Using kernlab
I've trained a model with code like the following:
my.model <- ksvm(result ~ f1+f2+f3, data=gold, kernel="vanilladot")
Since it's a linear model, I prefer at run-time to compute the scores as a simple weighted sum of the feature values rather than using the full SVM machinery. How can I convert the model to something like this (some made-up weights here):
> c(.bias=-2.7, f1=0.35, f2=-0.24, f3=2.31)
.bias f1 f2 f3
-2.70 0.35 -0.24 2.31
where .bias
is the bias term and the rest are feature weights?
EDIT:
Here's some example data.
gold <- structure(list(result = c(-1, -1, -1, -1, -1, -1, -1, -1, -1,
-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1), f1 = c(0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 1, 1, 0, 0,
1, 0, 0, 0, 0, 0, 0, 1, 0, 1, 1, 0, 0, 0, 1), f2 = c(13.4138113499447,
13.2216999857095, 12.964145772169, 13.1975227965938, 13.1031520152764,
13.59351759447, 13.1031520152764, 13.2700658838026, 12.964145772169,
13.1975227965938, 12.964145772169, 13.59351759447, 13.59351759447,
13.0897162110721, 13.364151238365, 12.9483051847806, 12.964145772169,
12.964145772169, 12.964145772169, 12.9483051847806, 13.0937231331592,
13.5362700880482, 13.3654209223623, 13.4356400945176, 13.59351759447,
13.2659406408724, 13.4228886221088, 13.5103065354936, 13.5642812689161,
13.3224757352068, 13.1779418771704, 13.5601730479315, 13.5457299603578,
13.3729010596517, 13.4823595997866, 13.0965264603473, 13.2710281801434,
13.4489887206797, 13.5132372154748, 13.5196188787197), f3 = c(0,
1, 0, 0, 0, 1, 0, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 0, 1, 0,
0, 0, 1, 0, 1, 1, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1, 0, 0)), .Names = c("result",
"f1", "f2", "f3"), class = "data.frame", row.names = c(NA, 40L
))
1) Recall that in linear SVM, the result is a hyperplane that separates the classes as best as possible. The weights represent this hyperplane, by giving you the coordinates of a vector which is orthogonal to the hyperplane - these are the coefficients given by svm. coef_.
The support vector machine is a machine learning algorithm that follows the supervised learning paradigm and can be used for both classifications as well as regression problems though this is primarily a classification algorithm.
To get the bias, just evaluate the model with a feature vector of all zeros. To get the coefficient of the first feature, evaluate the model with a feature vector with a "1" in the first position, and zeros everywhere else - and then subtract the bias, which you already know. I'm afraid I don't know R syntax, but conceptually you want something like this:
bias = my.model.eval([0, 0, 0])
f1 = my.model.eval([1, 0, 0]) - bias
f2 = my.model.eval([0, 1, 0]) - bias
f3 = my.model.eval([0, 0, 1]) - bias
To test that you did it correctly, you can try something like this:
assert(bias + f1 + f2 + f3 == my.model.eval([1, 1, 1]))
If you love us? You can donate to us via Paypal or buy me a coffee so we can maintain and grow! Thank you!
Donate Us With