More advanced network analysis enhancements #204
Description
10. Add possibility to record history of synapse weight value. This is tougher because all the synapse weight info is in the engine
11. Add breakpoints on some event like neuron fires, value reaches some threshold or value change. Interesting
12. Possibility / way to make a module from one neural network, so it can be used as a module in another network. Related to cut/paste Similar, but this willl hide inner structure so higher level structures will be more obvious. Something similar like submodules in Matlab Simulink or LabView .
14. Run for a specific number of cycles. Record all neuron states for each cycle and synapse weights. This will enable post analysis of network behaviour, batch input testing and comparison of various network structures. I'm not expecting to record and post analyse large scale networks. This is just meant to be used to tune some basic structures. And also this will enable me to do time plots of whatever from the network. This could be an extension of the existing Firing History mechanism. Since it is only done on a relatively small number of neurons, the performance impact isn’t too great.
15. Topic for further research and development: Use plain time instead of cycles. This should enable us to simulate more complex timing scenarios. And it is intended to be used only with recording of specific period of time. Then do post analysis or recorded behaviour. This could be easy. User already gets to define the Refractory period. In biology, this is about 4ms so the conversion of cycles to ms is straightforward. If you say the refractory is 0, then cycles are 4ms. If you say it is 1, then cycles are 8ms. Etc.