def inner_product(example1, example2): return example1.h_add * example2.h_add + example1.h_max * example2.h_max + example1.h_ff * example2.h_ff def gram_matrix(examples1, examples2): gram = zeros((len(examples1), len(examples2))) for i in range(len(examples1)): for j in range(len(examples2)): gram[i, j] = inner_product(examples1[i], examples2[j]) return gram print examples[1].h_ff from sklearn.svm import SVR, NuSVR model = SVR(C=1000, epsilon=0.1, kernel="precomputed") # model = SVR(kernel=kernel) print gram_matrix(examples, examples) model.fit(gram_matrix(examples, examples), array([example.cost for example in examples])) print array([example.cost for example in examples]) print model.predict(gram_matrix(examples[:1], examples)) from mlpy import KernelRidge model = KernelRidge(lmb=0.01) model.learn(gram_matrix(examples, examples), array([example.cost for example in examples])) print model.pred(gram_matrix(examples[:1], examples))