{ "id": 223610, "title": "Some explanations about predicted submissions?", "url": "/competitions/hpa-single-cell-image-classification/discussion/223610", "author_name": "GitMach", "post_date": "2021-03-04T16:30:42.923000", "votes": 1, "comment_count": 3, "views": 0, "content": "

Hi,

\n

I'm quite confused about the submission form, more precisely between what we're predicting and what we should submit.
\nI'll try ro do my best to explain:
\nThis is about a multi label classification problem, right, in our train.csv we've the file names for the images and the associated labels.
\nFor each image we've one or more labels associated.
\nFor instance first row from train.csv,
\nFile = 5c27f04c-bb99-11e8-b2b9-ac1f6b6435d0
\nLabels = 8|5|0
\nBut the submission file are in this format :
\nImageID,ImageWidth,ImageHeight,PredictionString
\nI've seen, this submission format

\n

0 0.0178 eNqtVL0OgyAQfiVIz4TBoYODiVdKE2odGDqYR.

\n

Following discussion via the link :
\nhttps://www.kaggle.com/c/hpa-single-cell-image-classification/discussion/219375

\n

I understood :

\n

0 is the predicted class-label corresponding to Nucleoplasm

\n

1 is the confidence of the predicted class label Nucleoplasm

\n

eNoLCAgIMA…EABJkBdQ== is the RLE, ZLIB compressed and Base64 encoded segmentation mask for the cell we are predicting on (that we are saying has the Nucleoplasm class).

\n

My question is :
\nGoing back to first image, we've 3 labels for this image, and we're submitting predictions for only one label? What's happened with the other 2 labels \"8\" and \"5\"?
\nShall we only get the higher probability ? I mean, predictions return an array with probabilities for each class right ? so shall we only get the higher probability ?

\n

It's quite confusing, so if you could help, that would be great.

", "messages": [ { "id": 1226552, "postDate": "2021-03-04T16:30:42.923Z", "content": "

Hi,

\n

I'm quite confused about the submission form, more precisely between what we're predicting and what we should submit.
\nI'll try ro do my best to explain:
\nThis is about a multi label classification problem, right, in our train.csv we've the file names for the images and the associated labels.
\nFor each image we've one or more labels associated.
\nFor instance first row from train.csv,
\nFile = 5c27f04c-bb99-11e8-b2b9-ac1f6b6435d0
\nLabels = 8|5|0
\nBut the submission file are in this format :
\nImageID,ImageWidth,ImageHeight,PredictionString
\nI've seen, this submission format

\n

0 0.0178 eNqtVL0OgyAQfiVIz4TBoYODiVdKE2odGDqYR.

\n

Following discussion via the link :
\nhttps://www.kaggle.com/c/hpa-single-cell-image-classification/discussion/219375

\n

I understood :

\n

0 is the predicted class-label corresponding to Nucleoplasm

\n

1 is the confidence of the predicted class label Nucleoplasm

\n

eNoLCAgIMA…EABJkBdQ== is the RLE, ZLIB compressed and Base64 encoded segmentation mask for the cell we are predicting on (that we are saying has the Nucleoplasm class).

\n

My question is :
\nGoing back to first image, we've 3 labels for this image, and we're submitting predictions for only one label? What's happened with the other 2 labels \"8\" and \"5\"?
\nShall we only get the higher probability ? I mean, predictions return an array with probabilities for each class right ? so shall we only get the higher probability ?

\n

It's quite confusing, so if you could help, that would be great.

", "rawMarkdown": "Hi,\n\nI'm quite confused about the submission form, more precisely between what we're predicting and what we should submit.\nI'll try ro do my best to explain:\nThis is about a multi label classification problem, right, in our train.csv we've the file names for the images and the associated labels. \nFor each image we've one or more labels associated. \nFor instance first row from train.csv,\nFile = 5c27f04c-bb99-11e8-b2b9-ac1f6b6435d0 \nLabels = 8|5|0\nBut the submission file are in this format :\nImageID,ImageWidth,ImageHeight,PredictionString\nI've seen, this submission format \n\n`0 0.0178 eNqtVL0OgyAQfiVIz4TBoYODiVdKE2odGDqYR.`\n\nFollowing discussion via the link :\nhttps://www.kaggle.com/c/hpa-single-cell-image-classification/discussion/219375\n\nI understood :\n\n0 is the predicted class-label corresponding to Nucleoplasm\n\n1 is the confidence of the predicted class label Nucleoplasm\n\neNoLCAgIMA...EABJkBdQ== is the RLE, ZLIB compressed and Base64 encoded segmentation mask for the cell we are predicting on (that we are saying has the Nucleoplasm class).\n\n\nMy question is : \nGoing back to first image, we've 3 labels for this image, and we're submitting predictions for only one label? What's happened with the other 2 labels \"8\" and \"5\"?\nShall we only get the higher probability ? I mean, predictions return an array with probabilities for each class right ? so shall we only get the higher probability ?\n\nIt's quite confusing, so if you could help, that would be great.", "votes": 1 }, { "id": 1227371, "postDate": "2021-03-05T13:13:07.180Z", "content": "

In one image, for each cell (cell mask), you have to predict labels (0~18, not necessarily only one) with confidence.

", "rawMarkdown": "In one image, for each cell (cell mask), you have to predict labels (0~18, not necessarily only one) with confidence.", "replies": [ { "id": 1229595, "postDate": "2021-03-07T13:21:38.300Z", "content": "

Great, we're talking about mAP, right?

", "rawMarkdown": "Great, we're talking about mAP, right?" }, { "id": 1230796, "postDate": "2021-03-08T13:13:28.330Z", "content": "

Yes, we are using the mAP metric.

", "rawMarkdown": "Yes, we are using the mAP metric." } ] } ], "comments": [ { "id": 1227371, "author_name": "cool_rabbit", "author_url": "", "post_date": "2021-03-05T13:13:07.180000", "content": "

In one image, for each cell (cell mask), you have to predict labels (0~18, not necessarily only one) with confidence.

", "votes": 0, "replies": [ { "id": 1229595, "author_name": "GitMach", "author_url": "", "post_date": "2021-03-07T13:21:38.300000", "content": "

Great, we're talking about mAP, right?

", "votes": 0, "replies": [] }, { "id": 1230796, "author_name": "Casper Winsnes", "author_url": "", "post_date": "2021-03-08T13:13:28.330000", "content": "

Yes, we are using the mAP metric.

", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "1226552": "Hi,\n\nI'm quite confused about the submission form, more precisely between what we're predicting and what we should submit.\nI'll try ro do my best to explain:\nThis is about a multi label classification problem, right, in our train.csv we've the file names for the images and the associated labels. \nFor each image we've one or more labels associated. \nFor instance first row from train.csv,\nFile = 5c27f04c-bb99-11e8-b2b9-ac1f6b6435d0 \nLabels = 8|5|0\nBut the submission file are in this format :\nImageID,ImageWidth,ImageHeight,PredictionString\nI've seen, this submission format \n\n`0 0.0178 eNqtVL0OgyAQfiVIz4TBoYODiVdKE2odGDqYR.`\n\nFollowing discussion via the link :\nhttps://www.kaggle.com/c/hpa-single-cell-image-classification/discussion/219375\n\nI understood :\n\n0 is the predicted class-label corresponding to Nucleoplasm\n\n1 is the confidence of the predicted class label Nucleoplasm\n\neNoLCAgIMA...EABJkBdQ== is the RLE, ZLIB compressed and Base64 encoded segmentation mask for the cell we are predicting on (that we are saying has the Nucleoplasm class).\n\n\nMy question is : \nGoing back to first image, we've 3 labels for this image, and we're submitting predictions for only one label? What's happened with the other 2 labels \"8\" and \"5\"?\nShall we only get the higher probability ? I mean, predictions return an array with probabilities for each class right ? so shall we only get the higher probability ?\n\nIt's quite confusing, so if you could help, that would be great.", "1227371": "In one image, for each cell (cell mask), you have to predict labels (0~18, not necessarily only one) with confidence." } }