{ "id": 242263, "title": "Code Requirements Understanding", "url": "/competitions/siim-covid19-detection/discussion/242263", "author_name": "", "post_date": "2021-05-28T08:20:51.368288700Z", "votes": 1, "comment_count": 12, "views": 0, "content": "

Can someone explain to me what is \"GPU Notebook <= 9 hours run-time\" means?

\n", "messages": [ { "id": "1326113", "postDate": "05/28/2021 08:20:51", "content": "

Can someone explain to me what is \"GPU Notebook <= 9 hours run-time\" means?

\n", "rawMarkdown": "Can someone explain to me what is \"GPU Notebook <= 9 hours run-time\" means? \n- GPU has multiple capabilities such as GTX1080, 2080, 3090. Can I use a maximum of 3090?\n- How about if my training time > 9 hours because some modern networks might take much time for training (>9 hours)?\n- I can use only the model in which training time > 9 hours? but inference < 9 hours? \nI'm really new with notebook submitting. Hope someone will clear to me. Thank you all", "votes": null }, { "id": "1326267", "postDate": "05/28/2021 11:10:23", "content": "

This is a code competition, which means that the inference for public and private test set has to be done using a kaggle notebook. The code requirements are talking about Kaggle notebooks, not notebooks on your local machine. \"GPU Notebook\" means using a kaggle notebook with GPU enabled. Kaggle only provides Tesla P100-PCIE-16GB. You can train your models for however long you want, but you must use less than 9 hours while doing inference in the Kaggle notebook.

", "rawMarkdown": "This is a code competition, which means that the inference for public and private test set has to be done using a kaggle notebook. The code requirements are talking about Kaggle notebooks, not notebooks on your local machine. \"GPU Notebook\" means using a kaggle notebook with GPU enabled. Kaggle only provides Tesla P100-PCIE-16GB. You can train your models for however long you want, but you must use less than 9 hours while doing inference in the Kaggle notebook.", "votes": null }, { "id": "1327460", "postDate": "05/29/2021 10:09:45", "content": "

@hoangduyloc

\n

cool profile picture, Cocomelon, all day… all night….with my son :)

\n

in code competition:
\n1- You can train your model/s anywhere you want by using any hardware.
\n2- Upload your model/s as Kaggle Dataset
\n3- Use Kaggle Notebook with your models to Submit your code by inferencing the private test set in less than 9 hours.

\n

Note: Kaggle Notebook will always stop after 9 hours ( for training or inference)

", "rawMarkdown": "hoangduyloc \n\ncool profile picture, Cocomelon, all day... all night....with my son :)\n\nin code competition:\n1- You can train your model/s anywhere you want by using any hardware. \n2- Upload your model/s as Kaggle Dataset\n3- Use Kaggle Notebook with your models to Submit your code by inferencing the private test set in less than 9 hours.\n\nNote: Kaggle Notebook will always stop after 9 hours ( for training or inference)", "votes": null }, { "id": "1328695", "postDate": "05/30/2021 13:01:08", "content": "

Thank you so much for the clear explanation. I really appreciate that.

", "rawMarkdown": "Thank you so much for the clear explanation. I really appreciate that.", "votes": null }, { "id": "1328701", "postDate": "05/30/2021 13:02:51", "content": "

Thank you so much \"DAD\", I now understand. Thank you.

", "rawMarkdown": "Thank you so much \"DAD\", I now understand. Thank you.", "votes": null }, { "id": "1328706", "postDate": "05/30/2021 13:07:13", "content": "

Can I ask you one more question. An example, I use multiple models for doing ensemble (ex: Efficientnet, Yolo…) So I need to do inference for every single model and ensemble them using only 1 Kaggle notebook while executing time < 9 hours, is it right?

", "rawMarkdown": "Can I ask you one more question. An example, I use multiple models for doing ensemble (ex: Efficientnet, Yolo...) So I need to do inference for every single model and ensemble them using only 1 Kaggle notebook while executing time < 9 hours, is it right?", "votes": null }, { "id": "1328909", "postDate": "05/30/2021 16:27:49", "content": "

You can do whatever you want and submit. if there is a problem you will get a submission error.

\n
Notebook Timeout: Your submission notebook exceeded the allowed runtime. Review the competition's Code Requirements page for the time limits. Note that the hidden dataset can be larger/smaller/different than the public dataset.\n\nNotebook Exceeded Allowed Compute: This indicates you have violated a code requirement constraint during the rerun. This includes limitations in the execution environment, for example requesting more RAM or disk than available, or competition constraints, such as input data source type or size limits\n
\n

.

\n

I suggest reading this
\nhttps://www.kaggle.com/code-competition-debugging

", "rawMarkdown": "You can do whatever you want and submit. if there is a problem you will get a submission error.\n\n```\nNotebook Timeout: Your submission notebook exceeded the allowed runtime. Review the competition's Code Requirements page for the time limits. Note that the hidden dataset can be larger/smaller/different than the public dataset.\n\nNotebook Exceeded Allowed Compute: This indicates you have violated a code requirement constraint during the rerun. This includes limitations in the execution environment, for example requesting more RAM or disk than available, or competition constraints, such as input data source type or size limits\n```.\n\nI suggest reading this \nhttps://www.kaggle.com/code-competition-debugging", "votes": null }, { "id": "1329184", "postDate": "05/31/2021 00:09:19", "content": "

Thank you so much

", "rawMarkdown": "Thank you so much", "votes": null }, { "id": "1329549", "postDate": "05/31/2021 07:57:11", "content": "

I read it, but there is one thing I can not understand. Am I right if: I have 3 models A, B, C, and I trained and output out_1.csv, out_2.csv, out_3.csv in my local machines. Can I just load \"out_1.csv, out_2.csv, out_3.csv\" to Kaggle notebook to ensemble them (very fast, I can add more models) or I need to load \"pre-trained model A, B, C and do inference them then ensemble them using 1 Kaggle notebook (lower, limitation in time 9 hours)? thankyou.

", "rawMarkdown": "I read it, but there is one thing I can not understand. Am I right if: I have 3 models A, B, C, and I trained and output out_1.csv, out_2.csv, out_3.csv in my local machines. Can I just load \"out_1.csv, out_2.csv, out_3.csv\" to Kaggle notebook to ensemble them (very fast, I can add more models) or I need to load \"pre-trained model A, B, C and do inference them then ensemble them using 1 Kaggle notebook (lower, limitation in time 9 hours)? thankyou.", "votes": null }, { "id": "1329624", "postDate": "05/31/2021 08:58:43", "content": "

you are asking a very important question

\n

See the leaderboards here, you will find out more than 500 teams got 0.00 on the final leaderboard
\nWhy?
\nhttps://www.kaggle.com/c/hpa-single-cell-image-classification/leaderboard
\nhttps://www.kaggle.com/c/hubmap-kidney-segmentation/leaderboard

\n

in code competition, there is a private test set that we don't have access to. Sometimes by mistake team forgot to make the code run on a private test and they end up 0.00 on the final leaderboard.

", "rawMarkdown": "you are asking a very important question\n\nSee the leaderboards here, you will find out more than 500 teams got 0.00 on the final leaderboard \nWhy?\nhttps://www.kaggle.com/c/hpa-single-cell-image-classification/leaderboard\nhttps://www.kaggle.com/c/hubmap-kidney-segmentation/leaderboard\n\nin code competition, there is a private test set that we don't have access to. Sometimes by mistake team forgot to make the code run on a private test and they end up 0.00 on the final leaderboard.", "votes": null }, { "id": "1330539", "postDate": "06/01/2021 00:08:08", "content": "

Thank you, I really appreciate it!

", "rawMarkdown": "Thank you, I really appreciate it!", "votes": null }, { "id": "1385229", "postDate": "07/12/2021 14:31:02", "content": "

so we can train model offline and load the model's weight into notebook as pre-trained model and only performed inference with test set?

", "rawMarkdown": "so we can train model offline and load the model's weight into notebook as pre-trained model and only performed inference with test set?", "votes": null }, { "id": "1385720", "postDate": "07/13/2021 01:29:48", "content": "

Yes. perform inference with public test set and private test set.

", "rawMarkdown": "Yes. perform inference with public test set and private test set.", "votes": null } ], "comments": [ { "id": 1326267, "author_name": "novice03", "author_url": "", "post_date": "05/28/2021 11:10:23", "content": "

This is a code competition, which means that the inference for public and private test set has to be done using a kaggle notebook. The code requirements are talking about Kaggle notebooks, not notebooks on your local machine. \"GPU Notebook\" means using a kaggle notebook with GPU enabled. Kaggle only provides Tesla P100-PCIE-16GB. You can train your models for however long you want, but you must use less than 9 hours while doing inference in the Kaggle notebook.

", "votes": null, "replies": [ { "id": 1328695, "author_name": "hoangduyloc", "author_url": "", "post_date": "05/30/2021 13:01:08", "content": "

Thank you so much for the clear explanation. I really appreciate that.

", "votes": null, "replies": [] }, { "id": 1328706, "author_name": "hoangduyloc", "author_url": "", "post_date": "05/30/2021 13:07:13", "content": "

Can I ask you one more question. An example, I use multiple models for doing ensemble (ex: Efficientnet, Yolo…) So I need to do inference for every single model and ensemble them using only 1 Kaggle notebook while executing time < 9 hours, is it right?

", "votes": null, "replies": [] }, { "id": 1328909, "author_name": "faisalalsrheed", "author_url": "", "post_date": "05/30/2021 16:27:49", "content": "

You can do whatever you want and submit. if there is a problem you will get a submission error.

\n
Notebook Timeout: Your submission notebook exceeded the allowed runtime. Review the competition's Code Requirements page for the time limits. Note that the hidden dataset can be larger/smaller/different than the public dataset.\n\nNotebook Exceeded Allowed Compute: This indicates you have violated a code requirement constraint during the rerun. This includes limitations in the execution environment, for example requesting more RAM or disk than available, or competition constraints, such as input data source type or size limits\n
\n

.

\n

I suggest reading this
\nhttps://www.kaggle.com/code-competition-debugging

", "votes": null, "replies": [] }, { "id": 1329184, "author_name": "hoangduyloc", "author_url": "", "post_date": "05/31/2021 00:09:19", "content": "

Thank you so much

", "votes": null, "replies": [] }, { "id": 1329549, "author_name": "hoangduyloc", "author_url": "", "post_date": "05/31/2021 07:57:11", "content": "

I read it, but there is one thing I can not understand. Am I right if: I have 3 models A, B, C, and I trained and output out_1.csv, out_2.csv, out_3.csv in my local machines. Can I just load \"out_1.csv, out_2.csv, out_3.csv\" to Kaggle notebook to ensemble them (very fast, I can add more models) or I need to load \"pre-trained model A, B, C and do inference them then ensemble them using 1 Kaggle notebook (lower, limitation in time 9 hours)? thankyou.

", "votes": null, "replies": [] }, { "id": 1329624, "author_name": "faisalalsrheed", "author_url": "", "post_date": "05/31/2021 08:58:43", "content": "

you are asking a very important question

\n

See the leaderboards here, you will find out more than 500 teams got 0.00 on the final leaderboard
\nWhy?
\nhttps://www.kaggle.com/c/hpa-single-cell-image-classification/leaderboard
\nhttps://www.kaggle.com/c/hubmap-kidney-segmentation/leaderboard

\n

in code competition, there is a private test set that we don't have access to. Sometimes by mistake team forgot to make the code run on a private test and they end up 0.00 on the final leaderboard.

", "votes": null, "replies": [] }, { "id": 1330539, "author_name": "hoangduyloc", "author_url": "", "post_date": "06/01/2021 00:08:08", "content": "

Thank you, I really appreciate it!

", "votes": null, "replies": [] }, { "id": 1385229, "author_name": "hawkeat", "author_url": "", "post_date": "07/12/2021 14:31:02", "content": "

so we can train model offline and load the model's weight into notebook as pre-trained model and only performed inference with test set?

", "votes": null, "replies": [] }, { "id": 1385720, "author_name": "faisalalsrheed", "author_url": "", "post_date": "07/13/2021 01:29:48", "content": "

Yes. perform inference with public test set and private test set.

", "votes": null, "replies": [] } ] }, { "id": 1327460, "author_name": "faisalalsrheed", "author_url": "", "post_date": "05/29/2021 10:09:45", "content": "

@hoangduyloc

\n

cool profile picture, Cocomelon, all day… all night….with my son :)

\n

in code competition:
\n1- You can train your model/s anywhere you want by using any hardware.
\n2- Upload your model/s as Kaggle Dataset
\n3- Use Kaggle Notebook with your models to Submit your code by inferencing the private test set in less than 9 hours.

\n

Note: Kaggle Notebook will always stop after 9 hours ( for training or inference)

", "votes": null, "replies": [ { "id": 1328701, "author_name": "hoangduyloc", "author_url": "", "post_date": "05/30/2021 13:02:51", "content": "

Thank you so much \"DAD\", I now understand. Thank you.

", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1326113": "Can someone explain to me what is \"GPU Notebook <= 9 hours run-time\" means? \n- GPU has multiple capabilities such as GTX1080, 2080, 3090. Can I use a maximum of 3090?\n- How about if my training time > 9 hours because some modern networks might take much time for training (>9 hours)?\n- I can use only the model in which training time > 9 hours? but inference < 9 hours? \nI'm really new with notebook submitting. Hope someone will clear to me. Thank you all", "1326267": "This is a code competition, which means that the inference for public and private test set has to be done using a kaggle notebook. The code requirements are talking about Kaggle notebooks, not notebooks on your local machine. \"GPU Notebook\" means using a kaggle notebook with GPU enabled. Kaggle only provides Tesla P100-PCIE-16GB. You can train your models for however long you want, but you must use less than 9 hours while doing inference in the Kaggle notebook.", "1327460": "hoangduyloc \n\ncool profile picture, Cocomelon, all day... all night....with my son :)\n\nin code competition:\n1- You can train your model/s anywhere you want by using any hardware. \n2- Upload your model/s as Kaggle Dataset\n3- Use Kaggle Notebook with your models to Submit your code by inferencing the private test set in less than 9 hours.\n\nNote: Kaggle Notebook will always stop after 9 hours ( for training or inference)", "1328695": "Thank you so much for the clear explanation. I really appreciate that.", "1328701": "Thank you so much \"DAD\", I now understand. Thank you.", "1328706": "Can I ask you one more question. An example, I use multiple models for doing ensemble (ex: Efficientnet, Yolo...) So I need to do inference for every single model and ensemble them using only 1 Kaggle notebook while executing time < 9 hours, is it right?", "1328909": "You can do whatever you want and submit. if there is a problem you will get a submission error.\n\n```\nNotebook Timeout: Your submission notebook exceeded the allowed runtime. Review the competition's Code Requirements page for the time limits. Note that the hidden dataset can be larger/smaller/different than the public dataset.\n\nNotebook Exceeded Allowed Compute: This indicates you have violated a code requirement constraint during the rerun. This includes limitations in the execution environment, for example requesting more RAM or disk than available, or competition constraints, such as input data source type or size limits\n```.\n\nI suggest reading this \nhttps://www.kaggle.com/code-competition-debugging", "1329184": "Thank you so much", "1329549": "I read it, but there is one thing I can not understand. Am I right if: I have 3 models A, B, C, and I trained and output out_1.csv, out_2.csv, out_3.csv in my local machines. Can I just load \"out_1.csv, out_2.csv, out_3.csv\" to Kaggle notebook to ensemble them (very fast, I can add more models) or I need to load \"pre-trained model A, B, C and do inference them then ensemble them using 1 Kaggle notebook (lower, limitation in time 9 hours)? thankyou.", "1329624": "you are asking a very important question\n\nSee the leaderboards here, you will find out more than 500 teams got 0.00 on the final leaderboard \nWhy?\nhttps://www.kaggle.com/c/hpa-single-cell-image-classification/leaderboard\nhttps://www.kaggle.com/c/hubmap-kidney-segmentation/leaderboard\n\nin code competition, there is a private test set that we don't have access to. Sometimes by mistake team forgot to make the code run on a private test and they end up 0.00 on the final leaderboard.", "1330539": "Thank you, I really appreciate it!", "1385229": "so we can train model offline and load the model's weight into notebook as pre-trained model and only performed inference with test set?", "1385720": "Yes. perform inference with public test set and private test set." }, "source": "meta" }