{ "id": 122286, "title": "What is the purpose of \"test_videos\"?", "url": "/competitions/deepfake-detection-challenge/discussion/122286", "author_name": "", "post_date": "2019-12-19T04:03:32.179574300Z", "votes": 4, "comment_count": 9, "views": 0, "content": "

I don't understand what is the purpose of \"test_videos\" if I can't see my LogLoss on them, I can't find any metadata.json for it.

\n\n

I want to see my LogLoss on test_videos before using one of my 2 submissions.

\n\n

I made my own test_videos out of training videos but my LogLoss is way more different.

", "messages": [ { "id": "698313", "postDate": "12/19/2019 04:03:32", "content": "

I don't understand what is the purpose of \"test_videos\" if I can't see my LogLoss on them, I can't find any metadata.json for it.

\n\n

I want to see my LogLoss on test_videos before using one of my 2 submissions.

\n\n

I made my own test_videos out of training videos but my LogLoss is way more different.

", "rawMarkdown": "I don't understand what is the purpose of \"test_videos\" if I can't see my LogLoss on them, I can't find any metadata.json for it.\n\nI want to see my LogLoss on test_videos before using one of my 2 submissions.\n\nI made my own test_videos out of training videos but my LogLoss is way more different.", "votes": null }, { "id": "698445", "postDate": "12/19/2019 08:45:19", "content": "
\n

but my LogLoss is way more different.

\n
\n\n

Welcome to Kaggle

", "rawMarkdown": "> but my LogLoss is way more different.\n\nWelcome to Kaggle", "votes": null }, { "id": "698461", "postDate": "12/19/2019 09:17:32", "content": "

", "rawMarkdown": "", "votes": null }, { "id": "698495", "postDate": "12/19/2019 10:12:25", "content": "

In that case they're not off to a bad start:

\n\n

\"\"

", "rawMarkdown": "In that case they're not off to a bad start:\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F575831%2F4fba62badef4f80b9ba03b5201210664%2FScreen%20Shot%202019-12-19%20at%2011.11.35.png?generation=1576750329265503&alt=media)", "votes": null }, { "id": "698496", "postDate": "12/19/2019 10:13:32", "content": "

Note that the mp4 files from test_videos are also part of the full training set, so you can get the labels from there. (It also means that if you're training on the full set, you're also training on the test videos and so any score you compute on those test videos will be too optimistic.)

", "rawMarkdown": "Note that the mp4 files from `test_videos` are also part of the full training set, so you can get the labels from there. (It also means that if you're training on the full set, you're also training on the test videos and so any score you compute on those test videos will be too optimistic.)", "votes": null }, { "id": "699216", "postDate": "12/20/2019 07:17:37", "content": "

I'm also confused. May be it's just designed for debugging......

", "rawMarkdown": "I'm also confused. May be it's just designed for debugging......", "votes": null }, { "id": "701724", "postDate": "12/23/2019 20:51:41", "content": "

test_videos in your interactive session is just a placeholder when committing your code. It is swapped out for the public test set's test_videos when making a submission through the \"Submit to Competition\" button. So it doesn't serve a scoring purpose during commit. But it does serve a purpose as a placeholder for the set of videos that we run your code against in the background during the private re-run of your code to return the public leaderboard score.

", "rawMarkdown": "`test_videos` in your interactive session is just a placeholder when committing your code. It is swapped out for the public test set's `test_videos` when making a submission through the \"Submit to Competition\" button. So it doesn't serve a scoring purpose during commit. But it does serve a purpose as a placeholder for the set of videos that we run your code against in the background during the private re-run of your code to return the public leaderboard score.", "votes": null }, { "id": "725116", "postDate": "01/21/2020 20:29:51", "content": "

deleted [how can I delete this comment?]

", "rawMarkdown": "deleted [how can I delete this comment?]", "votes": null }, { "id": "725743", "postDate": "01/22/2020 13:12:04", "content": "

Do you know how many real and fake videos are these 400 videos?

", "rawMarkdown": "Do you know how many real and fake videos are these 400 videos?", "votes": null }, { "id": "725932", "postDate": "01/22/2020 16:20:07", "content": "

50% fake and 50% real. Like I said, their labels are in the metadata for the full training set. The videos in test_videos are all taken from subdirs 0, 1, 2.

", "rawMarkdown": "50% fake and 50% real. Like I said, their labels are in the metadata for the full training set. The videos in `test_videos` are all taken from subdirs 0, 1, 2.", "votes": null } ], "comments": [ { "id": 698445, "author_name": "bibek777", "author_url": "", "post_date": "12/19/2019 08:45:19", "content": "
\n

but my LogLoss is way more different.

\n
\n\n

Welcome to Kaggle

", "votes": null, "replies": [ { "id": 698461, "author_name": "init27", "author_url": "", "post_date": "12/19/2019 09:17:32", "content": "

", "votes": null, "replies": [] }, { "id": 698495, "author_name": "humananalog", "author_url": "", "post_date": "12/19/2019 10:12:25", "content": "

In that case they're not off to a bad start:

\n\n

\"\"

", "votes": null, "replies": [] } ] }, { "id": 698496, "author_name": "humananalog", "author_url": "", "post_date": "12/19/2019 10:13:32", "content": "

Note that the mp4 files from test_videos are also part of the full training set, so you can get the labels from there. (It also means that if you're training on the full set, you're also training on the test videos and so any score you compute on those test videos will be too optimistic.)

", "votes": null, "replies": [ { "id": 725743, "author_name": "feifeizaici", "author_url": "", "post_date": "01/22/2020 13:12:04", "content": "

Do you know how many real and fake videos are these 400 videos?

", "votes": null, "replies": [] }, { "id": 725932, "author_name": "humananalog", "author_url": "", "post_date": "01/22/2020 16:20:07", "content": "

50% fake and 50% real. Like I said, their labels are in the metadata for the full training set. The videos in test_videos are all taken from subdirs 0, 1, 2.

", "votes": null, "replies": [] } ] }, { "id": 699216, "author_name": "fionalxd", "author_url": "", "post_date": "12/20/2019 07:17:37", "content": "

I'm also confused. May be it's just designed for debugging......

", "votes": null, "replies": [] }, { "id": 701724, "author_name": "juliaelliott", "author_url": "", "post_date": "12/23/2019 20:51:41", "content": "

test_videos in your interactive session is just a placeholder when committing your code. It is swapped out for the public test set's test_videos when making a submission through the \"Submit to Competition\" button. So it doesn't serve a scoring purpose during commit. But it does serve a purpose as a placeholder for the set of videos that we run your code against in the background during the private re-run of your code to return the public leaderboard score.

", "votes": null, "replies": [ { "id": 725116, "author_name": "aknirala", "author_url": "", "post_date": "01/21/2020 20:29:51", "content": "

deleted [how can I delete this comment?]

", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "698313": "I don't understand what is the purpose of \"test_videos\" if I can't see my LogLoss on them, I can't find any metadata.json for it.\n\nI want to see my LogLoss on test_videos before using one of my 2 submissions.\n\nI made my own test_videos out of training videos but my LogLoss is way more different.", "698445": "> but my LogLoss is way more different.\n\nWelcome to Kaggle", "698461": "", "698495": "In that case they're not off to a bad start:\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F575831%2F4fba62badef4f80b9ba03b5201210664%2FScreen%20Shot%202019-12-19%20at%2011.11.35.png?generation=1576750329265503&alt=media)", "698496": "Note that the mp4 files from `test_videos` are also part of the full training set, so you can get the labels from there. (It also means that if you're training on the full set, you're also training on the test videos and so any score you compute on those test videos will be too optimistic.)", "699216": "I'm also confused. May be it's just designed for debugging......", "701724": "`test_videos` in your interactive session is just a placeholder when committing your code. It is swapped out for the public test set's `test_videos` when making a submission through the \"Submit to Competition\" button. So it doesn't serve a scoring purpose during commit. But it does serve a purpose as a placeholder for the set of videos that we run your code against in the background during the private re-run of your code to return the public leaderboard score.", "725116": "deleted [how can I delete this comment?]", "725743": "Do you know how many real and fake videos are these 400 videos?", "725932": "50% fake and 50% real. Like I said, their labels are in the metadata for the full training set. The videos in `test_videos` are all taken from subdirs 0, 1, 2." }, "source": "meta" }