If you are using kaggle kernels(like me) to train, I recommend to have your model train at what is night for the U.S. I guess there's less load on kaggle's servers and the training is significantly faster(2-3x for training in my scenario).\nAnother tip is to use an inference kernel like so many public kernels have done, no need to retrain your model and it's better to save models locally(or thru a kernel output).\nAlso, if you want to train models for a long time but want to check progress(which you cant do in committed kernels), you can just train in the draft session and just use some nifty html to download your model from there. The downside is that you will have to keep a tab of your machine open(for the draft session)\nEdit: Also, for testing, follow the advice here: https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/98879#latest-574594
", "messages": [ { "id": "575532", "postDate": "07/15/2019 15:00:39", "content": "If you are using kaggle kernels(like me) to train, I recommend to have your model train at what is night for the U.S. I guess there's less load on kaggle's servers and the training is significantly faster(2-3x for training in my scenario).\nAnother tip is to use an inference kernel like so many public kernels have done, no need to retrain your model and it's better to save models locally(or thru a kernel output).\nAlso, if you want to train models for a long time but want to check progress(which you cant do in committed kernels), you can just train in the draft session and just use some nifty html to download your model from there. The downside is that you will have to keep a tab of your machine open(for the draft session)\nEdit: Also, for testing, follow the advice here: https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/98879#latest-574594
", "rawMarkdown": "If you are using kaggle kernels(like me) to train, I recommend to have your model train at what is night for the U.S. I guess there's less load on kaggle's servers and the training is significantly faster(2-3x for training in my scenario).\nAnother tip is to use an inference kernel like so many public kernels have done, no need to retrain your model and it's better to save models locally(or thru a kernel output).\nAlso, if you want to train models for a long time but want to check progress(which you cant do in committed kernels), you can just train in the draft session and just use some nifty html to download your model from there. The downside is that you will have to keep a tab of your machine open(for the draft session)\nEdit: Also, for testing, follow the advice here: https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/98879#latest-574594", "votes": null }, { "id": "575536", "postDate": "07/15/2019 15:06:23", "content": "We can also use Google drive to save the trained models weights, afterwards download them from there and upload them to inference kernel.
", "rawMarkdown": "We can also use Google drive to save the trained models weights, afterwards download them from there and upload them to inference kernel.", "votes": null }, { "id": "576847", "postDate": "07/16/2019 05:10:50", "content": "very nice!
\n\nI found the following discussion as a way to download without commit. This way seems to work for files < 2MB\nhttps://www.kaggle.com/product-feedback/61487359813
\n\nIs there any other way?
", "rawMarkdown": "very nice!\n\nI found the following discussion as a way to download without commit. This way seems to work for files < 2MB\nhttps://www.kaggle.com/product-feedback/61487359813\n\nIs there any other way?", "votes": null }, { "id": "577982", "postDate": "07/17/2019 08:47:22", "content": "This works great to download files of any size https://www.kaggle.com/kashnitsky/how-to-download-submission-file-without-committing
", "rawMarkdown": "This works great to download files of any size https://www.kaggle.com/kashnitsky/how-to-download-submission-file-without-committing", "votes": null }, { "id": "579655", "postDate": "07/19/2019 04:20:33", "content": "Thank you very much! \nI used to calculate the pytorch model again, which was very helpful.
", "rawMarkdown": "Thank you very much! \nI used to calculate the pytorch model again, which was very helpful.", "votes": null }, { "id": "580901", "postDate": "07/21/2019 03:07:25", "content": "what i've been using is html tags in the markdown boxes so\n<a href=filetodownload>DOWNLOAD</a>
thank you!
", "rawMarkdown": "thank you!", "votes": null }, { "id": "1598502", "postDate": "11/28/2021 14:18:50", "content": "I tried all the given approaches, but none of them seems working. I get following output:
\n404 page not found
", "rawMarkdown": "I tried all the given approaches, but none of them seems working. I get following output:\n\n**404 page not found**", "votes": null } ], "comments": [ { "id": 1598502, "author_name": "ashishrose", "author_url": "", "post_date": "11/28/2021 14:18:50", "content": "I tried all the given approaches, but none of them seems working. I get following output:
\n404 page not found
", "votes": null, "replies": [] }, { "id": 575536, "author_name": "rishabhiitbhu", "author_url": "", "post_date": "07/15/2019 15:06:23", "content": "We can also use Google drive to save the trained models weights, afterwards download them from there and upload them to inference kernel.
", "votes": null, "replies": [] }, { "id": 576847, "author_name": "currypurin", "author_url": "", "post_date": "07/16/2019 05:10:50", "content": "very nice!
\n\nI found the following discussion as a way to download without commit. This way seems to work for files < 2MB\nhttps://www.kaggle.com/product-feedback/61487359813
\n\nIs there any other way?
", "votes": null, "replies": [ { "id": 577982, "author_name": "vassilybar", "author_url": "", "post_date": "07/17/2019 08:47:22", "content": "This works great to download files of any size https://www.kaggle.com/kashnitsky/how-to-download-submission-file-without-committing
", "votes": null, "replies": [] }, { "id": 579655, "author_name": "currypurin", "author_url": "", "post_date": "07/19/2019 04:20:33", "content": "Thank you very much! \nI used to calculate the pytorch model again, which was very helpful.
", "votes": null, "replies": [] }, { "id": 580901, "author_name": "sidhanthholalkere", "author_url": "", "post_date": "07/21/2019 03:07:25", "content": "what i've been using is html tags in the markdown boxes so\n<a href=filetodownload>DOWNLOAD</a>
thank you!
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "575532": "If you are using kaggle kernels(like me) to train, I recommend to have your model train at what is night for the U.S. I guess there's less load on kaggle's servers and the training is significantly faster(2-3x for training in my scenario).\nAnother tip is to use an inference kernel like so many public kernels have done, no need to retrain your model and it's better to save models locally(or thru a kernel output).\nAlso, if you want to train models for a long time but want to check progress(which you cant do in committed kernels), you can just train in the draft session and just use some nifty html to download your model from there. The downside is that you will have to keep a tab of your machine open(for the draft session)\nEdit: Also, for testing, follow the advice here: https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/98879#latest-574594", "575536": "We can also use Google drive to save the trained models weights, afterwards download them from there and upload them to inference kernel.", "576847": "very nice!\n\nI found the following discussion as a way to download without commit. This way seems to work for files < 2MB\nhttps://www.kaggle.com/product-feedback/61487359813\n\nIs there any other way?", "577982": "This works great to download files of any size https://www.kaggle.com/kashnitsky/how-to-download-submission-file-without-committing", "579655": "Thank you very much! \nI used to calculate the pytorch model again, which was very helpful.", "580901": "what i've been using is html tags in the markdown boxes so\n`DOWNLOAD`", "586293": "thank you!", "1598502": "I tried all the given approaches, but none of them seems working. I get following output:\n\n**404 page not found**" }, "source": "meta" }