{ "id": 539949, "title": "Why use Regressors instead of Classifier?", "url": "/competitions/child-mind-institute-problematic-internet-use/discussion/539949", "author_name": "Calvin Hui", "post_date": "2024-10-11T15:04:02.718000", "votes": 8, "comment_count": 3, "views": 0, "content": "

Based on my understanding, the dependent variable sii is supposed to be {0, 1, 2, 3}, but when I looked at other people's code, I noticed that most of the high-scoring models use a regressor instead of a classifier. I'd like to ask why a Regressor performs better in this classification problem?

", "messages": [ { "id": 3014763, "postDate": "2024-10-11T15:04:02.717Z", "content": "

Based on my understanding, the dependent variable sii is supposed to be {0, 1, 2, 3}, but when I looked at other people's code, I noticed that most of the high-scoring models use a regressor instead of a classifier. I'd like to ask why a Regressor performs better in this classification problem?

", "rawMarkdown": "Based on my understanding, the dependent variable **sii** is supposed to be {0, 1, 2, 3}, but when I looked at other people's code, I noticed that most of the high-scoring models use a regressor instead of a classifier. I'd like to ask why a Regressor performs better in this classification problem?", "votes": 8 }, { "id": 3014977, "postDate": "2024-10-11T18:47:52.933Z", "content": "

@ccyhui this is an ordinal regressor problem where the relation between classes 0-1-2-3 indicate an ordinal increase in domain characteristics. In this case, as we go up from 0-3, we are sure that the child became more addicted to internet leading to physical and mental health problems. This type of problems need a special type of treatment not always possible with classifiers. I tried to use a classifier as my first submission a while ago and landed with a CV score of 0.33 and a LB score of 0.34x. Regressors gave me my current CV and LB score as on date @ccyhui

", "rawMarkdown": "@ccyhui this is an ordinal regressor problem where the relation between classes 0-1-2-3 indicate an **ordinal increase in domain characteristics**. In this case, as we go up from 0-3, we are sure that the child became more addicted to internet leading to physical and mental health problems. This type of problems need a special type of treatment not always possible with classifiers. I tried to use a classifier as my first submission a while ago and landed with a CV score of 0.33 and a LB score of 0.34x. Regressors gave me my current CV and LB score as on date @ccyhui ", "votes": 6 }, { "id": 3014837, "postDate": "2024-10-11T16:22:57Z", "content": "

In the case of classification, the order of classes is not considered as you can see the loss function.
\nHowever in this competition, there is an ordinal relationship between class labels.
\nSuch a problem setting is generally referred to as ordinal regression.

", "rawMarkdown": "In the case of classification, the order of classes is not considered as you can see the loss function.\nHowever in this competition, there is an ordinal relationship between class labels. \nSuch a problem setting is generally referred to as ordinal regression.", "votes": 2 }, { "id": 3014832, "postDate": "2024-10-11T16:16:02.620Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 3014977, "author_name": "Ravi Ramakrishnan", "author_url": "", "post_date": "2024-10-11T18:47:52.933000", "content": "

@ccyhui this is an ordinal regressor problem where the relation between classes 0-1-2-3 indicate an ordinal increase in domain characteristics. In this case, as we go up from 0-3, we are sure that the child became more addicted to internet leading to physical and mental health problems. This type of problems need a special type of treatment not always possible with classifiers. I tried to use a classifier as my first submission a while ago and landed with a CV score of 0.33 and a LB score of 0.34x. Regressors gave me my current CV and LB score as on date @ccyhui

", "votes": 6, "replies": [] }, { "id": 3014837, "author_name": "Aurora_blue", "author_url": "", "post_date": "2024-10-11T16:22:57", "content": "

In the case of classification, the order of classes is not considered as you can see the loss function.
\nHowever in this competition, there is an ordinal relationship between class labels.
\nSuch a problem setting is generally referred to as ordinal regression.

", "votes": 2, "replies": [] }, { "id": 3014832, "author_name": "", "author_url": "", "post_date": "2024-10-11T16:16:02.620000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "3014763": "Based on my understanding, the dependent variable **sii** is supposed to be {0, 1, 2, 3}, but when I looked at other people's code, I noticed that most of the high-scoring models use a regressor instead of a classifier. I'd like to ask why a Regressor performs better in this classification problem?", "3014977": "@ccyhui this is an ordinal regressor problem where the relation between classes 0-1-2-3 indicate an **ordinal increase in domain characteristics**. In this case, as we go up from 0-3, we are sure that the child became more addicted to internet leading to physical and mental health problems. This type of problems need a special type of treatment not always possible with classifiers. I tried to use a classifier as my first submission a while ago and landed with a CV score of 0.33 and a LB score of 0.34x. Regressors gave me my current CV and LB score as on date @ccyhui ", "3014837": "In the case of classification, the order of classes is not considered as you can see the loss function.\nHowever in this competition, there is an ordinal relationship between class labels. \nSuch a problem setting is generally referred to as ordinal regression.", "3014832": "" } }