Hi everyone,
\nI made 5 different submissions with 5 different outputs but my score is still equals to -13.1045. Someone has any idea on why it does this ?
\nThank you
Hi everyone,
\nI made 5 different submissions with 5 different outputs but my score is still equals to -13.1045. Someone has any idea on why it does this ?
\nThank you
Hi @berniechou your fixed it?
", "rawMarkdown": "Hi @berniechou your fixed it?", "votes": null }, { "id": "1002717", "postDate": "09/08/2020 11:26:21", "content": "yes i was using an external dataset (the same images but as png and not dcm) to make my submission.csv, we have to use the test dataset provided by kaggle.
", "rawMarkdown": "yes i was using an external dataset (the same images but as png and not dcm) to make my submission.csv, we have to use the test dataset provided by kaggle.", "votes": null }, { "id": "1021519", "postDate": "09/21/2020 23:58:21", "content": "Hello @berniechou . Im having the same problem and always scoring -13.1045. On inference I am using the oficial test dataset from imput to make some png images, which I save on the output directory, then transform the images, then load the png and make predictions. Is this the same as your case that you say is not allowed?
", "rawMarkdown": "Hello @berniechou . Im having the same problem and always scoring -13.1045. On inference I am using the oficial test dataset from imput to make some png images, which I save on the output directory, then transform the images, then load the png and make predictions. Is this the same as your case that you say is not allowed?", "votes": null }, { "id": "1021524", "postDate": "09/22/2020 00:00:48", "content": "You can transform to png but it must be within the committed notebook. You cannot save the transformed images in a dataset and try and access them in the committed notebook.
", "rawMarkdown": "You can transform to png but it must be within the committed notebook. You cannot save the transformed images in a dataset and try and access them in the committed notebook.", "votes": null }, { "id": "1021830", "postDate": "09/22/2020 06:40:51", "content": "Yes, that's what I'm doing in the commited notebook where I just save the test images on a provisional working directory where I do heavy transformations of the image before accessing them at predict stage. And I delete these provisional working directories in the same notebook.
", "rawMarkdown": "Yes, that's what I'm doing in the commited notebook where I just save the test images on a provisional working directory where I do heavy transformations of the image before accessing them at predict stage. And I delete these provisional working directories in the same notebook.", "votes": null }, { "id": "1022283", "postDate": "09/22/2020 12:28:50", "content": "Hi @lwendo,
\nwhen Kaggle evaluates your notebook to give it a score it just replaces the file \"test.csv\" by another one that i much bigger and they put the additional data in their test dataset, alright ?
\nTherefore, if you use another dataset folder (even if you build it with Kaggle) or another csv to make you submission you gonna get the baseline score which is -13.1045 (which is the score that you have when you submit the submission.csv file).
\nIn order to solve your problem, you could use a get_img() function that apply your \"heavy transformations\" before feeding your model (or you can create the dataset in your dataset notebook but I am afraid that you overpass the time limit).
\nI hope I helped :)
I understand. Thanks for your advise. Will try implementing a get_image function while trying to control the time limit.
", "rawMarkdown": "I understand. Thanks for your advise. Will try implementing a get_image function while trying to control the time limit.", "votes": null } ], "comments": [ { "id": 1002579, "author_name": "cswwp347724", "author_url": "", "post_date": "09/08/2020 08:37:54", "content": "Hi @berniechou your fixed it?
", "votes": null, "replies": [ { "id": 1002717, "author_name": "berniechou", "author_url": "", "post_date": "09/08/2020 11:26:21", "content": "yes i was using an external dataset (the same images but as png and not dcm) to make my submission.csv, we have to use the test dataset provided by kaggle.
", "votes": null, "replies": [] }, { "id": 1021519, "author_name": "lwendo", "author_url": "", "post_date": "09/21/2020 23:58:21", "content": "Hello @berniechou . Im having the same problem and always scoring -13.1045. On inference I am using the oficial test dataset from imput to make some png images, which I save on the output directory, then transform the images, then load the png and make predictions. Is this the same as your case that you say is not allowed?
", "votes": null, "replies": [ { "id": 1021524, "author_name": "richardepstein", "author_url": "", "post_date": "09/22/2020 00:00:48", "content": "You can transform to png but it must be within the committed notebook. You cannot save the transformed images in a dataset and try and access them in the committed notebook.
", "votes": null, "replies": [] }, { "id": 1021830, "author_name": "lwendo", "author_url": "", "post_date": "09/22/2020 06:40:51", "content": "Yes, that's what I'm doing in the commited notebook where I just save the test images on a provisional working directory where I do heavy transformations of the image before accessing them at predict stage. And I delete these provisional working directories in the same notebook.
", "votes": null, "replies": [] }, { "id": 1022283, "author_name": "berniechou", "author_url": "", "post_date": "09/22/2020 12:28:50", "content": "Hi @lwendo,
\nwhen Kaggle evaluates your notebook to give it a score it just replaces the file \"test.csv\" by another one that i much bigger and they put the additional data in their test dataset, alright ?
\nTherefore, if you use another dataset folder (even if you build it with Kaggle) or another csv to make you submission you gonna get the baseline score which is -13.1045 (which is the score that you have when you submit the submission.csv file).
\nIn order to solve your problem, you could use a get_img() function that apply your \"heavy transformations\" before feeding your model (or you can create the dataset in your dataset notebook but I am afraid that you overpass the time limit).
\nI hope I helped :)
I understand. Thanks for your advise. Will try implementing a get_image function while trying to control the time limit.
", "votes": null, "replies": [] } ] } ] } ], "raw_markdown_by_id": { "969473": "Hi everyone, \nI made 5 different submissions with 5 different outputs but my score is still equals to -13.1045. Someone has any idea on why it does this ?\nThank you", "1002579": "Hi @berniechou your fixed it?", "1002717": "yes i was using an external dataset (the same images but as png and not dcm) to make my submission.csv, we have to use the test dataset provided by kaggle.", "1021519": "Hello @berniechou . Im having the same problem and always scoring -13.1045. On inference I am using the oficial test dataset from imput to make some png images, which I save on the output directory, then transform the images, then load the png and make predictions. Is this the same as your case that you say is not allowed?", "1021524": "You can transform to png but it must be within the committed notebook. You cannot save the transformed images in a dataset and try and access them in the committed notebook.", "1021830": "Yes, that's what I'm doing in the commited notebook where I just save the test images on a provisional working directory where I do heavy transformations of the image before accessing them at predict stage. And I delete these provisional working directories in the same notebook.", "1022283": "Hi @lwendo, \nwhen Kaggle evaluates your notebook to give it a score it just replaces the file \"test.csv\" by another one that i much bigger and they put the additional data in their test dataset, alright ? \nTherefore, if you use another dataset folder (even if you build it with Kaggle) or another csv to make you submission you gonna get the baseline score which is -13.1045 (which is the score that you have when you submit the submission.csv file).\nIn order to solve your problem, you could use a get_img() function that apply your \"heavy transformations\" before feeding your model (or you can create the dataset in your dataset notebook but I am afraid that you overpass the time limit). \nI hope I helped :)", "1022428": "I understand. Thanks for your advise. Will try implementing a get_image function while trying to control the time limit." }, "source": "meta" }