![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1975114%2F47498d5eca012e60cefd618898aad94f%2FHourglass%20Network.png?generation=1578900912541877&alt=media\")
First time to post here. I got a stupid question. Does the kernel for the final submission have to include the training process? or we can just load a pre-trained model that can make prediction. Thanks.
\n\nI made a figure showing the structure of the Hourglass network. I hope you find it is helpful.\n
First time to post here. I got a stupid question. Does the kernel for the final submission have to include the training process? or we can just load a pre-trained model that can make prediction. Thanks.
\n\nI made a figure showing the structure of the Hourglass network. I hope you find it is helpful.\n
I think kernel submission means that prediction.csv needs to be generated in the kernel, and we can upload and use pretrained model.
\n\nAnd I have another question about the description in Timeline: January 14, 2020 - Pre-trained model and external data disclosure deadline. Participants must disclose any external data or pre-trained models used in the official forum thread in adherence with competition rules. \nI do not understand the meaning of disclose any external data or pre-trained models used in the official forum thread, does the pre-trained models here include the model we trained outside.
maybe you don't have to. (first of all, this is not Code-Competition.)\njust a reminder, you can upload your csv-file results\n1. using kaggle api with kaggle competitions submit -c pku-autonomous-driving -f submission.csv -m \"Message\" command\nor\n2. click Submit Predictions
@welkinfeng \n\"pre-trained models\" means some model-weight which was trained with other dataset like ImageNet, i think.
", "rawMarkdown": "maybe you don't have to. (first of all, this is not `Code-Competition`.)\njust a reminder, you can upload your csv-file results\n1. using [kaggle api](https://github.com/Kaggle/kaggle-api) with `kaggle competitions submit -c pku-autonomous-driving -f submission.csv -m \"Message\"` command\nor\n2. click [Submit Predictions](https://www.kaggle.com/c/pku-autonomous-driving/submit)\n\n@welkinfeng \n\"pre-trained models\" means some model-weight which was trained with other dataset like ImageNet, i think.", "votes": null }, { "id": "717652", "postDate": "01/13/2020 12:10:23", "content": "Thanks for your reply.
\n\nSo... if we only upload csv file, they will not know if we use a pre-trained model...
", "rawMarkdown": "Thanks for your reply.\n\nSo... if we only upload csv file, they will not know if we use a pre-trained model...", "votes": null }, { "id": "717692", "postDate": "01/13/2020 13:37:17", "content": "Sorry for not making it clear. For the pre-trained model, actually I mean a model trained on the dataset of this competition. I know that we can upload csv file for submission as @ryomak mentioned.
\n\nSince training a model in kernel takes long time, I would rather train it locally. Then I upload this model to kernel, make predictions and submit. In this case, model is not trained in kernel but only does the test part. So I do not have to include the code of training process in kernel. Is this okay for the final submission?
", "rawMarkdown": "Sorry for not making it clear. For the pre-trained model, actually I mean a model trained on the dataset of this competition. I know that we can upload csv file for submission as @ryomak mentioned. \n\nSince training a model in kernel takes long time, I would rather train it locally. Then I upload this model to kernel, make predictions and submit. In this case, model is not trained in kernel but only does the test part. So I do not have to include the code of training process in kernel. Is this okay for the final submission?", "votes": null }, { "id": "717710", "postDate": "01/13/2020 14:04:11", "content": "All you have to do is submit a csv file directly, no need for any kernals. If you want to use a kernal for inference that's fine.
", "rawMarkdown": "All you have to do is submit a csv file directly, no need for any kernals. If you want to use a kernal for inference that's fine.", "votes": null } ], "comments": [ { "id": 717556, "author_name": "welkinfeng", "author_url": "", "post_date": "01/13/2020 09:30:07", "content": "I think kernel submission means that prediction.csv needs to be generated in the kernel, and we can upload and use pretrained model.
\n\nAnd I have another question about the description in Timeline: January 14, 2020 - Pre-trained model and external data disclosure deadline. Participants must disclose any external data or pre-trained models used in the official forum thread in adherence with competition rules. \nI do not understand the meaning of disclose any external data or pre-trained models used in the official forum thread, does the pre-trained models here include the model we trained outside.
maybe you don't have to. (first of all, this is not Code-Competition.)\njust a reminder, you can upload your csv-file results\n1. using kaggle api with kaggle competitions submit -c pku-autonomous-driving -f submission.csv -m \"Message\" command\nor\n2. click Submit Predictions
@welkinfeng \n\"pre-trained models\" means some model-weight which was trained with other dataset like ImageNet, i think.
", "votes": null, "replies": [ { "id": 717652, "author_name": "welkinfeng", "author_url": "", "post_date": "01/13/2020 12:10:23", "content": "Thanks for your reply.
\n\nSo... if we only upload csv file, they will not know if we use a pre-trained model...
", "votes": null, "replies": [] }, { "id": 717692, "author_name": "quanchengzhang", "author_url": "", "post_date": "01/13/2020 13:37:17", "content": "Sorry for not making it clear. For the pre-trained model, actually I mean a model trained on the dataset of this competition. I know that we can upload csv file for submission as @ryomak mentioned.
\n\nSince training a model in kernel takes long time, I would rather train it locally. Then I upload this model to kernel, make predictions and submit. In this case, model is not trained in kernel but only does the test part. So I do not have to include the code of training process in kernel. Is this okay for the final submission?
", "votes": null, "replies": [] }, { "id": 717710, "author_name": "greatgamedota", "author_url": "", "post_date": "01/13/2020 14:04:11", "content": "All you have to do is submit a csv file directly, no need for any kernals. If you want to use a kernal for inference that's fine.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "717494": "First time to post here. I got a stupid question. Does the kernel for the final submission have to include the training process? or we can just load a pre-trained model that can make prediction. Thanks. \n\nI made a figure showing the structure of the Hourglass network. I hope you find it is helpful.\n", "717556": "I think **kernel submission** means that prediction.csv needs to be generated in the kernel, and we can upload and use pretrained model.\n\nAnd I have another question about the description in `Timeline: January 14, 2020 - Pre-trained model and external data disclosure deadline. Participants must disclose any external data or pre-trained models used in the official forum thread in adherence with competition rules.` \nI do not understand the meaning of `disclose any external data or pre-trained models used in the official forum thread`, does the pre-trained models here include the model we trained outside.", "717626": "maybe you don't have to. (first of all, this is not `Code-Competition`.)\njust a reminder, you can upload your csv-file results\n1. using [kaggle api](https://github.com/Kaggle/kaggle-api) with `kaggle competitions submit -c pku-autonomous-driving -f submission.csv -m \"Message\"` command\nor\n2. click [Submit Predictions](https://www.kaggle.com/c/pku-autonomous-driving/submit)\n\n@welkinfeng \n\"pre-trained models\" means some model-weight which was trained with other dataset like ImageNet, i think.", "717652": "Thanks for your reply.\n\nSo... if we only upload csv file, they will not know if we use a pre-trained model...", "717692": "Sorry for not making it clear. For the pre-trained model, actually I mean a model trained on the dataset of this competition. I know that we can upload csv file for submission as @ryomak mentioned. \n\nSince training a model in kernel takes long time, I would rather train it locally. Then I upload this model to kernel, make predictions and submit. In this case, model is not trained in kernel but only does the test part. So I do not have to include the code of training process in kernel. Is this okay for the final submission?", "717710": "All you have to do is submit a csv file directly, no need for any kernals. If you want to use a kernal for inference that's fine." }, "source": "meta" }