TsodyksMarkramSynapse broken

[neworderofjamie]

Like with the GIF neuron, whether the various functions of this model are activated is indicated by setting their time constants to zero. However, currently the model uses a static code string which results in divide by zeros. The same dynamic code generation approach used on the GIF neuron needs to be applied here (will also make, for example, purely depressing synapses much more efficient)

[alishdipani]

@neworderofjamie could you provide an example for this issue?

[neworderofjamie]

You can reproduce this issue using the official example (modified to disable the recording of "gsyn_inh" because of #16) which results in a figure like:

It would be good to verify the correctness of the facilitating case by installing NEST and overlaying the plots, but the "depressing" case is definitely broken. The reason for this is that the facilitation mechanism gets turned off by tau_facil=0 which causes a divide by zero here. The solution is to employ a simpler version of the strategy used by the GIF neuron to customize the model based on its parameters.

[alishdipani]

Thank you for this. I'll get back with the results for verification soon.