At first i thought that CVC would be either be normal, abnormal or neither making this a multi-class problem where we can use softmax output. However, i notice that some train examples are simultaneously normal and abnormal, thus perhaps this is multi-label and we should use sigmoid output.
What does it mean for CVC to be both normal and abnormal?
I do notice that ETT is multi-class. It is either normal, abnormal, borderline, or neither. And it takes exactly 1 and only 1 of those classes.
At first i thought that CVC would be either be normal, abnormal or neither making this a multi-class problem where we can use softmax output. However, i notice that some train examples are simultaneously normal and abnormal, thus perhaps this is multi-label and we should use sigmoid output.
What does it mean for CVC to be both normal and abnormal?
I do notice that ETT is multi-class. It is either normal, abnormal, borderline, or neither. And it takes exactly 1 and only 1 of those classes.
Nice question - it seems you do not need to know what the acual problem is to score 0.968 on LB :D
", "rawMarkdown": "Nice question - it seems you do not need to know what the acual problem is to score 0.968 on LB :D", "votes": 21, "replies": [ { "id": 1216240, "postDate": "2021-02-24T06:47:58.717Z", "content": "Once he understands it, he will get to the top positions. :p
\nBest of luck to both of you!
", "rawMarkdown": "Once he understands it, he will get to the top positions. :p\n\nBest of luck to both of you!", "votes": 1 }, { "id": 1216389, "postDate": "2021-02-24T08:41:12.787Z", "content": "\n\nNice question - it seems you do not need to know what the acual problem is to score 0.968 on LB :D
\n
In a lousy day at work, I had a good laugh reading this :) Thank you
", "rawMarkdown": "> Nice question - it seems you do not need to know what the acual problem is to score 0.968 on LB :D\n\nIn a lousy day at work, I had a good laugh reading this :) Thank you ", "votes": 8 }, { "id": 1216939, "postDate": "2021-02-24T16:22:34.720Z", "content": "Sure, that's machine learning 😉
", "rawMarkdown": "Sure, that's machine learning 😉", "votes": 4 }, { "id": 1220039, "postDate": "2021-02-27T13:36:58.417Z", "content": "@raddar ,
\nstill laughing… =)
", "rawMarkdown": "@raddar ,\n\nstill laughing... =)" }, { "id": 1220160, "postDate": "2021-02-27T16:42:51.910Z", "content": "Whether what's the acutal problem, I think bce and ensemble always work
", "rawMarkdown": "Whether what's the acutal problem, I think bce and ensemble always work", "votes": 2 } ] }, { "id": 1214566, "postDate": "2021-02-23T01:15:55.617Z", "content": "Is multi label binary classification, i did the same question here and you can read the host's answer.
", "rawMarkdown": "Is **multi label binary classification**, i did the same question [here](https://www.kaggle.com/c/ranzcr-clip-catheter-line-classification/discussion/203339) and you can read the host's answer.", "votes": 3, "replies": [ { "id": 1214568, "postDate": "2021-02-23T01:19:32.087Z", "content": "Thanks for the link. Yes, the host says this is multi label. But what does it mean to be both normal and abnormal. Can any doctors shed some light on this?
", "rawMarkdown": "Thanks for the link. Yes, the host says this is multi label. But what does it mean to be both normal and abnormal. Can any doctors shed some light on this?", "votes": 6 }, { "id": 1214574, "postDate": "2021-02-23T01:28:10.457Z", "content": "normal: Includes lines (tubes) that were appropriately positioned and did not require repositioningabnormal: Is the reason of this competition, lines (tubes) placed so bad that could lead a medical complication and required immediate repositioning.borderline: Are lines that would ideally require some repositioning but would in most cases still function adequately in their current position.Then it is strange that lines (tubes) are both appropriately positioned and placed so bad simultaneously
In the \"Overview\" section they wrote \"lines\" as plural, i think (maybe i'm wrong) that one is labeled as normal and the other as abnormal. A quick visualization could help. Someone soon is going to answer your question properly.
You can see this Reason why a person can have CVC(Normal) And CVC(Abnormal) values at the same time?.
", "rawMarkdown": "You can see this [Reason why a person can have CVC(Normal) And CVC(Abnormal) values at the same time?](https://www.kaggle.com/c/ranzcr-clip-catheter-line-classification/discussion/216887).", "votes": 4 }, { "id": 1214662, "postDate": "2021-02-23T03:12:20.420Z", "content": "Thanks @h053473666 It sounds like a person can have two tubes whereas one is position correctly and the second is not. I'll review some more images to confirm this.
", "rawMarkdown": "Thanks @h053473666 It sounds like a person can have two tubes whereas one is position correctly and the second is not. I'll review some more images to confirm this.", "votes": 11 }, { "id": 1216943, "postDate": "2021-02-24T16:25:56.820Z", "content": "Maybe this evaluation is helpful: https://www.kaggle.com/docxian/ranzcr-clip-target-structure-analysis
", "rawMarkdown": "Maybe this evaluation is helpful: https://www.kaggle.com/docxian/ranzcr-clip-target-structure-analysis", "votes": 2 }, { "id": 1217164, "postDate": "2021-02-24T20:02:54.710Z", "content": "These are very interesting insights. Thanks to all. 👌
", "rawMarkdown": "These are very interesting insights. Thanks to all. 👌 " }, { "id": 1219770, "postDate": "2021-02-27T08:15:30.300Z", "content": "Yes, it is the case that people can have multiple lines. Some even have >5 lines in this dataset
", "rawMarkdown": "Yes, it is the case that people can have multiple lines. Some even have >5 lines in this dataset" }, { "id": 1220388, "postDate": "2021-02-27T23:02:17.373Z", "content": "Some images contain more than 1 tube of the same kind.
", "rawMarkdown": "Some images contain more than 1 tube of the same kind." } ] }, { "id": 1227927, "postDate": "2021-03-06T00:03:23.287Z", "content": "Thanks everyone for answering my question. It appears the explanation for having seemingly contradicting labels is that some images have more than one CVC inserted. I displayed some EDA here
", "rawMarkdown": "Thanks everyone for answering my question. It appears the explanation for having seemingly contradicting labels is that some images have more than one CVC inserted. I displayed some EDA [here][1]\n\n[1]: https://www.kaggle.com/c/ranzcr-clip-catheter-line-classification/discussion/223880", "votes": 1 }, { "id": 1219771, "postDate": "2021-02-27T08:16:48.393Z", "content": "Regardless of whether it is multi label or multi class because we are being scored against individual columns auc I would think that softmax would yield some undesirable properties. Confidence of one class would affect another and change the sort order.
", "rawMarkdown": "Regardless of whether it is multi label or multi class because we are being scored against individual columns auc I would think that softmax would yield some undesirable properties. Confidence of one class would affect another and change the sort order. ", "votes": 1 }, { "id": 1216238, "postDate": "2021-02-24T06:47:24.680Z", "content": "I had exactly the same question regarding the CVC classes. Thanks for asking the question. :)
\nI think one approach to explore is to train a separate model for CVC classes one that outputs multi-labels.
Or maybe the problem is with the annotated data?
", "rawMarkdown": "I had exactly the same question regarding the CVC classes. Thanks for asking the question. :)\nI think one approach to explore is to train a separate model for CVC classes one that outputs multi-labels.\n\nOr maybe the problem is with the annotated data?", "votes": 1 }, { "id": 1219187, "postDate": "2021-02-26T14:28:09.133Z", "content": "Does system configuration affects your score? in a cometition please answer?
", "rawMarkdown": "Does system configuration affects your score? in a cometition please answer?", "replies": [ { "id": 1219224, "postDate": "2021-02-26T15:19:38.950Z", "content": "What do you mean by system configuration ? Can you elaborate?
Are you asking about the resources one is using, i.e. local GPU vs cloud GPU? Or rather the software used i.e. Pytorch vs Tensorflow? @santoshd3
", "rawMarkdown": "Are you asking about the resources one is using, i.e. local GPU vs cloud GPU? Or rather the software used i.e. Pytorch vs Tensorflow? @santoshd3 " } ] } ], "comments": [ { "id": 1215426, "author_name": "raddar", "author_url": "", "post_date": "2021-02-23T15:44:01.863000", "content": "Nice question - it seems you do not need to know what the acual problem is to score 0.968 on LB :D
", "votes": 21, "replies": [ { "id": 1216240, "author_name": "Yassine Alouini", "author_url": "", "post_date": "2021-02-24T06:47:58.717000", "content": "Once he understands it, he will get to the top positions. :p
\nBest of luck to both of you!
", "votes": 1, "replies": [] }, { "id": 1216389, "author_name": "yukiya", "author_url": "", "post_date": "2021-02-24T08:41:12.787000", "content": "\n\nNice question - it seems you do not need to know what the acual problem is to score 0.968 on LB :D
\n
In a lousy day at work, I had a good laugh reading this :) Thank you
", "votes": 8, "replies": [] }, { "id": 1216939, "author_name": "Chris X", "author_url": "", "post_date": "2021-02-24T16:22:34.720000", "content": "Sure, that's machine learning 😉
", "votes": 4, "replies": [] }, { "id": 1220039, "author_name": "dr. Konya", "author_url": "", "post_date": "2021-02-27T13:36:58.417000", "content": "@raddar ,
\nstill laughing… =)
", "votes": 0, "replies": [] }, { "id": 1220160, "author_name": "sheep", "author_url": "", "post_date": "2021-02-27T16:42:51.910000", "content": "Whether what's the acutal problem, I think bce and ensemble always work
", "votes": 2, "replies": [] } ] }, { "id": 1214566, "author_name": "Hiram Coria 🧬", "author_url": "", "post_date": "2021-02-23T01:15:55.617000", "content": "Is multi label binary classification, i did the same question here and you can read the host's answer.
", "votes": 3, "replies": [ { "id": 1214568, "author_name": "Chris Deotte", "author_url": "", "post_date": "2021-02-23T01:19:32.087000", "content": "Thanks for the link. Yes, the host says this is multi label. But what does it mean to be both normal and abnormal. Can any doctors shed some light on this?
", "votes": 6, "replies": [] }, { "id": 1214574, "author_name": "Hiram Coria 🧬", "author_url": "", "post_date": "2021-02-23T01:28:10.457000", "content": "normal: Includes lines (tubes) that were appropriately positioned and did not require repositioningabnormal: Is the reason of this competition, lines (tubes) placed so bad that could lead a medical complication and required immediate repositioning.borderline: Are lines that would ideally require some repositioning but would in most cases still function adequately in their current position.Then it is strange that lines (tubes) are both appropriately positioned and placed so bad simultaneously
In the \"Overview\" section they wrote \"lines\" as plural, i think (maybe i'm wrong) that one is labeled as normal and the other as abnormal. A quick visualization could help. Someone soon is going to answer your question properly.
You can see this Reason why a person can have CVC(Normal) And CVC(Abnormal) values at the same time?.
", "votes": 4, "replies": [] }, { "id": 1214662, "author_name": "Chris Deotte", "author_url": "", "post_date": "2021-02-23T03:12:20.420000", "content": "Thanks @h053473666 It sounds like a person can have two tubes whereas one is position correctly and the second is not. I'll review some more images to confirm this.
", "votes": 11, "replies": [] }, { "id": 1216943, "author_name": "Chris X", "author_url": "", "post_date": "2021-02-24T16:25:56.820000", "content": "Maybe this evaluation is helpful: https://www.kaggle.com/docxian/ranzcr-clip-target-structure-analysis
", "votes": 2, "replies": [] }, { "id": 1217164, "author_name": "Yassine Alouini", "author_url": "", "post_date": "2021-02-24T20:02:54.710000", "content": "These are very interesting insights. Thanks to all. 👌
", "votes": 0, "replies": [] }, { "id": 1219770, "author_name": "ryches", "author_url": "", "post_date": "2021-02-27T08:15:30.300000", "content": "Yes, it is the case that people can have multiple lines. Some even have >5 lines in this dataset
", "votes": 0, "replies": [] }, { "id": 1220388, "author_name": "Tolga", "author_url": "", "post_date": "2021-02-27T23:02:17.373000", "content": "Some images contain more than 1 tube of the same kind.
", "votes": 0, "replies": [] } ] }, { "id": 1227927, "author_name": "Chris Deotte", "author_url": "", "post_date": "2021-03-06T00:03:23.287000", "content": "Thanks everyone for answering my question. It appears the explanation for having seemingly contradicting labels is that some images have more than one CVC inserted. I displayed some EDA here
", "votes": 1, "replies": [] }, { "id": 1219771, "author_name": "ryches", "author_url": "", "post_date": "2021-02-27T08:16:48.393000", "content": "Regardless of whether it is multi label or multi class because we are being scored against individual columns auc I would think that softmax would yield some undesirable properties. Confidence of one class would affect another and change the sort order.
", "votes": 1, "replies": [] }, { "id": 1216238, "author_name": "Yassine Alouini", "author_url": "", "post_date": "2021-02-24T06:47:24.680000", "content": "I had exactly the same question regarding the CVC classes. Thanks for asking the question. :)
\nI think one approach to explore is to train a separate model for CVC classes one that outputs multi-labels.
Or maybe the problem is with the annotated data?
", "votes": 1, "replies": [] }, { "id": 1219187, "author_name": "Santosh kumar", "author_url": "", "post_date": "2021-02-26T14:28:09.133000", "content": "Does system configuration affects your score? in a cometition please answer?
", "votes": 0, "replies": [ { "id": 1219224, "author_name": "Chris Deotte", "author_url": "", "post_date": "2021-02-26T15:19:38.950000", "content": "What do you mean by system configuration ? Can you elaborate?
Are you asking about the resources one is using, i.e. local GPU vs cloud GPU? Or rather the software used i.e. Pytorch vs Tensorflow? @santoshd3
", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "1214561": "At first i thought that CVC would be either be `normal`, `abnormal` or `neither` making this a multi-class problem where we can use `softmax` output. However, i notice that some train examples are simultaneously `normal` and `abnormal`, thus perhaps this is multi-label and we should use `sigmoid` output.\n\nWhat does it mean for CVC to be both `normal` and `abnormal`?\n\nI do notice that ETT is multi-class. It is either `normal`, `abnormal`, `borderline`, or `neither`. And it takes exactly 1 and only 1 of those classes.", "1215426": "Nice question - it seems you do not need to know what the acual problem is to score 0.968 on LB :D", "1214566": "Is **multi label binary classification**, i did the same question [here](https://www.kaggle.com/c/ranzcr-clip-catheter-line-classification/discussion/203339) and you can read the host's answer.", "1227927": "Thanks everyone for answering my question. It appears the explanation for having seemingly contradicting labels is that some images have more than one CVC inserted. I displayed some EDA [here][1]\n\n[1]: https://www.kaggle.com/c/ranzcr-clip-catheter-line-classification/discussion/223880", "1219771": "Regardless of whether it is multi label or multi class because we are being scored against individual columns auc I would think that softmax would yield some undesirable properties. Confidence of one class would affect another and change the sort order. ", "1216238": "I had exactly the same question regarding the CVC classes. Thanks for asking the question. :)\nI think one approach to explore is to train a separate model for CVC classes one that outputs multi-labels.\n\nOr maybe the problem is with the annotated data?", "1219187": "Does system configuration affects your score? in a cometition please answer?" } }