def variational_classifier_modular(kl_inference,train_fname=traindat,test_fname=testdat, label_fname=label_binary_traindat,kernel_log_sigma=0,kernel_log_scale=0,noise_factor=1e-5, min_coeff_kernel=1e-2,max_attempt=0): from math import exp features_train=RealFeatures(CSVFile(train_fname)) labels_train=BinaryLabels(CSVFile(label_fname)) likelihood=LogitDVGLikelihood() error_eval=ErrorRateMeasure() mean_func=ConstMean() kernel_sigma=2*exp(2*kernel_log_sigma); kernel_func=GaussianKernel(10, kernel_sigma) inf=kl_inference(kernel_func, features_train, mean_func, labels_train, likelihood) try: inf.set_noise_factor(noise_factor) inf.set_min_coeff_kernel(min_coeff_kernel) inf.set_max_attempt(max_attempt) except: pass inf.set_scale(exp(kernel_log_scale)) gp=GaussianProcessClassification(inf) gp.train() pred_labels_train=gp.apply_binary(features_train) error_train=error_eval.evaluate(pred_labels_train, labels_train) #print "\nInference name:%s"%inf.get_name(), #print "marginal likelihood:%.10f"%inf.get_negative_log_marginal_likelihood(), #print "Training error %.4f"%error_train return pred_labels_train, gp, pred_labels_train.get_labels()
def variational_classifier_modular(kl_inference,
train_fname=traindat,
test_fname=testdat,
label_fname=label_binary_traindat,
kernel_log_sigma=0,
kernel_log_scale=0,
noise_factor=1e-5,
min_coeff_kernel=1e-2,
max_attempt=0):
from math import exp
features_train = RealFeatures(CSVFile(train_fname))
labels_train = BinaryLabels(CSVFile(label_fname))
likelihood = LogitDVGLikelihood()
error_eval = ErrorRateMeasure()
mean_func = ConstMean()
kernel_sigma = 2 * exp(2 * kernel_log_sigma)
kernel_func = GaussianKernel(10, kernel_sigma)
inf = kl_inference(kernel_func, features_train, mean_func, labels_train,
likelihood)
try:
inf.set_noise_factor(noise_factor)
inf.set_min_coeff_kernel(min_coeff_kernel)
inf.set_max_attempt(max_attempt)
except:
pass
inf.set_scale(exp(kernel_log_scale))
gp = GaussianProcessClassification(inf)
gp.train()
pred_labels_train = gp.apply_binary(features_train)
error_train = error_eval.evaluate(pred_labels_train, labels_train)
#print "\nInference name:%s"%inf.get_name(),
#print "marginal likelihood:%.10f"%inf.get_negative_log_marginal_likelihood(),
#print "Training error %.4f"%error_train
return pred_labels_train, gp, pred_labels_train.get_labels()
