def test_gridminimise():
from milksets.wine import load
features, labels = load()
x = gridminimise(milk.supervised.svm_simple(kernel=np.dot, C=2.),
features[::2], labels[::2] == 0, {'C': (0.5, )})
cval, = x
assert cval == ('C', .5)
def test_multi_view():
from milksets.wine import load
features, labels = load()
features0 = features[::10]
features1 = features[1::10]
features2 = features[2::10]
labels0 = labels[::10]
labels1 = labels[1::10]
labels2 = labels[2::10]
assert np.all(labels0 == labels1)
assert np.all(labels1 == labels2)
labels = labels0
train_features = zip(features0, features1, features2)
test_features = zip(features[3::10], features[4::10], features[5::10])
base = milk.supervised.classifier.ctransforms(
feature_selection_simple(),
milk.supervised.svm.svm_raw(C=128,
kernel=milk.supervised.svm.rbf_kernel(4.)),
milk.supervised.svm.svm_sigmoidal_correction())
classifier = milk.supervised.multi_view.multi_view_classifier(
[base, base, base])
model = classifier.train(train_features, labels == 0)
assert ([model.apply(f)
for f in test_features] == (labels == 0)).mean() > .9
def test_default_small():
features, labels = load()
selected = np.concatenate( [np.where(labels < 2)[0], np.where(labels == 2)[0][:6]] )
features = features[selected]
labels = labels[selected]
learner = defaultclassifier('fast')
# For version 0.3.8, the line below led to an error
milk.nfoldcrossvalidation(features, labels, classifier=learner)
def test_defaultclassifier():
from milksets import wine
features, labels = wine.load()
C = milk.supervised.defaultclassifier()
model = C.train(features, labels)
labelset = set(labels)
for f in features:
assert model.apply(f) in labelset
def test_defaultclassifier():
from milksets import wine
features, labels = wine.load()
C = milk.supervised.defaultclassifier()
model = C.train(features,labels)
labelset = set(labels)
for f in features:
assert model.apply(f) in labelset
def test_gridminimise_return():
from milksets.wine import load
features,labels = load()
learner = fast_classifier()
gridminimise(learner, features, labels, { 'ignore' : [0] })
_,error = gridminimise(learner, features, labels, { 'ignore' : [0] }, return_value=True, nfolds=5)
cmat,_ = milk.nfoldcrossvalidation(features, labels, learner=learner, nfolds=5)
assert error == cmat.sum()-cmat.trace()
def test_assign_cids():
from milksets.wine import load
features, _ = load()
assigns, centroids = milk.unsupervised.kmeans(features,
3,
R=2,
max_iters=10)
assert np.all(assign_centroids(features, centroids) == assigns)
def test_tree():
from milksets import wine
features, labels = wine.load()
selected = (labels < 2)
features = features[selected]
labels = labels[selected]
C = milk.supervised.tree.tree_classifier()
model = C.train(features,labels)
assert (np.array([model.apply(f) for f in features]) == labels).mean() > .5
def test_with_error():
from milksets.wine import load
features, labels = load()
learner = error_learner()
G = milk.supervised.gridsearch(
error_learner(),
params = { 'error' : range(3), 'error2' : range(5) }
)
G.train(features,labels)
def test_select_n():
from milksets.wine import load
features,labels = load()
for n in (1,2,4,8):
select = select_n_best(n, rank_corr)
model = select.train(features,labels)
f = model.apply(features[3])
assert len(f) == n
def test_nfoldcrossvalidation_simple_list():
from milksets import wine
features, labels = wine.load()
features = features[::2]
labels = labels[::2]
cmat,clabels = nfoldcrossvalidation(list(features), list(labels), classifier=fast_classifier())
assert cmat.shape == (3,3)
assert len(clabels) == 3
def test_tree():
from milksets import wine
features, labels = wine.load()
selected = (labels < 2)
features = features[selected]
labels = labels[selected]
C = milk.supervised.tree.tree_classifier()
model = C.train(features, labels)
assert (np.array([model.apply(f) for f in features]) == labels).mean() > .5
def test_select_n():
from milksets.wine import load
features, labels = load()
for n in (1, 2, 4, 8):
select = select_n_best(n, rank_corr)
model = select.train(features, labels)
f = model.apply(features[3])
assert len(f) == n
def test_rf():
from milksets import wine
features, labels = wine.load()
features = features[labels < 2]
labels = labels[labels < 2]
learner = randomforest.rf_learner()
model = learner.train(features[::5], labels[::5])
test = [model.apply(f) for f in features]
assert np.mean(labels == test) > .7
def test_with_error():
from milksets.wine import load
features, labels = load()
learner = error_learner()
G = milk.supervised.gridsearch(error_learner(),
params={
'error': list(range(3)),
'error2': list(range(5))
})
G.train(features, labels)
def test_non_contiguous_fmatrix():
from milksets.wine import load
features,_ = load()
features = features[:,::2]
assigns, centroids = milk.unsupervised.kmeans(features, 3, R=2, max_iters=10)
assert np.all(assign_centroids(features, centroids) == assigns)
features = features.astype(np.int32)
assigns, centroids = milk.unsupervised.kmeans(features, 3, R=2, max_iters=10)
assert np.all(assign_centroids(features, centroids) == assigns)
def test_learner():
from milksets import wine
learner = milk.supervised.adaboost.boost_learner(milk.supervised.tree.stump_learner())
features, labels = wine.load()
features = features[labels < 2]
labels = labels[labels < 2] == 0
labels = labels.astype(int)
model = learner.train(features[::2], labels[::2])
train_out = np.array(map(model.apply, features))
assert (train_out == labels).mean() > .9
def test_default_small():
features, labels = load()
selected = np.concatenate(
[np.where(labels < 2)[0],
np.where(labels == 2)[0][:6]])
features = features[selected]
labels = labels[selected]
learner = defaultclassifier('fast')
# For version 0.3.8, the line below led to an error
milk.nfoldcrossvalidation(features, labels, classifier=learner)
def test_non_contiguous_fmatrix():
from milksets.wine import load
features,_ = load()
features = features[:,::2]
assigns, centroids = milk.unsupervised.kmeans(features, 3, R=2, max_iters=10)
assert np.all(assign_centroids(features, centroids) == assigns)
features = features.astype(np.int32)
assigns, centroids = milk.unsupervised.kmeans(features, 3, R=2, max_iters=10)
assert np.all(assign_centroids(features, centroids) == assigns)
def test_learner():
from milksets import wine
learner = milk.supervised.adaboost.boost_learner(
milk.supervised.tree.stump_learner())
features, labels = wine.load()
features = features[labels < 2]
labels = labels[labels < 2] == 0
labels = labels.astype(int)
model = learner.train(features[::2], labels[::2])
train_out = np.array(map(model.apply, features))
assert (train_out == labels).mean() > .9
def test_split_subsample():
import random
from milksets import wine
features, labels = wine.load()
seen = set()
for i in xrange(20):
random.seed(2)
i,s = milk.supervised.tree._split(features[::10], labels[::10], None, milk.supervised.tree.information_gain, 2, random)
seen.add(i)
assert len(seen) <= 2
def test_split_subsample():
import random
from milksets import wine
features, labels = wine.load()
labels = labels.astype(np.int)
seen = set()
for i in xrange(20):
random.seed(2)
i,s = milk.supervised.tree._split(features[::10], labels[::10], None, milk.supervised.tree.information_gain, 2, random)
seen.add(i)
assert len(seen) <= 2
def test_gridsearch():
from milksets import wine
features, labels = wine.load()
selected = (labels < 2)
features = features[selected]
labels = labels[selected]
G = milk.supervised.gridsearch(
milk.supervised.svm.svm_raw(),
params={'C':[.01,.1,1.,10.],
'kernel':[milk.supervised.svm.rbf_kernel(0.1),milk.supervised.svm.rbf_kernel(1.)]
})
model = G.train(features,labels)
reslabels = [model.apply(f) for f in features]
assert len(reslabels) == len(features)
def test_gridminimise_return():
from milksets.wine import load
features, labels = load()
learner = fast_classifier()
gridminimise(learner, features, labels, {'ignore': [0]})
_, error = gridminimise(learner,
features,
labels, {'ignore': [0]},
return_value=True,
nfolds=5)
cmat, _ = milk.nfoldcrossvalidation(features,
labels,
learner=learner,
nfolds=5)
assert error == cmat.sum() - cmat.trace()
def test_gridsearch():
from milksets import wine
features, labels = wine.load()
selected = (labels < 2)
features = features[selected]
labels = labels[selected]
G = milk.supervised.gridsearch(milk.supervised.svm.svm_raw(),
params={
'C': [.01, .1, 1., 10.],
'kernel': [
milk.supervised.svm.rbf_kernel(0.1),
milk.supervised.svm.rbf_kernel(1.)
]
})
model = G.train(features, labels)
reslabels = [model.apply(f) for f in features]
assert len(reslabels) == len(features)
def test_multi_view():
from milksets.wine import load
features, labels = load()
features0 = features[::10]
features1 = features[1::10]
features2 = features[2::10]
labels0 = labels[::10]
labels1 = labels[1::10]
labels2 = labels[2::10]
assert np.all(labels0 == labels1)
assert np.all(labels1 == labels2)
labels = labels0
train_features = list(zip(features0,features1,features2))
test_features = list(zip(features[3::10], features[4::10], features[5::10]))
base = milk.supervised.classifier.ctransforms(
feature_selection_simple(),
milk.supervised.svm.svm_raw(C=128, kernel=milk.supervised.svm.rbf_kernel(4.)),
milk.supervised.svm.svm_sigmoidal_correction()
)
classifier = milk.supervised.multi_view.multi_view_classifier([base,base,base])
model = classifier.train(train_features, labels == 0)
assert ([model.apply(f) for f in test_features] == (labels == 0)).mean() > .9
def test_gridminimise():
from milksets.wine import load
features, labels = load()
x = gridminimise(milk.supervised.svm_simple(kernel=np.dot, C=2.), features[::2], labels[::2] == 0, {'C' : (0.5,) })
cval, = x
assert cval == ('C', .5)
def test_compare_rank_corr():
from milksets.wine import load
features,labels = load()
r0 = rank_corr(features,labels)
r1 = slow_rank_corr(features,labels)
assert np.allclose(r0,r1)
from pylab import *
import numpy as np
from milksets.wine import load
import milk.supervised
import milk.unsupervised.pca
import milk.supervised.svm
features, labels = load()
features = features[labels < 2]
labels = labels[labels < 2]
features, _ = milk.unsupervised.pca(features)
features = features[:, :2]
learner = milk.supervised.svm.svm_raw(kernel=np.dot, C=12)
model = learner.train(features, labels)
w = np.dot(model.svs.T, model.Yw)
b = model.b
x = np.linspace(-.5, .1, 100)
y = -w[0] / w[1] * x + b / w[1]
plot(features[labels == 1][:, 0], features[labels == 1][:, 1], 'bx')
plot(features[labels == 0][:, 0], features[labels == 0][:, 1], 'ro')
plot(x, y)
savefig('svm-demo-points.pdf')
clf()
learner = milk.supervised.svm.svm_raw(
kernel=milk.supervised.svm.rbf_kernel(1.), C=12)
model = learner.train(features, labels)
Y, X = (np.mgrid[:101, :101] - 50) / 12.5
values = [model.apply((y, x)) for y, x in zip(Y.ravel(), X.ravel())]
def test_fisher_approx():
from milksets import wine
features,labels = wine.load()
f = milk.supervised.svm.sigma_value_fisher(features,labels)
for sigma in (2.**-4,2.,16.,32.):
assert abs(f(sigma) - _slow_f(features,labels,sigma)) < 1e-6
def test_sda():
from milksets import wine
features, labels = wine.load()
selected = milk.supervised.featureselection.sda(features, labels)
for sel in selected:
assert sel <= features.shape[1]
def test_extra_arg():
from milksets.wine import load
features, labels = load()
learner = milk.defaultlearner()
model = learner.train(features[::2], labels[::2], extra_arg=5)
assert model.apply(features[1]) < 12.
def test_sda():
from milksets import wine
features, labels = wine.load()
selected = milk.supervised.featureselection.sda(features,labels)
for sel in selected:
assert sel <= features.shape[1]
def test_extra_arg():
from milksets.wine import load
features,labels = load()
learner = milk.defaultlearner()
model = learner.train(features[::2],labels[::2], extra_arg=5)
assert model.apply(features[1]) < 12.
def test_compare_rank_corr():
from milksets.wine import load
features, labels = load()
r0 = rank_corr(features, labels)
r1 = slow_rank_corr(features, labels)
assert np.allclose(r0, r1)
def test_assign_cids():
from milksets.wine import load
features,_ = load()
assigns, centroids = milk.unsupervised.kmeans(features, 3, R=2, max_iters=10)
assert np.all(assign_centroids(features, centroids) == assigns)
from milk.supervised import randomforest
from milk.supervised.multi import one_against_one
import milk.nfoldcrossvalidation
import milk.unsupervised
import pylab
from milksets import wine
# Load 'wine' dataset
features, labels = wine.load()
# random forest learner
rf_learner = randomforest.rf_learner()
# rf is a binary learner, so we transform it into a multi-class classifier
learner = one_against_one(rf_learner)
# cross validate with this learner and return predictions on left-out elements
cmat,names, preds = milk.nfoldcrossvalidation(features, labels, classifier=learner, return_predictions=1)
print('cross-validation accuracy:', cmat.trace()/float(cmat.sum()))
# dimensionality reduction for display
x,v = milk.unsupervised.pca(features)
colors = "rgb" # predicted colour
marks = "xo" # whether the prediction was correct
for (y,x),p,r in zip(x[:,:2], preds, labels):
c = colors[p]
m = marks[p == r]
pylab.plot(y,x,c+m)
pylab.show()
from pylab import *
import numpy as np
from milksets.wine import load
import milk.supervised
import milk.unsupervised.pca
import milk.supervised.svm
features, labels = load()
features = features[labels < 2]
labels = labels[labels < 2]
features,_ = milk.unsupervised.pca(features)
features = features[:,:2]
learner = milk.supervised.svm.svm_raw(kernel=np.dot, C=12)
model = learner.train(features, labels)
w = np.dot(model.svs.T, model.Yw)
b = model.b
x = np.linspace(-.5, .1, 100)
y = -w[0]/w[1]*x + b/w[1]
plot(features[labels == 1][:,0], features[labels == 1][:,1], 'bx')
plot(features[labels == 0][:,0], features[labels == 0][:,1], 'ro')
plot(x,y)
savefig('svm-demo-points.pdf')
clf()