search

Waikato / wekaDeeplearning4j

Weka package for the Deeplearning4j java library
https://deeplearning.cms.waikato.ac.nz/
GNU General Public License v3.0
185 stars 202 forks source link

NaN default parameters #17

Closed felipebravom closed 6 years ago

felipebravom commented 6 years ago

Hi, I'm finding the default values of many parameters as NaN (e.g., Convolution Layer, NeuralNet Configuration). The FileIterationListener is raising an exception.

Cheers, Felipe

braun-steven commented 6 years ago

The actual default parameters are not NaN but set as default parameters used in the dl4j backend. I just need to update the descriptions. Many of those parameters will vanish with the next version of dl4j anyway and it will be more clear which parameters are available for which object.

The FileIterationListener is raising an exception.

Could you provide me with a log of this exception?

felipebravom commented 6 years ago

Thanks. I'm getting a Class load failed message in the GUI and the following stack trace: I'm getting a java.lang.InstantiationException: weka.dl4j.listener.FileIterationListener java.lang.Class.newInstance(Class.java:427) weka.gui.GenericObjectEditor.classSelected(GenericObjectEditor.java:1865) weka.gui.GenericObjectEditor$3.valueChanged(GenericObjectEditor.java:1743) javax.swing.JTree.fireValueChanged(JTree.java:2927) javax.swing.JTree$TreeSelectionRedirector.valueChanged(JTree.java:3391)

braun-steven commented 6 years ago

The actual default parameters are not NaN but set as default parameters used in the dl4j backend.

I have to correct that: the backend is using the NaN value as a flag to indicate that this property should not be used. As soon as it is set to some value which is not NaN, the parameter will be used.