def noise_learning_test():
import numpy
from mclp import LPBoostMulticlassClassifier
# do all of the setup implicitly this time
c = LPBoostMulticlassClassifier(3, 0.1)
# ok to use numpy arrays
c.add_multiclass_classifier(numpy.eye(3))
# junk / noise
c.add_multiclass_classifier([[1.0, 0.4, 0.2],
[0.4, 0.4, 0.4],
[0.6, 0.1, 0.2]])
# junk / noise
c.add_multiclass_classifier([[0.3, 0.3, 0.3],
[0.4, 0.4, 0.4],
[0.5, 0.5, 0.5]])
c.update()
assert(c.weights[0][0] - 1.0 < 1e-8)
assert(c.rho - 1.0 < 1e-8)
def simple_learning_test():
import numpy
from mclp import *
c = LPBoostMulticlassClassifier(3, 0.1)
c.initialize_boosting([0,1,2], False, "clp")
c.add_multiclass_classifier([[1.0, 0.0, 0.0],
[0.0, 1.0, 0.0],
[0.0, 0.0, 1.0]])
c.update()
assert(c.weights[0][0] - 1.0 < 1e-8)
assert(c.rho - 1.0 < 1e-8)
fmean = train_features.mean(1)
train_features -= fmean[:, None]
test_features -= fmean[:, None]
fstd = train_features.std(1)
train_features /= fstd[:, None]
test_features /= fstd[:, None]
from mclp import LPBoostMulticlassClassifier
lp = LPBoostMulticlassClassifier(2, 0.1)
# lp.initialize_boosting([0,1], False, "clp")
for tf in train_features:
lp.add_multiclass_classifier([tf - 1, tf])
lp.update()
print "rho:", lp.rho
train_predictions = sp.zeros(train_labels.shape)
for i in xrange(len(train_features)):
# print i, lp.weights[0][i]
train_predictions += lp.weights[0][i] * train_features[i]
print "train perf=", 100.0 * (sp.sign(train_predictions) == train_labels).sum() / train_labels.size
test_predictions = sp.zeros(test_labels.shape)
for i in xrange(len(test_features)):
# print i, lp.weights[0][i]
test_predictions += lp.weights[0][i] * test_features[i]
print "test perf=", 100.0 * (sp.sign(test_predictions) == test_labels).sum() / test_labels.size