Math parsing of fit function

[ckoerber]

The function document method uses sympy to generate a latex expression for the fit function and returns None if it fails to generate the expression.

See also from lsqfitgui.util.function import parse_function_expression.

Unfortunately, parsing functions using numpy fail. This is due to

from sympy import S
import numpy as np

x = S("x")
np.exp(x)

fails.

A workaround is overloading the Symbol class

from sympy import Symbol, exp
import numpy as np

class MySymbol(Symbol):

   def exp(self):
       return exp(self)

x = MySymbol("x")
np.exp(x)

which may also work for other functions (maybe there is a more elegant way by looking up matching module attributes).

The main issue, I have however encountered is that, for example

x = MySymbol("x")
y = MySymbol("y")

np.exp(x + y)

fails again because this is now not a MySymbol but rather an Add class.

So, I believe, if it is somehow possible, it would be more effective to replace all the numpy expressions by sympy expressions within the function call. Maybe this possible in the spirit of unittest.mock.

[millernb]

A small suggestion: if lsqfit-gui is unable to properly parse the fitfcn as a latex expression, maybe fall-back to (optionally?) printing the python block. I know that, at least for me, my fit functions usually contain plenty of if/else statements (especially for xpt fits), which I wouldn't expect to be parsed correctly (though I would be tremendously impressed if they were).

[ckoerber]

This is a good idea, Though it can be a bit nested if it is also an entire class. I would probably make it a collapsible section (or comparable) to not pollute the GUI too much.

Generally speaking, if the if statements don't use the x or p values, it should be possible. It just uses sympy expressions instead of actual numbers or gvars.

[ckoerber]

The source code inspection is now implemented on the function-source-code branch. I'll wait with the merge because of related prior export dependencies.