When reviewing the logs of my submissions, the sample test set test_series.parquet shows up in the submission log. I know because it's length is so short.
\nHow is it physically possible that Kaggle uses a hidden test set on the same run -
\nI can only think that they run my code twice, once with the sample test set - displaying that log, and then a second time with the hidden test set, and not share that log.
Does anyone know what Kaggle really does?
", "messages": [ { "id": "2513741", "postDate": "11/05/2023 17:02:35", "content": "When reviewing the logs of my submissions, the sample test set test_series.parquet shows up in the submission log. I know because it's length is so short.
\nHow is it physically possible that Kaggle uses a hidden test set on the same run -
\nI can only think that they run my code twice, once with the sample test set - displaying that log, and then a second time with the hidden test set, and not share that log.
Does anyone know what Kaggle really does?
", "rawMarkdown": "When reviewing the logs of my submissions, the sample test set test_series.parquet shows up in the submission log. I know because it's length is so short.\nHow is it physically possible that Kaggle uses a hidden test set on the same run - \nI can only think that they run my code twice, once with the sample test set - displaying that log, and then a second time with the hidden test set, and not share that log.\n\nDoes anyone know what Kaggle really does?", "votes": null }, { "id": "2514996", "postDate": "11/06/2023 16:52:38", "content": "I can only think that they run my code twice, once with the sample test set - displaying that log, and then a second time with the hidden test set, and not share that log.
Your answer is correct
", "rawMarkdown": "`I can only think that they run my code twice, once with the sample test set - displaying that log, and then a second time with the hidden test set, and not share that log.`\n\nYour answer is correct", "votes": null }, { "id": "2515277", "postDate": "11/06/2023 20:04:35", "content": "Your interpretation is correct. There are 3 ways to run a Kaggle notebook.
\nCommit mode (occurs when we click \"Save Version\" and) uses the sample test data and these are the notebooks that we view when we view our notebooks or public notebooks. These notebooks have logs describing what happened during commit mode. When we \"submit to competition\", the our code runs on Kaggle server in submit mode. In submit mode, our notebook uses the hidden test and we cannot see the log nor the resultant notebook. Interactive mode is when we have the notebook open and run cells in real time.
", "rawMarkdown": "Your interpretation is correct. There are 3 ways to run a Kaggle notebook. \n* interactive mode\n* commit mode\n* submit mode\n\nCommit mode (occurs when we click \"Save Version\" and) uses the sample test data and these are the notebooks that we view when we view our notebooks or public notebooks. These notebooks have logs describing what happened during commit mode. When we \"submit to competition\", the our code runs on Kaggle server in submit mode. In submit mode, our notebook uses the hidden test and we cannot see the log nor the resultant notebook. Interactive mode is when we have the notebook open and run cells in real time.", "votes": null }, { "id": "2515629", "postDate": "11/07/2023 04:47:09", "content": "Just to be clear, do I need to 'Save Version' for the latest working code to be used by kaggle in 'submit' mode? I would expect the answer is 'no', but I just want to be certain. I am having still getting submission scoring errors, and maybe the code I think I'm running is not what's actually being run by Kaggle?
\nThanks!
The most common reason for submit error is because when we submit, our code runs with the hidden test which is about 200 series ids. That is much larger than the sample test and can cause memory errors.
\nThe best why to find errors is to run (either interactively mode or commit mode) inference on all 277 train series id. If your inference code doesn't have error doing inference on 277 series id, then it will most like submit successfully.
\n\n\ndo I need to 'Save Version'
\n
The typical way to submit, is to first click \"save version\", then after commit finishes, then click \"submit to comp\" on the output showing submission.csv. (So commit model runs first then submit mode runs after). However, there is a button on the right sidebar inside the interactive notebook that says \"submit to comp\". If you click this, then it starts both a commit mode and submit mode at the same time. These are two different ways to submit, but neither would be the cause of a submit error. Both send the same code to Kaggle server. (Also note there are a bunch of other ways to submit also which also send the same code to Kaggle server).
", "rawMarkdown": "The most common reason for submit error is because when we submit, our code runs with the hidden test which is about 200 series ids. That is much larger than the sample test and can cause memory errors.\n\nThe best why to find errors is to run (either interactively mode or commit mode) inference on all 277 train series id. If your inference code doesn't have error doing inference on 277 series id, then it will most like submit successfully.\n\n>do I need to 'Save Version' \n\nThe typical way to submit, is to first click \"save version\", then after commit finishes, then click \"submit to comp\" on the output showing submission.csv. (So commit model runs first then submit mode runs after). However, there is a button on the right sidebar inside the interactive notebook that says \"submit to comp\". If you click this, then it starts both a commit mode and submit mode at the same time. These are two different ways to submit, but neither would be the cause of a submit error. Both send the same code to Kaggle server. (Also note there are a bunch of other ways to submit also which also send the same code to Kaggle server).", "votes": null } ], "comments": [ { "id": 2514996, "author_name": "fahmiayari", "author_url": "", "post_date": "11/06/2023 16:52:38", "content": "I can only think that they run my code twice, once with the sample test set - displaying that log, and then a second time with the hidden test set, and not share that log.
Your answer is correct
", "votes": null, "replies": [] }, { "id": 2515277, "author_name": "cdeotte", "author_url": "", "post_date": "11/06/2023 20:04:35", "content": "Your interpretation is correct. There are 3 ways to run a Kaggle notebook.
\nCommit mode (occurs when we click \"Save Version\" and) uses the sample test data and these are the notebooks that we view when we view our notebooks or public notebooks. These notebooks have logs describing what happened during commit mode. When we \"submit to competition\", the our code runs on Kaggle server in submit mode. In submit mode, our notebook uses the hidden test and we cannot see the log nor the resultant notebook. Interactive mode is when we have the notebook open and run cells in real time.
", "votes": null, "replies": [ { "id": 2515629, "author_name": "lazarp", "author_url": "", "post_date": "11/07/2023 04:47:09", "content": "Just to be clear, do I need to 'Save Version' for the latest working code to be used by kaggle in 'submit' mode? I would expect the answer is 'no', but I just want to be certain. I am having still getting submission scoring errors, and maybe the code I think I'm running is not what's actually being run by Kaggle?
\nThanks!
The most common reason for submit error is because when we submit, our code runs with the hidden test which is about 200 series ids. That is much larger than the sample test and can cause memory errors.
\nThe best why to find errors is to run (either interactively mode or commit mode) inference on all 277 train series id. If your inference code doesn't have error doing inference on 277 series id, then it will most like submit successfully.
\n\n\ndo I need to 'Save Version'
\n
The typical way to submit, is to first click \"save version\", then after commit finishes, then click \"submit to comp\" on the output showing submission.csv. (So commit model runs first then submit mode runs after). However, there is a button on the right sidebar inside the interactive notebook that says \"submit to comp\". If you click this, then it starts both a commit mode and submit mode at the same time. These are two different ways to submit, but neither would be the cause of a submit error. Both send the same code to Kaggle server. (Also note there are a bunch of other ways to submit also which also send the same code to Kaggle server).
", "votes": null, "replies": [] } ] } ] } ], "raw_markdown_by_id": { "2513741": "When reviewing the logs of my submissions, the sample test set test_series.parquet shows up in the submission log. I know because it's length is so short.\nHow is it physically possible that Kaggle uses a hidden test set on the same run - \nI can only think that they run my code twice, once with the sample test set - displaying that log, and then a second time with the hidden test set, and not share that log.\n\nDoes anyone know what Kaggle really does?", "2514996": "`I can only think that they run my code twice, once with the sample test set - displaying that log, and then a second time with the hidden test set, and not share that log.`\n\nYour answer is correct", "2515277": "Your interpretation is correct. There are 3 ways to run a Kaggle notebook. \n* interactive mode\n* commit mode\n* submit mode\n\nCommit mode (occurs when we click \"Save Version\" and) uses the sample test data and these are the notebooks that we view when we view our notebooks or public notebooks. These notebooks have logs describing what happened during commit mode. When we \"submit to competition\", the our code runs on Kaggle server in submit mode. In submit mode, our notebook uses the hidden test and we cannot see the log nor the resultant notebook. Interactive mode is when we have the notebook open and run cells in real time.", "2515629": "Just to be clear, do I need to 'Save Version' for the latest working code to be used by kaggle in 'submit' mode? I would expect the answer is 'no', but I just want to be certain. I am having still getting submission scoring errors, and maybe the code I think I'm running is not what's actually being run by Kaggle? \nThanks!", "2515643": "The most common reason for submit error is because when we submit, our code runs with the hidden test which is about 200 series ids. That is much larger than the sample test and can cause memory errors.\n\nThe best why to find errors is to run (either interactively mode or commit mode) inference on all 277 train series id. If your inference code doesn't have error doing inference on 277 series id, then it will most like submit successfully.\n\n>do I need to 'Save Version' \n\nThe typical way to submit, is to first click \"save version\", then after commit finishes, then click \"submit to comp\" on the output showing submission.csv. (So commit model runs first then submit mode runs after). However, there is a button on the right sidebar inside the interactive notebook that says \"submit to comp\". If you click this, then it starts both a commit mode and submit mode at the same time. These are two different ways to submit, but neither would be the cause of a submit error. Both send the same code to Kaggle server. (Also note there are a bunch of other ways to submit also which also send the same code to Kaggle server)." }, "source": "meta" }