Please help me!!!!

[nycepero]

Hello, in our project we created a new node that implements a decision forest constructor algorithm for the prediction of the model, we use the Weka predictor but the one that is already deprecate because it is the one that is compatible with our node.
In one machine it works perfectly and in another it gives us this error …
“ERROR Weka Predictor (deprecated) 0:17 Could not create port object spec serializer for ‘org.knime.ext.weka.ports.WekaClassifierModelPortObjectSpec’ from plug-in ‘node_proactiveforest’: Executable extension definition for “specSerializer” not found.
ERROR PortTypeRegistry CODING PROBLEM Class “WekaClassifierModelPortObjectSpec” does not have a custom PortObjectSpecSerializer, using standard (but slow) Java serialization. Consider implementing a PortObjectSpecSerialzer.
ERROR Weka Predictor (deprecated) 0:17 Unable to clone input data at port 1 (Weka model): No value present”
Please, if you can think of something that can help us, we thank you in advance … this is a very important project and we have no idea what happens.
Thank you

[mlauber71]

First you could read about recent problems with a Weka node here and maybe get some ideas how to fix that:

ERROR Weka Predictor (3.7) Execute failed: java.lang.NullPointerException

I have experienced that the Weka nodes really do need a lot of memory and especially SVM models are very hungry for it. Besides just using a stronger machine you can try these things: test the whole thing with just 100 items, close all other workflows and restart KNIME before you do to see if it would work at all if it does work with smaller chunks you can think about looping over smaller numbers of lines or try streaming where KNIME would process only several lines at a time (encapsulate the …

Then if the workflow runs on one computer but not the other you could

• check the versions of Windows (if that is your OS), KNIME and Weka
• see if the performance settings of KNIME and the resources are identical since Weka nodes tend to use a lot of RAM
• reinstall KNIME and the necessary extensions
• check that your file system is healthy and can handle special characters KNIME likes to use (we have seen problems with some One Drive installations where admins restricted the use of these characters)

And then about the model itself:

• have you tried running it with the new classifier
• how large is the model file? That could very well affect the necessary power
• you could try to revert to an older version of KNIME and see if it works there

And of course if you could share a workflow where that happens or even the model we might get further insights - although I understand that that is often not possible. If it is a systematic bug it would be best to make a small workflow that can reproduce the problem without giving away privileged information.

If that is not possible you could try the following:

• clear you KNIME log
• restart KNIME
• just open the workflow in question and produce the error
• then post the LOG file here

We might have a further look at the problem.

If all of that does not help we might try further things like bringing in an R package that might be able to make use of a Weka model like “RWeka”. That is not a desired approach, it is not convenient (KNIMEey) but if the model/project is very important and there is no other way it could be possible to integrate that into a KNIME workflow (not promising anything though …).

[gab1one]

Hi @nycepero,
are you able to share the code of your project with us? Your error looks to me like there are some misconfigurations preset in your project setup.

best,
Gabriel