{ "id": 79593, "title": "URGENT. How much runtime buffer are you leaving for stage 2 ?", "url": "/competitions/quora-insincere-questions-classification/discussion/79593", "author_name": "", "post_date": "2019-02-05T20:46:03.768038Z", "votes": null, "comment_count": 5, "views": 0, "content": "

Hello !\nHow much runtime buffer are you guys leaving for stage 2 evaluation.

", "messages": [ { "id": "466698", "postDate": "02/05/2019 20:46:03", "content": "

Hello !\nHow much runtime buffer are you guys leaving for stage 2 evaluation.

", "rawMarkdown": "Hello !\nHow much runtime buffer are you guys leaving for stage 2 evaluation.", "votes": null }, { "id": "466717", "postDate": "02/05/2019 21:18:22", "content": "

Why are people downvoting this ? Is there anything wrong in this question ?

", "rawMarkdown": "Why are people downvoting this ? Is there anything wrong in this question ?", "votes": null }, { "id": "466721", "postDate": "02/05/2019 21:30:42", "content": "

Highly depends on your preprocessing and number of times the predictions are made of the test data. \nLet's say now you are taking 2 mins for preprocessing test data and making the predictions 5 time (30 secs each)

\n\n

So you should atleast save (2*60 + 5*30)*7 secs, since the 2nd Stage would be 7 times bigger. A save bet would be to have (2*60 + 5*30)*9 secs buffer to incorporate some other stuff, like if you are using tokenizer on test data as well as train data.

\n\n

Hope this helps. Although it's a little late now.

", "rawMarkdown": "Highly depends on your preprocessing and number of times the predictions are made of the test data. \nLet's say now you are taking 2 mins for preprocessing test data and making the predictions 5 time (30 secs each)\n\nSo you should atleast save (2*60 + 5*30)*7 secs, since the 2nd Stage would be 7 times bigger. A save bet would be to have (2*60 + 5*30)*9 secs buffer to incorporate some other stuff, like if you are using tokenizer on test data as well as train data.\n\nHope this helps. Although it's a little late now.", "votes": null }, { "id": "466794", "postDate": "02/05/2019 23:55:19", "content": "

I mark this question sincere

", "rawMarkdown": "I mark this question sincere", "votes": null }, { "id": "466797", "postDate": "02/05/2019 23:59:36", "content": "

To answer your question we simulated the increased test size by repeating the test file 6 more times. That is probably the easiest way to get an estimate. We are still allowing for some small buffer though just in case.

", "rawMarkdown": "To answer your question we simulated the increased test size by repeating the test file 6 more times. That is probably the easiest way to get an estimate. We are still allowing for some small buffer though just in case.", "votes": null }, { "id": "468143", "postDate": "02/08/2019 11:06:32", "content": "

I tested my kernel on the new test data, and it runs in time (with 200s to spare). I'm so relieved. Phew !

", "rawMarkdown": "I tested my kernel on the new test data, and it runs in time (with 200s to spare). I'm so relieved. Phew !", "votes": null } ], "comments": [ { "id": 466717, "author_name": "tarunpaparaju", "author_url": "", "post_date": "02/05/2019 21:18:22", "content": "

Why are people downvoting this ? Is there anything wrong in this question ?

", "votes": null, "replies": [ { "id": 466794, "author_name": "ryches", "author_url": "", "post_date": "02/05/2019 23:55:19", "content": "

I mark this question sincere

", "votes": null, "replies": [] } ] }, { "id": 466721, "author_name": "abdurrafae", "author_url": "", "post_date": "02/05/2019 21:30:42", "content": "

Highly depends on your preprocessing and number of times the predictions are made of the test data. \nLet's say now you are taking 2 mins for preprocessing test data and making the predictions 5 time (30 secs each)

\n\n

So you should atleast save (2*60 + 5*30)*7 secs, since the 2nd Stage would be 7 times bigger. A save bet would be to have (2*60 + 5*30)*9 secs buffer to incorporate some other stuff, like if you are using tokenizer on test data as well as train data.

\n\n

Hope this helps. Although it's a little late now.

", "votes": null, "replies": [] }, { "id": 466797, "author_name": "ryches", "author_url": "", "post_date": "02/05/2019 23:59:36", "content": "

To answer your question we simulated the increased test size by repeating the test file 6 more times. That is probably the easiest way to get an estimate. We are still allowing for some small buffer though just in case.

", "votes": null, "replies": [] }, { "id": 468143, "author_name": "tarunpaparaju", "author_url": "", "post_date": "02/08/2019 11:06:32", "content": "

I tested my kernel on the new test data, and it runs in time (with 200s to spare). I'm so relieved. Phew !

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "466698": "Hello !\nHow much runtime buffer are you guys leaving for stage 2 evaluation.", "466717": "Why are people downvoting this ? Is there anything wrong in this question ?", "466721": "Highly depends on your preprocessing and number of times the predictions are made of the test data. \nLet's say now you are taking 2 mins for preprocessing test data and making the predictions 5 time (30 secs each)\n\nSo you should atleast save (2*60 + 5*30)*7 secs, since the 2nd Stage would be 7 times bigger. A save bet would be to have (2*60 + 5*30)*9 secs buffer to incorporate some other stuff, like if you are using tokenizer on test data as well as train data.\n\nHope this helps. Although it's a little late now.", "466794": "I mark this question sincere", "466797": "To answer your question we simulated the increased test size by repeating the test file 6 more times. That is probably the easiest way to get an estimate. We are still allowing for some small buffer though just in case.", "468143": "I tested my kernel on the new test data, and it runs in time (with 200s to spare). I'm so relieved. Phew !" }, "source": "meta" }