even if I make the simplest of submission it takes around 1-2 hours to score my submission . Does anyone share the same problem ??
", "messages": [ { "id": "1571684", "postDate": "11/05/2021 04:49:50", "content": "even if I make the simplest of submission it takes around 1-2 hours to score my submission . Does anyone share the same problem ??
", "rawMarkdown": "even if I make the simplest of submission it takes around 1-2 hours to score my submission . Does anyone share the same problem ??", "votes": null }, { "id": "1571761", "postDate": "11/05/2021 06:21:10", "content": "I think this is obvious. Since the management of the competition is trying to avoid peeking onto future data, they only feed one line of data at a time. There's also the \"This version of the API is not optimized and should not be used to estimate the runtime of your code on the hidden test set.\" statement when using the API so you better take it to mind that it will be slow.
", "rawMarkdown": "I think this is obvious. Since the management of the competition is trying to avoid peeking onto future data, they only feed one line of data at a time. There's also the \"This version of the API is not optimized and should not be used to estimate the runtime of your code on the hidden test set.\" statement when using the API so you better take it to mind that it will be slow.", "votes": null }, { "id": "1572063", "postDate": "11/05/2021 11:35:57", "content": "The slow API is necessary to ensure no time leakage. We just had a time serie competition (Optiver) without such Time Serie API. At first I was happy about fast submission process. But then had a mitigated experience because both public and private LB were rewarding leakage. Most of my attempts at feature engineering / CV or model design were dwarfed in performance compared to introducing leakage. It was quite difficult to know what was good or bad. Honestly, I am quite happy to get back to a good TS API, because it is clearer that we are building somewhat good models. (Plus working on faster inference is a good engineering challenge).
", "rawMarkdown": "The slow API is necessary to ensure no time leakage. We just had a time serie competition (Optiver) without such Time Serie API. At first I was happy about fast submission process. But then had a mitigated experience because both public and private LB were rewarding leakage. Most of my attempts at feature engineering / CV or model design were dwarfed in performance compared to introducing leakage. It was quite difficult to know what was good or bad. Honestly, I am quite happy to get back to a good TS API, because it is clearer that we are building somewhat good models. (Plus working on faster inference is a good engineering challenge).", "votes": null }, { "id": "1572499", "postDate": "11/05/2021 17:58:38", "content": "With my model its taking around 8hrs. Any suggestions welcome. I tried to increase estimators and process dies
", "rawMarkdown": "With my model its taking around 8hrs. Any suggestions welcome. I tried to increase estimators and process dies", "votes": null } ], "comments": [ { "id": 1571761, "author_name": "seraphwedd18", "author_url": "", "post_date": "11/05/2021 06:21:10", "content": "I think this is obvious. Since the management of the competition is trying to avoid peeking onto future data, they only feed one line of data at a time. There's also the \"This version of the API is not optimized and should not be used to estimate the runtime of your code on the hidden test set.\" statement when using the API so you better take it to mind that it will be slow.
", "votes": null, "replies": [] }, { "id": 1572063, "author_name": "lucasmorin", "author_url": "", "post_date": "11/05/2021 11:35:57", "content": "The slow API is necessary to ensure no time leakage. We just had a time serie competition (Optiver) without such Time Serie API. At first I was happy about fast submission process. But then had a mitigated experience because both public and private LB were rewarding leakage. Most of my attempts at feature engineering / CV or model design were dwarfed in performance compared to introducing leakage. It was quite difficult to know what was good or bad. Honestly, I am quite happy to get back to a good TS API, because it is clearer that we are building somewhat good models. (Plus working on faster inference is a good engineering challenge).
", "votes": null, "replies": [] }, { "id": 1572499, "author_name": "neoistheone", "author_url": "", "post_date": "11/05/2021 17:58:38", "content": "With my model its taking around 8hrs. Any suggestions welcome. I tried to increase estimators and process dies
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1571684": "even if I make the simplest of submission it takes around 1-2 hours to score my submission . Does anyone share the same problem ??", "1571761": "I think this is obvious. Since the management of the competition is trying to avoid peeking onto future data, they only feed one line of data at a time. There's also the \"This version of the API is not optimized and should not be used to estimate the runtime of your code on the hidden test set.\" statement when using the API so you better take it to mind that it will be slow.", "1572063": "The slow API is necessary to ensure no time leakage. We just had a time serie competition (Optiver) without such Time Serie API. At first I was happy about fast submission process. But then had a mitigated experience because both public and private LB were rewarding leakage. Most of my attempts at feature engineering / CV or model design were dwarfed in performance compared to introducing leakage. It was quite difficult to know what was good or bad. Honestly, I am quite happy to get back to a good TS API, because it is clearer that we are building somewhat good models. (Plus working on faster inference is a good engineering challenge).", "1572499": "With my model its taking around 8hrs. Any suggestions welcome. I tried to increase estimators and process dies" }, "source": "meta" }