以变​​量为约束条件的约束回归Python

Question

I'm attempting to run a constrained regression in Python, using the sm.GLM model and then the model.fit_constrained code.

I am feeding in two variables alongside two dummy variables, the dummies are what I am trying to constrain. I want the two dummy variables coefficients multiplied by a weight to equal zero.

This works fine when I am multiply the coefficients by integer weights, as below

results = model.fit_constrained('BOATS * 1 + CARS * 0.5')

However, I want these integers to be variable, and depend on the proportion of my data with a 1 for each dummy variable. I have calculated the proportions in the series SectorWgt, but cannot work out how to then feed it in to the model.fit_constrained code.

This has been my best attempt

results = model.fit_constrained('SIZE*int(SectorWgt.iloc[0])+VQMadj*int(SectorWgt.iloc[1])')

But then I get the error

patsy.PatsyError: unrecognized token in constraint

due to the

int(SectorWgt.iloc[0])

part of code.

Does anyone have any thoughts? Thanks!

Answer 1

Use string formatting:

x = int(SectorWgt.iloc[0])
y = int(SectorWgt.iloc[1])

results = model.fit_constrained('SIZE*{}+VQMadj*{}'.format(x, y))

Answer 2

If using Python 3.6 or newer, you can take advantage of cleaner string interpolation syntax with Python's f-strings .

constraint_str = f"SIZE*{int(SectorWgt.iloc[0])}+VQMadj*{int(SectorWgt.iloc[1])}"
results = model.fit_constrained(constraint_str)