Due to recent updates, all users are required to create an Altair One account to login to the RapidMiner community. Click the Register button to create your account using the same email that you have previously used to login to the RapidMiner community. This will ensure that any previously created content will be synced to your Altair One account. Once you login, you will be asked to provide a username that identifies you to other Community users. Email us at Community with questions.

"Extracting NN weights for use in other application."

wesselwessel Member Posts: 537 Maven
edited May 2019 in Help
Dear All,

I wish to train a neural-network in RM, and export the weights to another Java program.

RM provides the following text output:
Node 1 (Sigmoid)    Node 2 (Sigmoid)  Output (Linear) 
----------------  ----------------  ------------------- 
att1: -0.198      att1: -1.080      Node 1: 1.210     
att2: 0.399        att2: -0.660      Node 2: -2.484       
att3: 0.736        att3: -0.370      Threshold: 0.615
att4: 0.472        att4: -0.740         
Threshold: -0.439  Threshold: 0.351         

But this is of no use, since these weights are not in full double precision.


When I use the model writer to write an XML file, I think I do get the proper weights,
somewhere in the file, but the question is "where exactly".
For example:
            <inputNodes id="73">
              <com.rapidminer.operator.learner.functions.neuralnet.InnerNode id="74">
                <weights id="75">
                  <double>-0.4394771578377415</double>
                  <double>-0.19801951972474785</double>
                  <double>0.3992843094056797</double>
                  <double>0.7363213055568706</double>
                  <double>0.4723782999051438</double>

                      <com.rapidminer.operator.learner.functions.neuralnet.InnerNode reference="74"/>
                      <com.rapidminer.operator.learner.functions.neuralnet.InnerNode id="91">
                        <weights id="92">
                          <double>0.3506573392426525</double>
                          <double>-1.0796773279531984</double>
                          <double>-0.6600984419916611</double>
                          <double>-0.36967194141103804</double>
                          <double>-0.7402183227344546</double>
                        </weights>
                        <weightChanges

These seem to correspond to the weights of hidden node 2.
But this XML file contains a lot more then just weights.

Best regards,

Wessel
Tagged:

Answers

  • landland RapidMiner Certified Analyst, RapidMiner Certified Expert, Member Posts: 2,531 Unicorn
    Hi Wessel,
    the easiest and somehow quirky way would be to just increase the number of digits RapidMiner presents to the user. This is possible in the properties menu. It holds true for almost all places in RapidMiner and hence it should work with the Neural Net, too.

    Greetings,
      Sebastian
Sign In or Register to comment.