Hi, I'm making a NN that has to be trained in an environment in which the correct outcome is not known before the next iteration (software agent environment).
In all documentation I find that the teacher component can be linked to the output layer. But is it also possible to store the output of the network and when the next iteration starts feed that through the teacher (and change the weights of the network accordingly) before I train the network for another iteration?
Basicly I want the teacher to tell the network how good or bad the network acted the last iteration, not the current.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
No, I'm sorry, it isn't possible, but it could be easily implemented by extending the TeachingSynapse component. I'll insert it in the feature requests list.
Paolo
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi, I'm making a NN that has to be trained in an environment in which the correct outcome is not known before the next iteration (software agent environment).
In all documentation I find that the teacher component can be linked to the output layer. But is it also possible to store the output of the network and when the next iteration starts feed that through the teacher (and change the weights of the network accordingly) before I train the network for another iteration?
Basicly I want the teacher to tell the network how good or bad the network acted the last iteration, not the current.
No, I'm sorry, it isn't possible, but it could be easily implemented by extending the TeachingSynapse component. I'll insert it in the feature requests list.
Paolo