Sigmoid.__call__ does not take lambda and gamma

[pberkes]

At the moment, Sigmoid.__call__ does not take the parameters lambda and gamma. It is odd, because the other public methods, Sigmoid.slope and Sigmoid.inverse to take those parameters into account.

The smell comes out in psignifit.tools.psychometric, which does take those parameters, and where the values are computed by calling Sigmoid, then rescaling:

    sigmoid = sigmoid_by_name(sigmoid_name)
    sigmoid_values = sigmoid(stimulus_level, threshold=threshold, width=width)
    psi = gamma + (1.0 - lambda_ - gamma) * sigmoid_values

I think this simple operation should be performed in Sigmoid.__call__

[pberkes]

Also worth noting that Sigmoid.__call__ does mention lambda and gamma in the docstring, but they are not used

[otizonaizit]

The other use of Sigmoid.__call__ that I could trace is here:

python-psignifit/psignifit/_posterior.py

Line 162 in f198fac

psi = sigmoid(level, thres, width) * scale + gamma

where scale is defined like this:

python-psignifit/psignifit/_posterior.py

Line 149 in f198fac

scale = 1 - gamma - lambd

so, yes, that seems to be a relic of the past and we should move lambda and gamma to the Sigmoid.__call__ method. It would be important to trace where we call Sigmoid objects though, as I wouldn't trust that we are testing all code paths where this may be the case. Not sure how to go about it.

And not that we are at it, maybe get rid of __call__ altogether and transform it into something more explicit like values() or so?
It probably looked like a good idea to use __call__ when the Sigmoid class was first developed, but I now think that saving a couple of characters when getting the values of the sigmoid is not worth the opacity of the intent of the call.

[otizonaizit]

by the way, this plan is exactly the opposite of what @pberkes proposed in #140 ;)

[pberkes]

Haha you got me. The point is that we need to do something about it.

In the end, it's related to #102: how is the user supposed to interact with a sigmoid, when they need to? it's a bit of a patchwork now

[pberkes]

Need to check psigniplot, as well:

            y = sigmoid(sigmoid_x, this_sigmoid_params['threshold'], this_sigmoid_params['width'])
            y = (1 - estimate['gamma'] - this_sigmoid_params['lambda']) * y + estimate['gamma']