Export SVM classifiers from sklearn to Java code

Question

I used sklearnto train a set of SVM classifiers (mostly linear using LinearSVM, but some of them use a class SVCwith rbf core), and I am very pleased with the results. Now I need to export classifiers into production to another code base that uses Java. I am looking for possible libraries that are published in maven that can easily be included in this new codebase.

What are you offering?

Answer 1

Linear classifiers are simple: they have coef_and intercept_are described in class docstrings. These are regular NumPy arrays, so you can flush them to disk with standard NumPy functions.

>>> from sklearn.datasets import load_iris
>>> iris = load_iris()
>>> from sklearn.svm import LinearSVC
>>> clf = LinearSVC().fit(iris.data, iris.target)

Now dump this into a pseudo file:

>>> from io import BytesIO
>>> outfile = BytesIO()
>>> np.savetxt(outfile, clf.coef_)
>>> print(outfile.getvalue())
1.842426121444650788e-01 4.512319840786759295e-01 -8.079381916413134190e-01 -4.507115611351246720e-01
5.201335313639676022e-02 -8.941985347763323766e-01 4.052446671573840531e-01 -9.380586070674181709e-01
-8.506908158338851722e-01 -9.867329247779884627e-01 1.380997337625912147e+00 1.865393234038096981e+00

What can you parse from Java, right?

Now, to get an estimate for the k'th class on the sample x, you need to evaluate

np.dot(x, clf.coef_[k]) + clf.intercept_[k]
# ==
(sum(x[i] * clf.coef_[k, i] for i in xrange(clf.coef_.shape[1]))
 + clf.intercept_[k])

which is also doable, hopefully. The class with the highest score wins.

For SVM cells, the situation is more complicated because you need to replicate the one-vs-one solution function , as well as the kernels in Java code. The SVM model is stored on objects SVCin the support_vectors_and attributes dual_coef_.

Answer 2

, SVM- , , - , SVM, () . , , SVM, , , .