BOLD: transformation lambda function about neurolib HOT 5 CLOSED

That sounds like a good idea. I guess, for many models that have been around for a while, people have already used transfer functions to produce firing rate output, which could then simply be used. This would also match the behaviour of the model in neurolib with the behaviour of the same model outside of neurolib. For example, someone reads papers about model A (for which transfer function f_A is traditionally being used) and wants to use it in neurolib. However, they are not aware of the different output scaling in neurolib right now. This could cause confusion and even wrong conclusions. The proposed change would prevent that.

from neurolib.

+1 for the user-defined lambda functions. for identity (i.e. no transform), it can be just lambda x: x

from neurolib.

People have already used transfer functions to produce firing rate output ... However, they are not aware of the different output scaling in neurolib right now

Right now, the output itself is not scaled before being "recorded" to the model.output object. This could be done in a very similar way but also done by the model integration itself. aln for example is originally implemented to output kHz but I convert it to Hz in a very hackish way (see Issue #53 ...). I don't quite like it I must admit because the aln model also expects kHz input. So once you start transforming outputs, I think you might also have to start transforming inputs...

from neurolib.

Thanks for the input from both of you. I will make a PR on this, stay tuned!

from neurolib.

Closed by #59

from neurolib.