Looking for multiple label regression

Hi, everybody.
I am looking for a way for "multiple label regression"
For example, there are two labels,
concentration(real attribute), type(nominal attribute)
and several independent attributes
So I want to predict concentration and type, given independent attributes
using Neural Network or Deep Learning or Regression or else.
What is the best strategy doing this using RapidMiner?
(But, I found RapidMiner cannot assign two or more labels to example sets)
I am a beginner of RapidMiner, so please help me with this difficult problem.
Thank you and have a nice weekend.
Best Answers
-
Correct, you can only have one label at a time in RapidMiner. So you will need to build two models, one for each of the labels separately, if you are going to do it using the native RapidMiner operators.
You just set the role of label to concentration and build that model. Then you set the role of label to type and build another model. You could do this sequentially in one process or split your analysis into two separate processes. At the end you should have two different predictions/scores for each example and you can merge them together and combine them in whatever way you want.
1 -
Hi @morphism!
In RapidMiner jargon, a label is the same thing as a target. Machine Learning classifiers can handle a single target variable. You can solve your issue in many ways:- Each label is calculated independently with its own algorithm.
- You can use an ensemble to intertwine the calculations.
- Use Multi-Target Regression. RapidMiner does not support MTR at the time, but you can easily implement the MTR algorithms from scikit-learn in Python and do the data preparation with RapidMiner.
All the best,
Rodrigo.2
Answers
-
Correct, you can only have one label at a time in RapidMiner. So you will need to build two models, one for each of the labels separately, if you are going to do it using the native RapidMiner operators.
You just set the role of label to concentration and build that model. Then you set the role of label to type and build another model. You could do this sequentially in one process or split your analysis into two separate processes. At the end you should have two different predictions/scores for each example and you can merge them together and combine them in whatever way you want.
1 -
Hello, Telcontar120.
Thank you for you comment.
I don't have experience with two labels setting, so I cannot figure what that means at this moment.
But I will follow your guidance.
Thank you and Have a nice weekend, Telcontar120.
See you later.
0 -
There is an operator called "Set Role" that you can use to change the role. So you simply use that operator and make your first target variable your label (e.g., concentration), and then use RapidMiner to build a predictive model for it. Then you use Set Role again, change the label to the other target variable (type), and then build another model. Then you have two predictive models, one for each of the attributes you want to predict.1
-
Thank you.~
0 -
Hi @morphism!
In RapidMiner jargon, a label is the same thing as a target. Machine Learning classifiers can handle a single target variable. You can solve your issue in many ways:- Each label is calculated independently with its own algorithm.
- You can use an ensemble to intertwine the calculations.
- Use Multi-Target Regression. RapidMiner does not support MTR at the time, but you can easily implement the MTR algorithms from scikit-learn in Python and do the data preparation with RapidMiner.
All the best,
Rodrigo.2 -
Thank you, rfuentealba.
I will send you email if it is hard to build processes as you teach me
See you soon and have a nice day!
0