test data, especially in this competition where the \"ground truth\" is not in the train or test series data files.\n\nFor example, something like this:\n```\n<... your code ...>\n\n# Get and check the number of rows in the test file\n# (various ways to do this without loading all the test data)\ntest_rows =