search

gkaguirrelab / temporalFittingEngine

A core computational engine, and associated models, for non-linear fitting of parametric models to time-series data
0 stars 0 forks source link

'errorType' key/value pair to fitResponse not used fully #74

Closed DavidBrainard closed 5 years ago

DavidBrainard commented 5 years ago

Someone added an errorType key/value pair to the fitResponse method but did not comment it.

This parameter is passed to the fitError method of the subclass when computing the final error, but does not appear to be passed during the actual fitting. This seems like a bug, or at least worth a careful comment.

I commented the key/value pair (along with others that had been added without updates to the header comment), and put a note about the fact that this does not seem to be fully used. Someone who uses this option might take a look.

gkaguirre commented 5 years ago

This issue intersects with the larger topic of how we wish to handle parameters within the tfe.

For now, I have modified fitResponse so that varargin is passed to the objective function of fmincon. We will want to switch to an alternative approach in which the object will store these properties instead of engaging in parameter passing.

DavidBrainard commented 5 years ago

Sounds good. Just an additional thought. There will be some key/value pairs that probably should be passed to methods. For example, for fitResponse we will probably want to preserve the current ability to override the search starting parameters (which are those returned by defaultParams method, with something that varies according to a particular problem. This I can see being varied after an object is created. There may be other examples like this.