{ "id": 250610, "title": "How can we predict four target variable values at the same time?", "url": "/competitions/mlb-player-digital-engagement-forecasting/discussion/250610", "author_name": "", "post_date": "2021-07-03T13:55:46.045212Z", "votes": null, "comment_count": 4, "views": 0, "content": "

Hello everyone, I'm just a beginner in Kaggle.

\n

I'm trying to make prediction on target variables, only to have realized that the given data have
\nfour target variables which I've never met.

\n

Should I predict values for four variables at the same time? Otherwise, what technique could I use to overcome this hardship? If you cannot give me a direct answer, would you give me some clues for this topic?

", "messages": [ { "id": "1374672", "postDate": "07/03/2021 13:55:46", "content": "

Hello everyone, I'm just a beginner in Kaggle.

\n

I'm trying to make prediction on target variables, only to have realized that the given data have
\nfour target variables which I've never met.

\n

Should I predict values for four variables at the same time? Otherwise, what technique could I use to overcome this hardship? If you cannot give me a direct answer, would you give me some clues for this topic?

", "rawMarkdown": "Hello everyone, I'm just a beginner in Kaggle.\n\nI'm trying to make prediction on target variables, only to have realized that the given data have\nfour target variables which I've never met.\n\nShould I predict values for four variables at the same time? Otherwise, what technique could I use to overcome this hardship? If you cannot give me a direct answer, would you give me some clues for this topic?", "votes": null }, { "id": "1374700", "postDate": "07/03/2021 14:22:29", "content": "

As a start you can just train 4 classifiers, one for each target.

\n

Another option is to use a neural network with 4 output nodes (see some of the public notebooks: e.g., https://www.kaggle.com/ulrich07/mlb-debug-ann).

\n

One final option is to look at sklearn.multioutput.

", "rawMarkdown": "As a start you can just train 4 classifiers, one for each target. \n\nAnother option is to use a neural network with 4 output nodes (see some of the public notebooks: e.g., https://www.kaggle.com/ulrich07/mlb-debug-ann).\n\nOne final option is to look at [sklearn.multioutput](https://scikit-learn.org/stable/modules/classes.html#module-sklearn.multioutput).", "votes": null }, { "id": "1375027", "postDate": "07/03/2021 19:30:55", "content": "

Youn can choose between 4 different models or 1 model for all targets, you have the MultipleOutputRegressor wrapper on sklearn or 4 units on final layer if you are using NN

", "rawMarkdown": "Youn can choose between 4 different models or 1 model for all targets, you have the MultipleOutputRegressor wrapper on sklearn or 4 units on final layer if you are using NN", "votes": null }, { "id": "1375215", "postDate": "07/04/2021 03:08:35", "content": "

Lightgbm and catboost can easily be used to run 4 models. Several shared kernels showing how to do lightgbm - not sure I have seen a catboost. You can use the same feature set for all 4 models. If your getting a decent amount up the leaderboard than a little bit of coding can get you using different feature sets and light parameters.

\n

This is a very ugly data set to put togeather and make sense. I would recommend using lightgbm and spend most of your time getting features built.

\n

The problem - the submission process is a real bitch - so start slow and make sure you can build a basic light that does 4 models and is successful in prediction. Add a FEW features at a time after that -

", "rawMarkdown": "Lightgbm and catboost can easily be used to run 4 models. Several shared kernels showing how to do lightgbm - not sure I have seen a catboost. You can use the same feature set for all 4 models. If your getting a decent amount up the leaderboard than a little bit of coding can get you using different feature sets and light parameters.\n\nThis is a very ugly data set to put togeather and make sense. I would recommend using lightgbm and spend most of your time getting features built.\n\nThe problem - the submission process is a real bitch - so start slow and make sure you can build a basic light that does 4 models and is successful in prediction. Add a FEW features at a time after that -", "votes": null }, { "id": "1375507", "postDate": "07/04/2021 09:07:16", "content": "

Hi @junyeongahnn,

\n

I believe you mean that you want to predict a single dependent variable by multiple independent variables, if that what you mean, so you can work a regression model, easy. but if you want to predict all four variables, so I may recommend doing so by using four different models and that will require setting each of your variables as a target variable once. That if you can be so sure about a circular dependence among your variables (every and each variable dependent on others) which's rare to happen to my little experience.

", "rawMarkdown": "Hi @junyeongahnn,\n\nI believe you mean that you want to predict a single dependent variable by multiple independent variables, if that what you mean, so you can work a regression model, easy. but if you want to predict all four variables, so I may recommend doing so by using four different models and that will require setting each of your variables as a target variable once. That if you can be so sure about a circular dependence among your variables (every and each variable dependent on others) which's rare to happen to my little experience.", "votes": null } ], "comments": [ { "id": 1374700, "author_name": "fchmiel", "author_url": "", "post_date": "07/03/2021 14:22:29", "content": "

As a start you can just train 4 classifiers, one for each target.

\n

Another option is to use a neural network with 4 output nodes (see some of the public notebooks: e.g., https://www.kaggle.com/ulrich07/mlb-debug-ann).

\n

One final option is to look at sklearn.multioutput.

", "votes": null, "replies": [] }, { "id": 1375027, "author_name": "enric1296", "author_url": "", "post_date": "07/03/2021 19:30:55", "content": "

Youn can choose between 4 different models or 1 model for all targets, you have the MultipleOutputRegressor wrapper on sklearn or 4 units on final layer if you are using NN

", "votes": null, "replies": [] }, { "id": 1375215, "author_name": "pcjimmmy", "author_url": "", "post_date": "07/04/2021 03:08:35", "content": "

Lightgbm and catboost can easily be used to run 4 models. Several shared kernels showing how to do lightgbm - not sure I have seen a catboost. You can use the same feature set for all 4 models. If your getting a decent amount up the leaderboard than a little bit of coding can get you using different feature sets and light parameters.

\n

This is a very ugly data set to put togeather and make sense. I would recommend using lightgbm and spend most of your time getting features built.

\n

The problem - the submission process is a real bitch - so start slow and make sure you can build a basic light that does 4 models and is successful in prediction. Add a FEW features at a time after that -

", "votes": null, "replies": [] }, { "id": 1375507, "author_name": "taricov", "author_url": "", "post_date": "07/04/2021 09:07:16", "content": "

Hi @junyeongahnn,

\n

I believe you mean that you want to predict a single dependent variable by multiple independent variables, if that what you mean, so you can work a regression model, easy. but if you want to predict all four variables, so I may recommend doing so by using four different models and that will require setting each of your variables as a target variable once. That if you can be so sure about a circular dependence among your variables (every and each variable dependent on others) which's rare to happen to my little experience.

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1374672": "Hello everyone, I'm just a beginner in Kaggle.\n\nI'm trying to make prediction on target variables, only to have realized that the given data have\nfour target variables which I've never met.\n\nShould I predict values for four variables at the same time? Otherwise, what technique could I use to overcome this hardship? If you cannot give me a direct answer, would you give me some clues for this topic?", "1374700": "As a start you can just train 4 classifiers, one for each target. \n\nAnother option is to use a neural network with 4 output nodes (see some of the public notebooks: e.g., https://www.kaggle.com/ulrich07/mlb-debug-ann).\n\nOne final option is to look at [sklearn.multioutput](https://scikit-learn.org/stable/modules/classes.html#module-sklearn.multioutput).", "1375027": "Youn can choose between 4 different models or 1 model for all targets, you have the MultipleOutputRegressor wrapper on sklearn or 4 units on final layer if you are using NN", "1375215": "Lightgbm and catboost can easily be used to run 4 models. Several shared kernels showing how to do lightgbm - not sure I have seen a catboost. You can use the same feature set for all 4 models. If your getting a decent amount up the leaderboard than a little bit of coding can get you using different feature sets and light parameters.\n\nThis is a very ugly data set to put togeather and make sense. I would recommend using lightgbm and spend most of your time getting features built.\n\nThe problem - the submission process is a real bitch - so start slow and make sure you can build a basic light that does 4 models and is successful in prediction. Add a FEW features at a time after that -", "1375507": "Hi @junyeongahnn,\n\nI believe you mean that you want to predict a single dependent variable by multiple independent variables, if that what you mean, so you can work a regression model, easy. but if you want to predict all four variables, so I may recommend doing so by using four different models and that will require setting each of your variables as a target variable once. That if you can be so sure about a circular dependence among your variables (every and each variable dependent on others) which's rare to happen to my little experience." }, "source": "meta" }