@admin\nAs a recent Kaggler, I don't really understand the limit to the Python packages we can/can't have on Notebooks. And why this is special to notebooks with GPU usage... Why couldn't we all use the famous MTCNN (for efficient face detection) in a GPU Notebook? It would contribute to all of us.\nBut again, I probably miss something important here (I'm not a big fan of notebooks anyway...) and I'm ready to learn (like my model).\nSimon
", "messages": [ { "id": "694614", "postDate": "12/13/2019 21:00:39", "content": "@admin\nAs a recent Kaggler, I don't really understand the limit to the Python packages we can/can't have on Notebooks. And why this is special to notebooks with GPU usage... Why couldn't we all use the famous MTCNN (for efficient face detection) in a GPU Notebook? It would contribute to all of us.\nBut again, I probably miss something important here (I'm not a big fan of notebooks anyway...) and I'm ready to learn (like my model).\nSimon
", "rawMarkdown": "admin\nAs a recent Kaggler, I don't really understand the limit to the Python packages we can/can't have on Notebooks. And why this is special to notebooks with GPU usage... Why couldn't we all use the famous MTCNN (for efficient face detection) in a GPU Notebook? It would contribute to all of us.\nBut again, I probably miss something important here (I'm not a big fan of notebooks anyway...) and I'm ready to learn (like my model).\nSimon", "votes": null }, { "id": "694698", "postDate": "12/14/2019 01:33:43", "content": "I don't see why we can't use the mtcnn package. I uploaded a dataset containing the wheel. click me\nYou add this dataset to your kernel and add this line to the top of your kernel:\n!pip install ../input/mtcnn-package/mtcnn-0.1.0-py3-none-any.whl
Then you can import it as normal.\nhope this helps!
", "rawMarkdown": "I don't see why we can't use the mtcnn package. I uploaded a dataset containing the wheel. [click me](https://www.kaggle.com/unkownhihi/mtcnn-package)\nYou add this dataset to your kernel and add this line to the top of your kernel:\n`!pip install ../input/mtcnn-package/mtcnn-0.1.0-py3-none-any.whl`\n\nThen you can import it as normal.\nhope this helps!", "votes": null }, { "id": "694975", "postDate": "12/14/2019 12:16:15", "content": "Thanks. It'll help me upload my current local model. Let's just hope that it'll be a \"legal\" solution.
", "rawMarkdown": "Thanks. It'll help me upload my current local model. Let's just hope that it'll be a \"legal\" solution.", "votes": null }, { "id": "695236", "postDate": "12/14/2019 20:21:34", "content": "According to \"Code Requirements\" :
\n\n\n\n\nExternal data is allowed up to 1 GB in size. External data must be freely & publicly available, including pre-trained models.\n No using other Kaggle notebooks or utility scripts as inputs to your submission notebook. External data also cannot be externally-referenced (i.e. via BigQuery, Github, external URL). Instead, load external models or datasets directly as an external data source.
\n
So I guess we just need to download it and upload it as normal data source would fit in the rules.
", "rawMarkdown": "According to \"Code Requirements\" : \n> External data is allowed up to 1 GB in size. External data must be freely & publicly available, including pre-trained models.\nNo using other Kaggle notebooks or utility scripts as inputs to your submission notebook. External data also cannot be externally-referenced (i.e. via BigQuery, Github, external URL). Instead, load external models or datasets directly as an external data source.\n\nSo I guess we just need to download it and upload it as normal data source would fit in the rules.", "votes": null }, { "id": "699108", "postDate": "12/20/2019 03:53:28", "content": "It's always impressive how fast Kagglers can get over hinderances. I thought no internet would stop the use of lots of packages - but serveral kernels show how you can include the code needed as part of your 1GB data. Only issue would be if you need to use more space for your model weights - but I agree with Simon - to some extent it is very STUPID to not include some of the current and effective libraries - sometimes I think Kaggle smarter than I give them credit for - but this is not one of the times.
\n\nSimon - did you make the request thru the support site ?? I have seen no evidence in the last year that Kaggle accepts discussion posts as official requests.
", "rawMarkdown": "It's always impressive how fast Kagglers can get over hinderances. I thought no internet would stop the use of lots of packages - but serveral kernels show how you can include the code needed as part of your 1GB data. Only issue would be if you need to use more space for your model weights - but I agree with Simon - to some extent it is very STUPID to not include some of the current and effective libraries - sometimes I think Kaggle smarter than I give them credit for - but this is not one of the times.\n\nSimon - did you make the request thru the support site ?? I have seen no evidence in the last year that Kaggle accepts discussion posts as official requests.", "votes": null }, { "id": "707391", "postDate": "12/31/2019 18:24:15", "content": "Doesn't this violate the rules?
\n\n\n\n\nNo custom packages enabled in your submission notebook
\n
Could the sponsors or Kaggle clarify the legality of using third-party software, such as this, in the submission notebook?
", "rawMarkdown": "Doesn't this violate the rules?\n\n> No custom packages enabled in your submission notebook\n\nCould the sponsors or Kaggle clarify the legality of using third-party software, such as this, in the submission notebook?", "votes": null }, { "id": "708793", "postDate": "01/02/2020 17:32:01", "content": "The “no custom packages” requirement is specific to not enabling “custom packages” through the settings menu on the notebook/script. But if you find alternative methods to use packages by loading them as external data sources and without needing internet enabled, then that’s permitted.
", "rawMarkdown": "The “no custom packages” requirement is specific to not enabling “custom packages” through the settings menu on the notebook/script. But if you find alternative methods to use packages by loading them as external data sources and without needing internet enabled, then that’s permitted.", "votes": null }, { "id": "708895", "postDate": "01/02/2020 20:10:58", "content": "@juliaelliott Thanks for the clarification!
", "rawMarkdown": "juliaelliott Thanks for the clarification!", "votes": null }, { "id": "723185", "postDate": "01/19/2020 15:50:57", "content": "I add !pip install /kaggle/input/mtcnnpackage/mtcnn-0.1.0-py3-none-any.whl to the top of my kernel and receive an 'invalid syntax' error. I can't seem to figure this out. Is something else needed to run pip inside the kernel? My kernel type is 'script' and not 'notebook', this appears to be the only difference between our examples.
The kernel type was the issue. You can only use the !pip install method inside a notebook.
I don't see why we can't use the mtcnn package. I uploaded a dataset containing the wheel. click me\nYou add this dataset to your kernel and add this line to the top of your kernel:\n!pip install ../input/mtcnn-package/mtcnn-0.1.0-py3-none-any.whl
Then you can import it as normal.\nhope this helps!
", "votes": null, "replies": [ { "id": 694975, "author_name": "simoninparis", "author_url": "", "post_date": "12/14/2019 12:16:15", "content": "Thanks. It'll help me upload my current local model. Let's just hope that it'll be a \"legal\" solution.
", "votes": null, "replies": [] }, { "id": 695236, "author_name": "unkownhihi", "author_url": "", "post_date": "12/14/2019 20:21:34", "content": "According to \"Code Requirements\" :
\n\n\n\n\nExternal data is allowed up to 1 GB in size. External data must be freely & publicly available, including pre-trained models.\n No using other Kaggle notebooks or utility scripts as inputs to your submission notebook. External data also cannot be externally-referenced (i.e. via BigQuery, Github, external URL). Instead, load external models or datasets directly as an external data source.
\n
So I guess we just need to download it and upload it as normal data source would fit in the rules.
", "votes": null, "replies": [] }, { "id": 707391, "author_name": "olegtrott", "author_url": "", "post_date": "12/31/2019 18:24:15", "content": "Doesn't this violate the rules?
\n\n\n\n\nNo custom packages enabled in your submission notebook
\n
Could the sponsors or Kaggle clarify the legality of using third-party software, such as this, in the submission notebook?
", "votes": null, "replies": [] }, { "id": 708793, "author_name": "juliaelliott", "author_url": "", "post_date": "01/02/2020 17:32:01", "content": "The “no custom packages” requirement is specific to not enabling “custom packages” through the settings menu on the notebook/script. But if you find alternative methods to use packages by loading them as external data sources and without needing internet enabled, then that’s permitted.
", "votes": null, "replies": [] }, { "id": 708895, "author_name": "olegtrott", "author_url": "", "post_date": "01/02/2020 20:10:58", "content": "@juliaelliott Thanks for the clarification!
", "votes": null, "replies": [] } ] }, { "id": 699108, "author_name": "pcjimmmy", "author_url": "", "post_date": "12/20/2019 03:53:28", "content": "It's always impressive how fast Kagglers can get over hinderances. I thought no internet would stop the use of lots of packages - but serveral kernels show how you can include the code needed as part of your 1GB data. Only issue would be if you need to use more space for your model weights - but I agree with Simon - to some extent it is very STUPID to not include some of the current and effective libraries - sometimes I think Kaggle smarter than I give them credit for - but this is not one of the times.
\n\nSimon - did you make the request thru the support site ?? I have seen no evidence in the last year that Kaggle accepts discussion posts as official requests.
", "votes": null, "replies": [] }, { "id": 723185, "author_name": "dalekube", "author_url": "", "post_date": "01/19/2020 15:50:57", "content": "I add !pip install /kaggle/input/mtcnnpackage/mtcnn-0.1.0-py3-none-any.whl to the top of my kernel and receive an 'invalid syntax' error. I can't seem to figure this out. Is something else needed to run pip inside the kernel? My kernel type is 'script' and not 'notebook', this appears to be the only difference between our examples.
The kernel type was the issue. You can only use the !pip install method inside a notebook.