{ "id": 241143, "title": "How to submit model results", "url": "/competitions/birdclef-2021/discussion/241143", "author_name": "", "post_date": "2021-05-23T08:53:25.389569100Z", "votes": 1, "comment_count": 7, "views": 0, "content": "

Hi, all. I am new to kaggle. I wonder how can I submmit my results in this competition.
\nAs Rules says, I have to test my model on-line because the test datasets are hided, so that means I have to use my model and its corresponding weights to predict my results.

\n

However, when I try to use a random initialized model to predict results, e.g. efficient net. It informed that I have no No module named 'efficientnet_pytorch' when I commit notebook. However, It runs correctly in my notebook.

\n

By the way, is there any possible way I can train my model off-line and predict it on-line?

\n

I got confused about the submission mechanism. Anybody can give a baseline solution or example pipeline? Thanks.

", "messages": [ { "id": "1319495", "postDate": "05/23/2021 08:53:25", "content": "

Hi, all. I am new to kaggle. I wonder how can I submmit my results in this competition.
\nAs Rules says, I have to test my model on-line because the test datasets are hided, so that means I have to use my model and its corresponding weights to predict my results.

\n

However, when I try to use a random initialized model to predict results, e.g. efficient net. It informed that I have no No module named 'efficientnet_pytorch' when I commit notebook. However, It runs correctly in my notebook.

\n

By the way, is there any possible way I can train my model off-line and predict it on-line?

\n

I got confused about the submission mechanism. Anybody can give a baseline solution or example pipeline? Thanks.

", "rawMarkdown": "Hi, all. I am new to kaggle. I wonder how can I submmit my results in this competition.\nAs Rules says, I have to test my model on-line because the test datasets are hided, so that means I have to use my model and its corresponding weights to predict my results.\n\nHowever, when I try to use a random initialized model to predict results, e.g. efficient net. It informed that I have no No module named 'efficientnet_pytorch' when I commit notebook. However, It runs correctly in my notebook.\n\nBy the way, is there any possible way I can train my model off-line and predict it on-line?\n\nI got confused about the submission mechanism. Anybody can give a baseline solution or example pipeline? Thanks.", "votes": null }, { "id": "1319548", "postDate": "05/23/2021 10:00:44", "content": "

You need to save your model weights and whatever pytorch package you are using in one or more kaggle datasets and add these datasets to your notebook. See an example here: https://www.kaggle.com/kneroma/clean-fast-simple-bird-identifier-inference

\n

I believe (but I am not fully certain) that the corresponding training is https://www.kaggle.com/kneroma/clean-fast-simple-bird-identifier-training-colab

", "rawMarkdown": "You need to save your model weights and whatever pytorch package you are using in one or more kaggle datasets and add these datasets to your notebook. See an example here: https://www.kaggle.com/kneroma/clean-fast-simple-bird-identifier-inference\n\nI believe (but I am not fully certain) that the corresponding training is https://www.kaggle.com/kneroma/clean-fast-simple-bird-identifier-training-colab", "votes": null }, { "id": "1319749", "postDate": "05/23/2021 13:25:24", "content": "

Thanks for your sharing! I am clear now.

", "rawMarkdown": "Thanks for your sharing! I am clear now.", "votes": null }, { "id": "1319761", "postDate": "05/23/2021 13:34:22", "content": "

I got another problem. If I train the network on notebook kernel. When I commit this version, do I need to re-train this net for a long time or I can simply use the trained net weights by some operation?

", "rawMarkdown": "I got another problem. If I train the network on notebook kernel. When I commit this version, do I need to re-train this net for a long time or I can simply use the trained net weights by some operation?", "votes": null }, { "id": "1320213", "postDate": "05/23/2021 21:43:37", "content": "

Devil is in details, but you can always reuse pretrained wights or weights you traine dyourself.

", "rawMarkdown": "Devil is in details, but you can always reuse pretrained wights or weights you traine dyourself.", "votes": null }, { "id": "1320252", "postDate": "05/23/2021 23:59:03", "content": "

but how to make it？ when i create anather notebook i cannot found the model weights . If I do it in original notebook ， it has to train model again when submit

", "rawMarkdown": "but how to make it？ when i create anather notebook i cannot found the model weights . If I do it in original notebook ， it has to train model again when submit", "votes": null }, { "id": "1320266", "postDate": "05/24/2021 00:37:54", "content": "

You have to save your weights in a file and turn this into a dataset. From Kaggle documentation:

\n
\n

Notebook Output File Datasets

\n

Creating a dataset from a Notebook’s output files will let you create reproducible data pipelines. To create a dataset from a Notebook’s output files, click on the icon in the uploader and search for your Notebook. Alternatively, you can click “Create Dataset” from the Output tab on your rendered Notebook. Then, select the files you want to use in your dataset.

\n
", "rawMarkdown": "You have to save your weights in a file and turn this into a dataset. From [Kaggle documentation](https://www.kaggle.com/docs/datasets):\n\n> Notebook Output File Datasets\n> \n> Creating a dataset from a Notebook’s output files will let you create reproducible data pipelines. To create a dataset from a Notebook’s output files, click on the icon in the uploader and search for your Notebook. Alternatively, you can click “Create Dataset” from the Output tab on your rendered Notebook. Then, select the files you want to use in your dataset.", "votes": null }, { "id": "1320464", "postDate": "05/24/2021 05:43:19", "content": "

Thanks a lot! Your warm reply helps me a lot😄

", "rawMarkdown": "Thanks a lot! Your warm reply helps me a lot😄", "votes": null } ], "comments": [ { "id": 1319548, "author_name": "cpmpml", "author_url": "", "post_date": "05/23/2021 10:00:44", "content": "

You need to save your model weights and whatever pytorch package you are using in one or more kaggle datasets and add these datasets to your notebook. See an example here: https://www.kaggle.com/kneroma/clean-fast-simple-bird-identifier-inference

\n

I believe (but I am not fully certain) that the corresponding training is https://www.kaggle.com/kneroma/clean-fast-simple-bird-identifier-training-colab

", "votes": null, "replies": [ { "id": 1319749, "author_name": "haoweiz", "author_url": "", "post_date": "05/23/2021 13:25:24", "content": "

Thanks for your sharing! I am clear now.

", "votes": null, "replies": [] } ] }, { "id": 1319761, "author_name": "haoweiz", "author_url": "", "post_date": "05/23/2021 13:34:22", "content": "

I got another problem. If I train the network on notebook kernel. When I commit this version, do I need to re-train this net for a long time or I can simply use the trained net weights by some operation?

", "votes": null, "replies": [ { "id": 1320213, "author_name": "cpmpml", "author_url": "", "post_date": "05/23/2021 21:43:37", "content": "

Devil is in details, but you can always reuse pretrained wights or weights you traine dyourself.

", "votes": null, "replies": [] }, { "id": 1320252, "author_name": "haoweiz", "author_url": "", "post_date": "05/23/2021 23:59:03", "content": "

but how to make it？ when i create anather notebook i cannot found the model weights . If I do it in original notebook ， it has to train model again when submit

", "votes": null, "replies": [] }, { "id": 1320266, "author_name": "cpmpml", "author_url": "", "post_date": "05/24/2021 00:37:54", "content": "

You have to save your weights in a file and turn this into a dataset. From Kaggle documentation:

\n
\n

Notebook Output File Datasets

\n

Creating a dataset from a Notebook’s output files will let you create reproducible data pipelines. To create a dataset from a Notebook’s output files, click on the icon in the uploader and search for your Notebook. Alternatively, you can click “Create Dataset” from the Output tab on your rendered Notebook. Then, select the files you want to use in your dataset.

\n
", "votes": null, "replies": [] }, { "id": 1320464, "author_name": "haoweiz", "author_url": "", "post_date": "05/24/2021 05:43:19", "content": "

Thanks a lot! Your warm reply helps me a lot😄

", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1319495": "Hi, all. I am new to kaggle. I wonder how can I submmit my results in this competition.\nAs Rules says, I have to test my model on-line because the test datasets are hided, so that means I have to use my model and its corresponding weights to predict my results.\n\nHowever, when I try to use a random initialized model to predict results, e.g. efficient net. It informed that I have no No module named 'efficientnet_pytorch' when I commit notebook. However, It runs correctly in my notebook.\n\nBy the way, is there any possible way I can train my model off-line and predict it on-line?\n\nI got confused about the submission mechanism. Anybody can give a baseline solution or example pipeline? Thanks.", "1319548": "You need to save your model weights and whatever pytorch package you are using in one or more kaggle datasets and add these datasets to your notebook. See an example here: https://www.kaggle.com/kneroma/clean-fast-simple-bird-identifier-inference\n\nI believe (but I am not fully certain) that the corresponding training is https://www.kaggle.com/kneroma/clean-fast-simple-bird-identifier-training-colab", "1319749": "Thanks for your sharing! I am clear now.", "1319761": "I got another problem. If I train the network on notebook kernel. When I commit this version, do I need to re-train this net for a long time or I can simply use the trained net weights by some operation?", "1320213": "Devil is in details, but you can always reuse pretrained wights or weights you traine dyourself.", "1320252": "but how to make it？ when i create anather notebook i cannot found the model weights . If I do it in original notebook ， it has to train model again when submit", "1320266": "You have to save your weights in a file and turn this into a dataset. From [Kaggle documentation](https://www.kaggle.com/docs/datasets):\n\n> Notebook Output File Datasets\n> \n> Creating a dataset from a Notebook’s output files will let you create reproducible data pipelines. To create a dataset from a Notebook’s output files, click on the icon in the uploader and search for your Notebook. Alternatively, you can click “Create Dataset” from the Output tab on your rendered Notebook. Then, select the files you want to use in your dataset.", "1320464": "Thanks a lot! Your warm reply helps me a lot😄" }, "source": "meta" }