search

OpenRoberta / openroberta-lab

The programming environment »Open Roberta Lab« by Fraunhofer IAIS enables children and adolescents to program robots. A variety of different programming blocks are provided to program motors and sensors of the robot. Open Roberta Lab uses an approach of graphical programming so that beginners can seamlessly start coding. As a cloud-based application, the platform can be used without prior installation of specific software but runs in any popular browser, independent of operating system and device.
Apache License 2.0
124 stars 118 forks source link

xNN - activation functions #1591

Closed Thorlogy closed 5 months ago

Thorlogy commented 7 months ago

Describe the bug It see

To Reproduce Steps to reproduce the behavior:

  1. Go to 'Open Roberta xNN'
  2. Click on 'Neuronal Network Define'
  3. Click on activation
  4. Click on ReLU, Tanh, Sigmoid
  5. Input trainings data
  6. Run forward propagation layer by layer
  7. See error

Expected behavior activation

Screenshots

xNN_activiation

Device information

rbudde commented 7 months ago

the activation function was remembered in the state of the network, but not in the current instance and thus not applied after being changed. Thus: it was applied to the output neuron, but not the right one.

heini208 commented 5 months ago

Output is now always updated. However if you switch activation while switching layers the input disappears.