Hi,
\nI'm having difficulties with some of the assumptions I made about the data, though I would like clarification if possible. The train.csv data had no missing values for any of the roster data on any date (including offseason). The example_test.csv file was the same.
However, it seems as though there are missing values for the rosters column in the private test set. The notebook I created tested this by implementing a 10 minute delay and raising an error if any of the test data were missing values in the rosters column. Running the notebook on the last five days of April did not yield a delay and error; running it on the private test set did.
I'm not entirely certain what other assumptions about the private test set may be false if all rosters data were present in train.csv but not in the private set. I'm also not certain if I can create rosters data from other columns if they may be missing, too. Could somebody please clarify if my assumptions about the private data are grossly off, and if the rosters data is not meant to be in the private test set?
Thank you!
", "messages": [ { "id": "1353181", "postDate": "06/17/2021 00:52:36", "content": "Hi,
\nI'm having difficulties with some of the assumptions I made about the data, though I would like clarification if possible. The train.csv data had no missing values for any of the roster data on any date (including offseason). The example_test.csv file was the same.
However, it seems as though there are missing values for the rosters column in the private test set. The notebook I created tested this by implementing a 10 minute delay and raising an error if any of the test data were missing values in the rosters column. Running the notebook on the last five days of April did not yield a delay and error; running it on the private test set did.
I'm not entirely certain what other assumptions about the private test set may be false if all rosters data were present in train.csv but not in the private set. I'm also not certain if I can create rosters data from other columns if they may be missing, too. Could somebody please clarify if my assumptions about the private data are grossly off, and if the rosters data is not meant to be in the private test set?
Thank you!
", "rawMarkdown": "Hi,\n\nI'm having difficulties with some of the assumptions I made about the data, though I would like clarification if possible. The `train.csv` data had no missing values for any of the roster data on any date (including offseason). The `example_test.csv` file was the same.\n\nHowever, it seems as though there are missing values for the `rosters` column in the private test set. The [notebook I created](https://www.kaggle.com/nczeak/mlb-private-test-column-check-1) tested this by implementing a 10 minute delay and raising an error if any of the test data were missing values in the `rosters` column. Running the notebook on the last five days of April did not yield a delay and error; running it on the private test set did.\n\nI'm not entirely certain what other assumptions about the private test set may be false if all `rosters` data were present in `train.csv` but not in the private set. I'm also not certain if I can create `rosters` data from other columns if they may be missing, too. Could somebody please clarify if my assumptions about the private data are grossly off, and if the `rosters` data is not meant to be in the private test set?\n\nThank you!", "votes": null }, { "id": "1357020", "postDate": "06/19/2021 12:12:54", "content": "There are now multiple notebooks reporting submission errors on otherwise functional notebooks. Moderators - could we please get somebody to look into this? It would be greatly appreciated! Thanks in advance!
\n", "rawMarkdown": "There are now [multiple](https://www.kaggle.com/c/mlb-player-digital-engagement-forecasting/discussion/246111) [notebooks](https://www.kaggle.com/c/mlb-player-digital-engagement-forecasting/discussion/247340) reporting submission errors on otherwise functional notebooks. Moderators - could we please get somebody to look into this? It would be greatly appreciated! Thanks in advance!\n\n@juliaelliott \n@wcukierski", "votes": null }, { "id": "1359821", "postDate": "06/21/2021 14:52:26", "content": "Hey Nick,
\nPlease check out https://www.kaggle.com/code-competition-debugging for more info.
\nFor context, it's typical in these style of competitions to have multiple reports of submission errors. It's the tradeoff we make for having an unseen test set with a locked down debugging loop. It's also a healthy, albeit frustrating, process that helps inoculate code against the vagaries of future data; we want to shake out the surprises now so that a higher percentage of notebooks succeed when we run them. Of the reports we receive about errors, some are newcomers figuring out how submission works, some are code bugs the author hasn't spotted, some are isolated data consistency surprises, and some are legit structural competition issues. We try to intervene if we suspect the latter is happening.
\nFor your specific suspicion of missing roster values, I recommend making your notebook robust to missing values in any of the fields. There are lots of ways to do this. This function decorators approach is one clever way to do it. Good luck!
", "rawMarkdown": "Hey Nick,\n\nPlease check out https://www.kaggle.com/code-competition-debugging for more info.\n\nFor context, it's typical in these style of competitions to have multiple reports of submission errors. It's the tradeoff we make for having an unseen test set with a locked down debugging loop. It's also a healthy, albeit frustrating, process that helps inoculate code against the vagaries of future data; we want to shake out the surprises now so that a higher percentage of notebooks succeed when we run them. Of the reports we receive about errors, some are newcomers figuring out how submission works, some are code bugs the author hasn't spotted, some are isolated data consistency surprises, and some are legit structural competition issues. We try to intervene if we suspect the latter is happening.\n\nFor your specific suspicion of missing roster values, I recommend making your notebook robust to missing values in any of the fields. There are lots of ways to do this. This [function decorators](https://www.kaggle.com/c/nfl-big-data-bowl-2020/discussion/120375#688496) approach is one clever way to do it. Good luck!", "votes": null } ], "comments": [ { "id": 1357020, "author_name": "nczeak", "author_url": "", "post_date": "06/19/2021 12:12:54", "content": "There are now multiple notebooks reporting submission errors on otherwise functional notebooks. Moderators - could we please get somebody to look into this? It would be greatly appreciated! Thanks in advance!
\n", "votes": null, "replies": [ { "id": 1359821, "author_name": "wcukierski", "author_url": "", "post_date": "06/21/2021 14:52:26", "content": "Hey Nick,
\nPlease check out https://www.kaggle.com/code-competition-debugging for more info.
\nFor context, it's typical in these style of competitions to have multiple reports of submission errors. It's the tradeoff we make for having an unseen test set with a locked down debugging loop. It's also a healthy, albeit frustrating, process that helps inoculate code against the vagaries of future data; we want to shake out the surprises now so that a higher percentage of notebooks succeed when we run them. Of the reports we receive about errors, some are newcomers figuring out how submission works, some are code bugs the author hasn't spotted, some are isolated data consistency surprises, and some are legit structural competition issues. We try to intervene if we suspect the latter is happening.
\nFor your specific suspicion of missing roster values, I recommend making your notebook robust to missing values in any of the fields. There are lots of ways to do this. This function decorators approach is one clever way to do it. Good luck!
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1353181": "Hi,\n\nI'm having difficulties with some of the assumptions I made about the data, though I would like clarification if possible. The `train.csv` data had no missing values for any of the roster data on any date (including offseason). The `example_test.csv` file was the same.\n\nHowever, it seems as though there are missing values for the `rosters` column in the private test set. The [notebook I created](https://www.kaggle.com/nczeak/mlb-private-test-column-check-1) tested this by implementing a 10 minute delay and raising an error if any of the test data were missing values in the `rosters` column. Running the notebook on the last five days of April did not yield a delay and error; running it on the private test set did.\n\nI'm not entirely certain what other assumptions about the private test set may be false if all `rosters` data were present in `train.csv` but not in the private set. I'm also not certain if I can create `rosters` data from other columns if they may be missing, too. Could somebody please clarify if my assumptions about the private data are grossly off, and if the `rosters` data is not meant to be in the private test set?\n\nThank you!", "1357020": "There are now [multiple](https://www.kaggle.com/c/mlb-player-digital-engagement-forecasting/discussion/246111) [notebooks](https://www.kaggle.com/c/mlb-player-digital-engagement-forecasting/discussion/247340) reporting submission errors on otherwise functional notebooks. Moderators - could we please get somebody to look into this? It would be greatly appreciated! Thanks in advance!\n\n@juliaelliott \n@wcukierski", "1359821": "Hey Nick,\n\nPlease check out https://www.kaggle.com/code-competition-debugging for more info.\n\nFor context, it's typical in these style of competitions to have multiple reports of submission errors. It's the tradeoff we make for having an unseen test set with a locked down debugging loop. It's also a healthy, albeit frustrating, process that helps inoculate code against the vagaries of future data; we want to shake out the surprises now so that a higher percentage of notebooks succeed when we run them. Of the reports we receive about errors, some are newcomers figuring out how submission works, some are code bugs the author hasn't spotted, some are isolated data consistency surprises, and some are legit structural competition issues. We try to intervene if we suspect the latter is happening.\n\nFor your specific suspicion of missing roster values, I recommend making your notebook robust to missing values in any of the fields. There are lots of ways to do this. This [function decorators](https://www.kaggle.com/c/nfl-big-data-bowl-2020/discussion/120375#688496) approach is one clever way to do it. Good luck!" }, "source": "meta" }