forum.alglib.net

Hello,
I am a beginner in c # develop, I just want to run the solution alglib (multinomial logistic regression). is there anyone who can help me to run this example (the initial matrix attached). I want how to enter this matrix to the Main console application, and show output the program.
thank you for your help

I recommend you to read http://www.alglib.net/dataanalysis/gene ... ciples.php

It described different encoding techniques that you can use. You can encode X1-X4 "as is", and use 1-of-N to the last column.

I have read and understand the documentation, but my problem is:
- declaration of the matrix-based learning ...
- function that allows me to run the logistic regression multinomial.
can you give me the Main () which allows running this prog based on the matrix attached.

Thank you for your cooperation

Unfortunately, there is no examples for logit yet. However, I don't think that it should be complex problem.

First, you encode your problem (first line may look like {45, 250, 1300, 600, 0, 0, 1}).
Then you create logit model
Then you fit it
Then you use it, extract parameters, etc.

you can describe me the column headings (0,0,1 ...) of the first line

you can describe me the column headings (... 0,0,1) of the first line

Sorry, I've misinformed you. It will be {45, 250, 1300, 600, 2}.

First four values are X1-X4, last value is a class number (0-2).

Thank you,

can you help me to run the same example with Lineaire Discriminant Analysis!!!

Sample format will be the same. Just apply LDA, and you will get set of orthogonal vectors, which correspond to the most informational axes.

However, I don't think that LDA will be good for your problem. You have multiclass problem, so LDA won't give you linear classifier. It will just give you subspace which captures most of the information (variance) contained in the sample. But it will be subspace with dimensionality equal to 3, so you just will get 3-dimensional problem instead of 4-dimensional one. You can use LDA as preprocessor before applying another, more powerful algorithm. But I don't think that such preprocessing is necessary in your case (space dimensionality is already low).

is there an algorithm Factorial Discrminant Analysis in AlgLib???