Understanding predict_mu

[amw5g]

I'm trying to replicate the predict_mu function for a LinearGam outside of python. Specifically, I have the values from coef_ and a corresponding vector of candidate input values. I would expect to multiply the two, sum them, and get the predicted value. But something's going wrong. My coef_ looks like this (adjusted so that each row represents a feature, and multiple values in a row represent spline coefs):

	Col1	Col2	Col3	Col4	Col5	Col6	Col7	Col8	Col9	Col10	Col11	Col12	Col13	Col14	Col15
0	0.663752	0.208064	0.101053	0.405164	0.521302	0.38334	0.145932	0.527738	0.374967	-0.48859	-0.30235	0.16712	-0.05632	-0.07733	0.164027
1	0.42112	0.596672	0.772241	0.947837
2	-0.10411	0.074421	0.778614	1.98895
3	0.63753	0.672153	0.702262	0.725925
4	0.356707	0.465944	0.74609	1.169129
5	0.321534
6	-0.55888
7	0.51603	0.628271	0.740579	0.852991
8	0.01015
9	0.835706	0.677995	0.602635	0.621535
10	0.535754	0.634896	0.734039	0.833181
11	0.768789	0.729764	0.664955	0.574363
12	0.716694	0.695241	0.673739	0.65219
13	1.030868	0.799866	0.568991	0.338147
14	0.681974	0.681629	0.684924	0.689344
15	0.593	0.651409	0.713672	0.779789
16	0.625485	0.664807	0.704128	0.74345
17	0.243205	0.529593	0.823353	1.14172
intercept	2.73787

And my vector of inputs looks like this:

	input
0	52
1	161
2	109856
3	6.058527
4	127
5	1
6	0
7	222.5667
8	12.42475
9	233
10	0
11	0
12	1
13	2
14	1
15	1
16	0
17	2
intercept	1

Multiplying the input by each coefficient, and summing yields 302988. predict_mu yields 12.97.

What step am I missing?

[shyamcody]

@amw5g Look into the `pygam._linear_predictor' function. You are not creating the model matrix as par the code. After creating the model matrix, according to b-spline or whatever basis you have used to create the coefficients, you have to dot multiply the coefficients with the model matrix.