![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2262353%2F0e2fb8b43d2ec8cea211bf21daf9d064%2FCapture1.PNG?generation=1591938909716954&alt=media\")
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2262353%2Fec760fa0e8f4cfde4871152a55b39b93%2FCapture2.PNG?generation=1591938903711990&alt=media\")
\nHow can we prevent this from happening?My labels are of the form\narray([0., 0., 0., 0., 0., 0.], dtype=float32) . When I create an ImageList, the y value is truncated to the first two values of the label array. \n\n\nHow can we prevent this from happening?
How are you building your ImageList? My v1 is a bit out of practice but I can try to see anything noticeable. If possible I’d recommend learning to use the v2, it’s high and mid level API is much more user friendly
", "rawMarkdown": "How are you building your ImageList? My v1 is a bit out of practice but I can try to see anything noticeable. If possible I’d recommend learning to use the v2, it’s high and mid level API is much more user friendly ", "votes": 1, "replies": [ { "id": 885375, "postDate": "2020-06-14T06:27:34.960Z", "content": "I used the code used by @iafoss in his https://www.kaggle.com/iafoss/panda-concat-tile-pooling-starter-0-79-lb kernel. Instead of isup_grade. I created a column called ord_label which one hot encodes the isup grade column in an ordinal fashion. For instance, 0 will be like [0,0,0,0,0,0] and 3 will be like [1,1,1,0,0,0].
", "rawMarkdown": "I used the code used by @iafoss in his https://www.kaggle.com/iafoss/panda-concat-tile-pooling-starter-0-79-lb kernel. Instead of isup_grade. I created a column called ord_label which one hot encodes the isup grade column in an ordinal fashion. For instance, 0 will be like [0,0,0,0,0,0] and 3 will be like [1,1,1,0,0,0]." }, { "id": 905386, "postDate": "2020-06-28T14:05:21.900Z", "content": "I'm currently using FastAI version 1 so my results might be a little different. Instead of creating a column which has all the one hot encoded predictions I created a target column for each.
\n\nIf your using FastAIs databunch object you can simply add this line to it once you've expanded your dataframe appropriately.\nlabel_col=['grade_1', 'grade_2', 'grade_3', 'grade_4', 'grade_5']
I did the same and overcame the issue.
", "rawMarkdown": "I did the same and overcame the issue. " } ] }, { "id": 882734, "postDate": "2020-06-12T05:15:59.083Z", "content": "My labels are of the form\narray([0., 0., 0., 0., 0., 0.], dtype=float32) . When I create an ImageList, the y value is truncated to the first two values of the label array. \n\n\nHow can we prevent this from happening?
@venky2506 \nI dont face such issue\n\nlearn2.data.train_ds.x.items[-3],data.train_ds.y[-3].data\n('../input/tiles_set_200/train_images/train_images/ffcee00fd033d3ece1408035a7fd4ea7.png',\n array([1., 1., 0., 0., 0., 0.], dtype=float32),\n
Are you using fastaiv1 or fastaiv2?
", "rawMarkdown": "Are you using fastaiv1 or fastaiv2?", "replies": [ { "id": 883456, "postDate": "2020-06-12T16:40:55.553Z", "content": "Version 1
", "rawMarkdown": "Version 1\n" } ] }, { "id": 885374, "postDate": "2020-06-14T06:27:10.953Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 885135, "author_name": "Zach Mueller", "author_url": "", "post_date": "2020-06-13T23:54:40.577000", "content": "How are you building your ImageList? My v1 is a bit out of practice but I can try to see anything noticeable. If possible I’d recommend learning to use the v2, it’s high and mid level API is much more user friendly
", "votes": 1, "replies": [ { "id": 885375, "author_name": "Son of Anton v3.0", "author_url": "", "post_date": "2020-06-14T06:27:34.960000", "content": "I used the code used by @iafoss in his https://www.kaggle.com/iafoss/panda-concat-tile-pooling-starter-0-79-lb kernel. Instead of isup_grade. I created a column called ord_label which one hot encodes the isup grade column in an ordinal fashion. For instance, 0 will be like [0,0,0,0,0,0] and 3 will be like [1,1,1,0,0,0].
", "votes": 0, "replies": [] }, { "id": 905386, "author_name": "Neil Kloper", "author_url": "", "post_date": "2020-06-28T14:05:21.900000", "content": "I'm currently using FastAI version 1 so my results might be a little different. Instead of creating a column which has all the one hot encoded predictions I created a target column for each.
\n\nIf your using FastAIs databunch object you can simply add this line to it once you've expanded your dataframe appropriately.\nlabel_col=['grade_1', 'grade_2', 'grade_3', 'grade_4', 'grade_5']
I did the same and overcame the issue.
", "votes": 0, "replies": [] } ] }, { "id": 899510, "author_name": "Jaideep", "author_url": "", "post_date": "2020-06-24T09:29:53.447000", "content": "@venky2506 \nI dont face such issue\n\nlearn2.data.train_ds.x.items[-3],data.train_ds.y[-3].data\n('../input/tiles_set_200/train_images/train_images/ffcee00fd033d3ece1408035a7fd4ea7.png',\n array([1., 1., 0., 0., 0., 0.], dtype=float32),\n
Are you using fastaiv1 or fastaiv2?
", "votes": 0, "replies": [ { "id": 883456, "author_name": "Son of Anton v3.0", "author_url": "", "post_date": "2020-06-12T16:40:55.553000", "content": "Version 1
", "votes": 0, "replies": [] } ] }, { "id": 885374, "author_name": "", "author_url": "", "post_date": "2020-06-14T06:27:10.953000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "885135": "How are you building your ImageList? My v1 is a bit out of practice but I can try to see anything noticeable. If possible I’d recommend learning to use the v2, it’s high and mid level API is much more user friendly ", "882734": " My labels are of the form\narray([0., 0., 0., 0., 0., 0.], dtype=float32) . When I create an ImageList, the y value is truncated to the first two values of the label array. \n\n\nHow can we prevent this from happening?\n", "899510": "@venky2506 \nI dont face such issue\n```\nlearn2.data.train_ds.x.items[-3],data.train_ds.y[-3].data\n('../input/tiles_set_200/train_images/train_images/ffcee00fd033d3ece1408035a7fd4ea7.png',\n array([1., 1., 0., 0., 0., 0.], dtype=float32),\n```\n \n ", "883430": "Are you using fastaiv1 or fastaiv2?", "885374": "" } }