Options

# Using (bayesian network) models?

Hello there,

First of all, I am quite new to RapidMiner so I apologise for any nomenclature mistakes.

Until now I have made very simple analyses with rapidminer. My modelling diagrams consisted of:

Retrieve -> select attributes -> decision tree

or

Retrieve -> select attributes -> correlation matrix

The data set has a mix of reals, nominals, etc. however binning is an option.

Now, I would like to use bayesian network models, and preferably in an interactive way like this:

1) Make a conditional dependency map (untargeted learning)

2) Make a targeted conditional dependency (so define a label variable)

- Use the result of 2) and do the following:

a) Look at the conditional probability distributions if nothing is known of the input

b) Fix one of the input variables to a certain bin, and look how the conditional probability distributions change

Now my questions are:

1) Is this possible, and if so, what approach should I take for 1 and for 2? If not possible, what alternative ways can I use

to achieve the same?

a) How would my modelling diagram look like (and which building blocks should I use)?

b) How can I "apply model", do I have to load an excel file each time to change the input variables when applying the model, or is there

a built-in interface for this sort of use (so I can play very rapidly with the input variables)?

If these questions are not worded clearly I will try to explain better. Thank you in advance for taking the time to read it!

p.s. Originally I posted this in the getting started forum but I think this place is more appropriate so

I removed the other post and put it here, I hope that this is the correct place (and if not, my apologies).

First of all, I am quite new to RapidMiner so I apologise for any nomenclature mistakes.

Until now I have made very simple analyses with rapidminer. My modelling diagrams consisted of:

Retrieve -> select attributes -> decision tree

or

Retrieve -> select attributes -> correlation matrix

The data set has a mix of reals, nominals, etc. however binning is an option.

Now, I would like to use bayesian network models, and preferably in an interactive way like this:

1) Make a conditional dependency map (untargeted learning)

2) Make a targeted conditional dependency (so define a label variable)

- Use the result of 2) and do the following:

a) Look at the conditional probability distributions if nothing is known of the input

b) Fix one of the input variables to a certain bin, and look how the conditional probability distributions change

Now my questions are:

1) Is this possible, and if so, what approach should I take for 1 and for 2? If not possible, what alternative ways can I use

to achieve the same?

a) How would my modelling diagram look like (and which building blocks should I use)?

b) How can I "apply model", do I have to load an excel file each time to change the input variables when applying the model, or is there

a built-in interface for this sort of use (so I can play very rapidly with the input variables)?

If these questions are not worded clearly I will try to explain better. Thank you in advance for taking the time to read it!

p.s. Originally I posted this in the getting started forum but I think this place is more appropriate so

I removed the other post and put it here, I hope that this is the correct place (and if not, my apologies).

0