dependent attribute
Hello,
Is it possible to display an attribute, of which I know the value, as a function of other attributes?
This means that for each time series a separate function should be created that describes the attribute.
Thanks in advance
Simon
Is it possible to display an attribute, of which I know the value, as a function of other attributes?
This means that for each time series a separate function should be created that describes the attribute.
Thanks in advance
Simon
Find more posts tagged with
Sort by:
1 - 9 of
91
Hi @ceaperez
I have already looked at the Generate Attributes operator, but unfortunately have not found a function that could solve my problem.
Regards
Hi @ceaperez
I have 5 attributes and want to represent attribute 1 as a function of attributes 2-5. A separate function is to be formed in each data series.
Example:
A1 A2 A3 A4 A5 function (can be linear, polynomial, ...)
-----------------------------------------------------------------------------------
10 20 30 40 50 x1*20+x2*30+x3*40+x4*50=10
15 22 39 43 51 y1*22+y2*39+y3*43+y4*51=15
... ... ... ... ... ...
Regards
What you are asking for is essentially what regression modeling does. You have a numerical label and you want to calculate it as a function of other input attributes. Those functions can be linear (in linear regression), polynomial (in polynomial regression), or non-linear (complex functions like neural nets). So try setting the attribute you want to predict/calculate as the label and then you can use various regression modeling operators to see how it will work. You can view the linear regression operator tutorial if you need help setting this up.
I do not want to predict attribute 1, but only represent it by the other attributes (2-5).
Regards
@SimonK I understand, but don't get hung up on the vocabulary here. In this context "predict" is simply a word that means "represent as a functional combination of other attributes" even when it is already known. If your target attribute is actually a pure functional output of the other inputs, then your "model" will achieve 100% accuracy. If it does not, then you may not have a set of relationships that can be expressed with 100% consistency. Without seeing the dataset it is impossible to be more specific.
With your tips, I get a general function that is calculated from my entire data set.
However, I would be interested to know whether it is possible to generate a separate function in each time series.
Regards
Yes, of course, you simply need to filter for each series (whatever that means for your dataset, I guess it could be a single row or a group of related rows) and run the analysis for each one separately and store each resulting model using the Store attribute.
If you have a large number of series, this could be further automated with several of the Loop attributes, you might want to check those out and perhaps some of the tutorials regarding process control in RapidMiner.
If you have a large number of series, this could be further automated with several of the Loop attributes, you might want to check those out and perhaps some of the tutorials regarding process control in RapidMiner.
Sort by:
1 - 2 of
21
@SimonK I understand, but don't get hung up on the vocabulary here. In this context "predict" is simply a word that means "represent as a functional combination of other attributes" even when it is already known. If your target attribute is actually a pure functional output of the other inputs, then your "model" will achieve 100% accuracy. If it does not, then you may not have a set of relationships that can be expressed with 100% consistency. Without seeing the dataset it is impossible to be more specific.
Yes, of course, you simply need to filter for each series (whatever that means for your dataset, I guess it could be a single row or a group of related rows) and run the analysis for each one separately and store each resulting model using the Store attribute.
If you have a large number of series, this could be further automated with several of the Loop attributes, you might want to check those out and perhaps some of the tutorials regarding process control in RapidMiner.
If you have a large number of series, this could be further automated with several of the Loop attributes, you might want to check those out and perhaps some of the tutorials regarding process control in RapidMiner.
With the Generate Attributes operator you can create new attributes using several functions and other attributes from your dataset.
Regards