![\"old](\"https://i.ibb.co/pjqpfM0/supplemental-train.png]\")
Some people in the discussion forum noticed a drop off in score on the LB when submitting newer versions, which seems to be related to the change in supplemental_train.csv. I compared the new and old versions of the files and made a small sketch to show how the time intervals of the file changed:
\n
However, these are only the versions that are/were visible under the competition data. I believe that previously, the supplemental_train.csv file used when making a submission ended on June 13th, and I am not sure about changes in this file. But if changes in the hidden supplemental_train.csv occurred, this could also lead to changed scores if you build up data before the test set starts. To make sure to fetch data from right before the test set starts, you could just get the corresponding timeframe from the train.csv file, but make sure to change it back for the final submissions, as this won't work for the final submissions.
", "messages": [ { "id": "1654546", "postDate": "01/18/2022 15:59:26", "content": "Some people in the discussion forum noticed a drop off in score on the LB when submitting newer versions, which seems to be related to the change in supplemental_train.csv. I compared the new and old versions of the files and made a small sketch to show how the time intervals of the file changed:
\n
However, these are only the versions that are/were visible under the competition data. I believe that previously, the supplemental_train.csv file used when making a submission ended on June 13th, and I am not sure about changes in this file. But if changes in the hidden supplemental_train.csv occurred, this could also lead to changed scores if you build up data before the test set starts. To make sure to fetch data from right before the test set starts, you could just get the corresponding timeframe from the train.csv file, but make sure to change it back for the final submissions, as this won't work for the final submissions.
", "rawMarkdown": "Some people in the discussion forum noticed a drop off in score on the LB when submitting newer versions, which seems to be related to the change in supplemental_train.csv. I compared the new and old versions of the files and made a small sketch to show how the time intervals of the file changed: \n\n\n\nHowever, these are only the versions that are/were visible under the competition data. I believe that previously, the supplemental_train.csv file used when making a submission ended on June 13th, and I am not sure about changes in this file. But if changes in the hidden supplemental_train.csv occurred, this could also lead to changed scores if you build up data before the test set starts. To make sure to fetch data from right before the test set starts, you could just get the corresponding timeframe from the train.csv file, but make sure to change it back for the final submissions, as this won't work for the final submissions.", "votes": null }, { "id": "1654910", "postDate": "01/18/2022 23:53:07", "content": "One problem is that there ere are some wrong data in the new supplemental data. I compare the data with the actual market data, and I see that some values do not exist. For example, for one asset, its usual average close price is 50, but in the supplemental data, one row of this asset has the close price greater than 200 on Jan 2021 so that the target value reaches above 20. However, the close price of this asset has never reached above 200 during that specific minute by checking online, so I think the supplemental data recorded some wrong data.
", "rawMarkdown": "One problem is that there ere are some wrong data in the new supplemental data. I compare the data with the actual market data, and I see that some values do not exist. For example, for one asset, its usual average close price is 50, but in the supplemental data, one row of this asset has the close price greater than 200 on Jan 2021 so that the target value reaches above 20. However, the close price of this asset has never reached above 200 during that specific minute by checking online, so I think the supplemental data recorded some wrong data.", "votes": null }, { "id": "1654918", "postDate": "01/19/2022 00:20:36", "content": "Right, and it seems like it happend to multiple assets. Though I wonder where these spikes come from, in the past I have also seen some rare spikes in the charts of some cryptocurrency exchanges. I just hope it is safe to assume that this will not be the case in the final test set and they will fix it soon.
", "rawMarkdown": "Right, and it seems like it happend to multiple assets. Though I wonder where these spikes come from, in the past I have also seen some rare spikes in the charts of some cryptocurrency exchanges. I just hope it is safe to assume that this will not be the case in the final test set and they will fix it soon.", "votes": null } ], "comments": [ { "id": 1654910, "author_name": "yus002", "author_url": "", "post_date": "01/18/2022 23:53:07", "content": "One problem is that there ere are some wrong data in the new supplemental data. I compare the data with the actual market data, and I see that some values do not exist. For example, for one asset, its usual average close price is 50, but in the supplemental data, one row of this asset has the close price greater than 200 on Jan 2021 so that the target value reaches above 20. However, the close price of this asset has never reached above 200 during that specific minute by checking online, so I think the supplemental data recorded some wrong data.
", "votes": null, "replies": [ { "id": 1654918, "author_name": "fritzcremer", "author_url": "", "post_date": "01/19/2022 00:20:36", "content": "Right, and it seems like it happend to multiple assets. Though I wonder where these spikes come from, in the past I have also seen some rare spikes in the charts of some cryptocurrency exchanges. I just hope it is safe to assume that this will not be the case in the final test set and they will fix it soon.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1654546": "Some people in the discussion forum noticed a drop off in score on the LB when submitting newer versions, which seems to be related to the change in supplemental_train.csv. I compared the new and old versions of the files and made a small sketch to show how the time intervals of the file changed: \n\n\n\nHowever, these are only the versions that are/were visible under the competition data. I believe that previously, the supplemental_train.csv file used when making a submission ended on June 13th, and I am not sure about changes in this file. But if changes in the hidden supplemental_train.csv occurred, this could also lead to changed scores if you build up data before the test set starts. To make sure to fetch data from right before the test set starts, you could just get the corresponding timeframe from the train.csv file, but make sure to change it back for the final submissions, as this won't work for the final submissions.", "1654910": "One problem is that there ere are some wrong data in the new supplemental data. I compare the data with the actual market data, and I see that some values do not exist. For example, for one asset, its usual average close price is 50, but in the supplemental data, one row of this asset has the close price greater than 200 on Jan 2021 so that the target value reaches above 20. However, the close price of this asset has never reached above 200 during that specific minute by checking online, so I think the supplemental data recorded some wrong data.", "1654918": "Right, and it seems like it happend to multiple assets. Though I wonder where these spikes come from, in the past I have also seen some rare spikes in the charts of some cryptocurrency exchanges. I just hope it is safe to assume that this will not be the case in the final test set and they will fix it soon." }, "source": "meta" }