It seems versions of this have been asked in various parts in other discussion topics, with partial responses/comments, so I am looking for a clear statement from the admins.
\n\nThe rules stated that stage1_test data would be released at the end of stage1. However, the masks for stage1_test data were released in RLE format as a .csv file. Our team intended to add stage1_test data to our training data during stage 2, however we were expecting those stage1_test masks to be released in the same format as the stage1 training data - i.e. a separate .png file for each nuclei. We had basic code to try to re-construct a .png file from RLE/.csv data, however that code needs to be modified to support creating the stage1_test masks in the format we were expecting.
\n\nSo, my question: Are we allowed to write code to translate the released stage1_test RLE/.csv code into the expected mask format, so that we can use that stage1_test data to retrain our model?
\n\nThanks!
", "messages": [ { "id": "313497", "postDate": "04/13/2018 10:46:20", "content": "It seems versions of this have been asked in various parts in other discussion topics, with partial responses/comments, so I am looking for a clear statement from the admins.
\n\nThe rules stated that stage1_test data would be released at the end of stage1. However, the masks for stage1_test data were released in RLE format as a .csv file. Our team intended to add stage1_test data to our training data during stage 2, however we were expecting those stage1_test masks to be released in the same format as the stage1 training data - i.e. a separate .png file for each nuclei. We had basic code to try to re-construct a .png file from RLE/.csv data, however that code needs to be modified to support creating the stage1_test masks in the format we were expecting.
\n\nSo, my question: Are we allowed to write code to translate the released stage1_test RLE/.csv code into the expected mask format, so that we can use that stage1_test data to retrain our model?
\n\nThanks!
", "rawMarkdown": "It seems versions of this have been asked in various parts in other discussion topics, with partial responses/comments, so I am looking for a clear statement from the admins.\n\nThe rules stated that stage1_test data would be released at the end of stage1. However, the masks for stage1_test data were released in RLE format as a .csv file. Our team intended to add stage1_test data to our training data during stage 2, however we were expecting those stage1_test masks to be released in the same format as the stage1 training data - i.e. a separate .png file for each nuclei. We had basic code to try to re-construct a .png file from RLE/.csv data, however that code needs to be modified to support creating the stage1_test masks in the format we were expecting.\n\nSo, my question: Are we allowed to write code to translate the released stage1_test RLE/.csv code into the expected mask format, so that we can use that stage1_test data to retrain our model?\n\nThanks!", "votes": null }, { "id": "313536", "postDate": "04/13/2018 12:30:35", "content": "I think it's ok. This modification is not related to the science.
", "rawMarkdown": "I think it's ok. This modification is not related to the science.", "votes": null }, { "id": "313537", "postDate": "04/13/2018 12:32:25", "content": "Thanks @Yan, however, we need a crystal clear response from the host or the Kaggle admin before training/submitting anything.
", "rawMarkdown": "Thanks @Yan, however, we need a crystal clear response from the host or the Kaggle admin before training/submitting anything.", "votes": null }, { "id": "313737", "postDate": "04/13/2018 19:12:52", "content": "A Kaggle admin responded in another thread:
\n\nYes, you can retrain with stage 1 test data, as long as your modeling code doesn't change.
\n\nYou are allowed to add code to translate RLE encoded masks into masks.
", "rawMarkdown": "A Kaggle admin responded in another thread:\n\nYes, you can retrain with stage 1 test data, as long as your modeling code doesn't change.\n\nYou are allowed to add code to translate RLE encoded masks into masks.", "votes": null }, { "id": "313738", "postDate": "04/13/2018 19:17:16", "content": "Yes. That code (translating RLE masks to stage 1 masks format) is fine.
", "rawMarkdown": "Yes. That code (translating RLE masks to stage 1 masks format) is fine.", "votes": null }, { "id": "313746", "postDate": "04/13/2018 19:24:47", "content": "Excellent, thanks for the confirmation!
", "rawMarkdown": "Excellent, thanks for the confirmation!", "votes": null } ], "comments": [ { "id": 313536, "author_name": "wayne999", "author_url": "", "post_date": "04/13/2018 12:30:35", "content": "I think it's ok. This modification is not related to the science.
", "votes": null, "replies": [ { "id": 313537, "author_name": "nicolefinnie", "author_url": "", "post_date": "04/13/2018 12:32:25", "content": "Thanks @Yan, however, we need a crystal clear response from the host or the Kaggle admin before training/submitting anything.
", "votes": null, "replies": [] }, { "id": 313737, "author_name": "jliamfinnie", "author_url": "", "post_date": "04/13/2018 19:12:52", "content": "A Kaggle admin responded in another thread:
\n\nYes, you can retrain with stage 1 test data, as long as your modeling code doesn't change.
\n\nYou are allowed to add code to translate RLE encoded masks into masks.
", "votes": null, "replies": [] } ] }, { "id": 313738, "author_name": "wendykan", "author_url": "", "post_date": "04/13/2018 19:17:16", "content": "Yes. That code (translating RLE masks to stage 1 masks format) is fine.
", "votes": null, "replies": [ { "id": 313746, "author_name": "jliamfinnie", "author_url": "", "post_date": "04/13/2018 19:24:47", "content": "Excellent, thanks for the confirmation!
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "313497": "It seems versions of this have been asked in various parts in other discussion topics, with partial responses/comments, so I am looking for a clear statement from the admins.\n\nThe rules stated that stage1_test data would be released at the end of stage1. However, the masks for stage1_test data were released in RLE format as a .csv file. Our team intended to add stage1_test data to our training data during stage 2, however we were expecting those stage1_test masks to be released in the same format as the stage1 training data - i.e. a separate .png file for each nuclei. We had basic code to try to re-construct a .png file from RLE/.csv data, however that code needs to be modified to support creating the stage1_test masks in the format we were expecting.\n\nSo, my question: Are we allowed to write code to translate the released stage1_test RLE/.csv code into the expected mask format, so that we can use that stage1_test data to retrain our model?\n\nThanks!", "313536": "I think it's ok. This modification is not related to the science.", "313537": "Thanks @Yan, however, we need a crystal clear response from the host or the Kaggle admin before training/submitting anything.", "313737": "A Kaggle admin responded in another thread:\n\nYes, you can retrain with stage 1 test data, as long as your modeling code doesn't change.\n\nYou are allowed to add code to translate RLE encoded masks into masks.", "313738": "Yes. That code (translating RLE masks to stage 1 masks format) is fine.", "313746": "Excellent, thanks for the confirmation!" }, "source": "meta" }