Grasshopper

algorithmic modeling for Rhino

Hey first of all I want to thank you for this really exiting plug-in.

My main Problem is that the output i get from the back-propagation solver and the resulting trained net is different than expected and I am not able to find the reason why ... I hope you can help me. 

The basic idea is quite simple. I have a complex fabrication process and a quite complex and computation intensive way of testing if a peace is possible to fabricate. I generated 5000 training vectors for in and out put in[length,Angle1,Angle2] and out[bool-Possilble]. 


I trained the  net over 50 000 Cycles using a learning rate from 0.01 . 


My basic problem is that the vector classifier outputs the same output_vector for any input vector I choose event-hough if it is identical to the training Vectors ... any idea why ?

I uploaded the script as well ... :) 
Thank you for your help. 

P.S. I think a method showing the progress of the training while cycling would be very helpful. I know that this is difficult because looping components freeze the Grasshopper Canvas here I upload a easy python component which uses the Rhino Command line to output progress while cycling maybe this would be a possible solution ... 

Kind Regards 

Tobi

Views: 168

Attachments:

Replies to This Discussion

Hi all,

I have a very similar problem.
Keep getting the same output vector from the trained network when I change the input vector.

Any suggestions what could be the problem here?

Cheers,

Lukas

RSS

About

Translate

Search

Photos

  • Add Photos
  • View All

© 2021   Created by Scott Davidson.   Powered by

Badges  |  Report an Issue  |  Terms of Service