dasvm_libsvm = DomainAdaptationSVM(1.0, wdk, lab, presvm_libsvm, B) dasvm_libsvm.set_bias_enabled(False) dasvm_libsvm.train() dasvm_liblinear = DomainAdaptationSVMLinear(1.0, feat, lab, presvm_liblinear, B) dasvm_liblinear.io.set_loglevel(MSG_DEBUG) dasvm_liblinear.set_bias_enabled(False) dasvm_liblinear.train() print "##############" alphas = [] sv_ids = dasvm_libsvm.get_support_vectors() for (j, sv_id) in enumerate(sv_ids): alpha = dasvm_libsvm.get_alphas()[j] #get rid of label alpha = alpha * labels[sv_id] alphas.append(alpha) print "alphas libsvm", alphas #w = presvm_liblinear.get_w() #print "prew", w[0:10] #print "labels", labels ############################################# # checking linear term ############################################# #linterm_libsvm = dasvm_libsvm.get_linear_term_array()
dasvm_libsvm.set_bias_enabled(False) dasvm_libsvm.train() dasvm_liblinear = DomainAdaptationSVMLinear(1.0, feat, lab, presvm_liblinear, B) dasvm_liblinear.io.set_loglevel(MSG_DEBUG) dasvm_liblinear.set_bias_enabled(False) dasvm_liblinear.train() print "##############" alphas = [] sv_ids = dasvm_libsvm.get_support_vectors() for (j, sv_id) in enumerate(sv_ids): alpha = dasvm_libsvm.get_alphas()[j] #get rid of label alpha = alpha*labels[sv_id] alphas.append(alpha) print "alphas libsvm", alphas #w = presvm_liblinear.get_w() #print "prew", w[0:10] #print "labels", labels ############################################# # checking linear term #############################################
Math_init_random(1) dasvm.train() #dasvm = SVMLight(1.0, wdk, lab) #dasvm.set_linear_term(numpy.double(p)) #dasvm.train() lin_da = dasvm.get_linear_term() daobj = dasvm.get_objective() sv_idx_da = dasvm.get_support_vectors() alphas_da = dasvm.get_alphas() alphas_full_da = numpy.zeros(N) alphas_full_da[sv_idx_da] = alphas_da ################ #checking linear term presvm.set_bias(0.0) tmp_out = -B*presvm.classify(feat).get_labels()*tmp_lab - 1 for i in xrange(len(examples)): print lin_da[i], tmp_out[i] assert(abs(lin_da[i]-tmp_out[i])<= 0.001)
dasvm = DomainAdaptationSVM(1.0, wdk, lab, presvm, B) #dasvm = SVMLight(1.0, wdk, lab) Math_init_random(1) dasvm.train() #dasvm = SVMLight(1.0, wdk, lab) #dasvm.set_linear_term(numpy.double(p)) #dasvm.train() lin_da = dasvm.get_linear_term() daobj = dasvm.get_objective() sv_idx_da = dasvm.get_support_vectors() alphas_da = dasvm.get_alphas() alphas_full_da = numpy.zeros(N) alphas_full_da[sv_idx_da] = alphas_da ################ #checking linear term presvm.set_bias(0.0) tmp_out = -B * presvm.classify(feat).get_labels() * tmp_lab - 1 for i in xrange(len(examples)): print lin_da[i], tmp_out[i] assert (abs(lin_da[i] - tmp_out[i]) <= 0.001) for i in xrange(len(lin)):