Here I used matlab to compute the nonlinear regressions for the linear, polynomial and
gaussian kernels. The output below shows the sum of squared differences between the
modelled value and the expected value, so it should give a relative indication of fit.
The source for this problem, is provided below:
[p1b.m] [Gkernel.m]
ans =
2.9167 ( Gaussian )
3.3376 ( Linear )
3.2534 ( Polynomial )
Didn't have a great deal of success here; I need to reaccess my method after I get
some sleep. The source is a combination of python and matlab; it is included below along
with the data set.
[ps3.tgz]
Poster | Expected | Linear Kernel | Polynomial Kernel | Gaussian Kernel |
---|---|---|---|---|
American | (0.479) |
(0.680) |
(0.022) |
|
Spanish | Spanish (0.245) |
Spanish (0.077) |
Spanish (0.077) |
|
Spanish | Spanish (1.506) |
Spanish (2.274) |
Spanish (0.007) |
|
Spanish | Spanish (1.506) |
Spanish (0.884) |
Spanish (0.128) |
|
American | American (-0.506) |
(0.011) |
American (-0.049) |
|
Spanish | Spanish (1.040) |
Spanish (1.569) |
Spanish (0.001) |
|