\n
![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F1554871%2F57cda94195d506a1b150fdc4207d4c42%2FAnnotation%202020-09-06%20163927.png?generation=1599390736098749&alt=media\")
Hi, I was able to produce a submission without the use of the images. But since I included them, I can't produce any sucessful submission. I am able to predict with the small test set but it fails when I do the submission. In short, I am doing:
\n-Preparation of the tabular data
\n-DataGenerator for the images and the tabular
\n-Finally load a pre-trained model and do the prediction
Here is my notebook: https://www.kaggle.com/yohannwattiez/load-trained-model-and-predict
\nThank you in advance for the help :)
", "messages": [ { "id": "1000162", "postDate": "09/06/2020 10:54:40", "content": "Hi, I was able to produce a submission without the use of the images. But since I included them, I can't produce any sucessful submission. I am able to predict with the small test set but it fails when I do the submission. In short, I am doing:
\n-Preparation of the tabular data
\n-DataGenerator for the images and the tabular
\n-Finally load a pre-trained model and do the prediction
Here is my notebook: https://www.kaggle.com/yohannwattiez/load-trained-model-and-predict
\nThank you in advance for the help :)
", "rawMarkdown": "Hi, I was able to produce a submission without the use of the images. But since I included them, I can't produce any sucessful submission. I am able to predict with the small test set but it fails when I do the submission. In short, I am doing:\n\n-Preparation of the tabular data\n-DataGenerator for the images and the tabular\n-Finally load a pre-trained model and do the prediction\n\nHere is my notebook: https://www.kaggle.com/yohannwattiez/load-trained-model-and-predict\n\nThank you in advance for the help :)", "votes": null }, { "id": "1000187", "postDate": "09/06/2020 11:14:35", "content": "@yohannwattiez What are you loading using the following code in your inference notebook?
\n
It's a class I created to transform the tabular. The class is defined just above:
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F3321142%2F6b18ee15127ad2cead283b3d7ccec90a%2Fcomment.png?generation=1599391445877243&alt=media\")
\"Submission CSV not found\" is when your code encounters an error halfway through execution before the CSV file is generated. It could be due to GPU memory running out when using images or another simple reason could be is that you have accidentally provided a wrong path for submission or hard-coded something which should be dynamically processed.
", "rawMarkdown": "**\"Submission CSV not found\"** is when your code encounters an error halfway through execution before the CSV file is generated. It could be due to GPU memory running out when using images or another simple reason could be is that you have accidentally provided a wrong path for submission or hard-coded something which should be dynamically processed.", "votes": null }, { "id": "1000246", "postDate": "09/06/2020 12:09:29", "content": "Yes, I know but I can't find where is my mistake. I will try to reduce the batch size. It might be due to that.
", "rawMarkdown": "Yes, I know but I can't find where is my mistake. I will try to reduce the batch size. It might be due to that.", "votes": null }, { "id": "1001427", "postDate": "09/07/2020 10:21:32", "content": "So I used the following procedure to find where it failed: Debug Procedure. It was the batch size. Apparently, it could support the batch size on the small test set but not on the hiddent set
", "rawMarkdown": "So I used the following procedure to find where it failed: [Debug Procedure](https://www.kaggle.com/c/osic-pulmonary-fibrosis-progression/discussion/180173). It was the batch size. Apparently, it could support the batch size on the small test set but not on the hiddent set", "votes": null }, { "id": "1003207", "postDate": "09/08/2020 18:31:35", "content": "I can confirm, it was just the batch size. There might be more images for some patient in the hidden set than in the small given test set.
", "rawMarkdown": "I can confirm, it was just the batch size. There might be more images for some patient in the hidden set than in the small given test set.", "votes": null } ], "comments": [ { "id": 1000187, "author_name": "yovinyahathugoda", "author_url": "", "post_date": "09/06/2020 11:14:35", "content": "@yohannwattiez What are you loading using the following code in your inference notebook?
\n
It's a class I created to transform the tabular. The class is defined just above:
\n
\"Submission CSV not found\" is when your code encounters an error halfway through execution before the CSV file is generated. It could be due to GPU memory running out when using images or another simple reason could be is that you have accidentally provided a wrong path for submission or hard-coded something which should be dynamically processed.
", "votes": null, "replies": [] }, { "id": 1000246, "author_name": "yohannwattiez", "author_url": "", "post_date": "09/06/2020 12:09:29", "content": "Yes, I know but I can't find where is my mistake. I will try to reduce the batch size. It might be due to that.
", "votes": null, "replies": [] } ] }, { "id": 1001427, "author_name": "yohannwattiez", "author_url": "", "post_date": "09/07/2020 10:21:32", "content": "So I used the following procedure to find where it failed: Debug Procedure. It was the batch size. Apparently, it could support the batch size on the small test set but not on the hiddent set
", "votes": null, "replies": [] }, { "id": 1003207, "author_name": "yohannwattiez", "author_url": "", "post_date": "09/08/2020 18:31:35", "content": "I can confirm, it was just the batch size. There might be more images for some patient in the hidden set than in the small given test set.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1000162": "Hi, I was able to produce a submission without the use of the images. But since I included them, I can't produce any sucessful submission. I am able to predict with the small test set but it fails when I do the submission. In short, I am doing:\n\n-Preparation of the tabular data\n-DataGenerator for the images and the tabular\n-Finally load a pre-trained model and do the prediction\n\nHere is my notebook: https://www.kaggle.com/yohannwattiez/load-trained-model-and-predict\n\nThank you in advance for the help :)", "1000187": "yohannwattiez What are you loading using the following code in your inference notebook?\n", "1000198": "It's a class I created to transform the tabular. The class is defined just above: \n", "1000212": "**\"Submission CSV not found\"** is when your code encounters an error halfway through execution before the CSV file is generated. It could be due to GPU memory running out when using images or another simple reason could be is that you have accidentally provided a wrong path for submission or hard-coded something which should be dynamically processed.", "1000246": "Yes, I know but I can't find where is my mistake. I will try to reduce the batch size. It might be due to that.", "1001427": "So I used the following procedure to find where it failed: [Debug Procedure](https://www.kaggle.com/c/osic-pulmonary-fibrosis-progression/discussion/180173). It was the batch size. Apparently, it could support the batch size on the small test set but not on the hiddent set", "1003207": "I can confirm, it was just the batch size. There might be more images for some patient in the hidden set than in the small given test set." }, "source": "meta" }