def testGetstateSetstate(self):
nDims = 32 # need multiple of 8, because of sse
nClass = 4
size = 20
labels = _RGEN.random_integers(0, nClass - 1, size)
samples = np.zeros((size, nDims), dtype=_DTYPE)
centers = np.array([[0, 0], [0, 1], [1, 0], [1, 1]])
for i in range(0, size):
t = 6.28 * _RGEN.random_sample()
samples[i][
0] = 2 * centers[labels[i]][0] + 0.5 * _RGEN.rand() * np.cos(t)
samples[i][
1] = 2 * centers[labels[i]][1] + 0.5 * _RGEN.rand() * np.sin(t)
classifier = svm_dense(0, nDims, seed=_SEED, probability=True)
for y, xList in zip(labels, samples):
x = np.array(xList, dtype=_DTYPE)
classifier.add_sample(float(y), x)
classifier.train(gamma=1.0 / 3.0, C=100, eps=1e-1)
classifier.cross_validate(2, gamma=0.5, C=10, eps=1e-3)
s1 = classifier.__getstate__()
h1 = hashlib.md5(s1).hexdigest()
classifier2 = svm_dense(0, nDims)
classifier2.__setstate__(s1)
s2 = classifier2.__getstate__()
h2 = hashlib.md5(s2).hexdigest()
self.assertEqual(h1, h2)
with open("svm_test.bin", "wb") as f:
pickle.dump(classifier, f)
with open("svm_test.bin", "rb") as f:
classifier3 = pickle.load(f)
s3 = classifier3.__getstate__()
h3 = hashlib.md5(s3).hexdigest()
self.assertEqual(h1, h3)
os.unlink("svm_test.bin")
def testGetstateSetstate(self):
nDims = 32 # need multiple of 8, because of sse
nClass = 4
size = 20
labels = _RGEN.random_integers(0, nClass - 1, size)
samples = np.zeros((size, nDims), dtype=_DTYPE)
centers = np.array([[0, 0], [0, 1], [1, 0], [1, 1]])
for i in range(0, size):
t = 6.28 * _RGEN.random_sample()
samples[i][0] = 2 * centers[labels[i]][0] + 0.5*_RGEN.rand() * np.cos(t)
samples[i][1] = 2 * centers[labels[i]][1] + 0.5*_RGEN.rand() * np.sin(t)
classifier = svm_dense(0, nDims, seed=_SEED, probability = True)
for y, xList in zip(labels, samples):
x = np.array(xList, dtype=_DTYPE)
classifier.add_sample(float(y), x)
classifier.train(gamma=1.0/3.0, C=100, eps=1e-1)
classifier.cross_validate(2, gamma=0.5, C=10, eps=1e-3)
s1 = classifier.__getstate__()
h1 = hashlib.md5(s1).hexdigest()
classifier2 = svm_dense(0, nDims)
classifier2.__setstate__(s1)
s2 = classifier2.__getstate__()
h2 = hashlib.md5(s2).hexdigest()
self.assertEqual(h1, h2)
with open("svm_test.bin", "wb") as f:
pickle.dump(classifier, f)
with open("svm_test.bin", "rb") as f:
classifier3 = pickle.load(f)
s3 = classifier3.__getstate__()
h3 = hashlib.md5(s3).hexdigest()
self.assertEqual(h1, h3)
os.unlink("svm_test.bin")
def persistence():
print "Persistence"
numpy.random.seed(42)
n_dims = 2
n_class = 12
size = 100
labels = numpy.random.random_integers(0, 256, size)
samples = numpy.zeros((size, n_dims), dtype=type)
print "Generating data"
for i in range(0, size):
t = 6.28 * numpy.random.random_sample()
samples[i][0] = 2 * labels[i] + 1.5 * numpy.cos(t)
samples[i][1] = 2 * labels[i] + 1.5 * numpy.sin(t)
print "Creating dense classifier"
classifier = algo.svm_dense(0, n_dims=n_dims, seed=42)
print "Adding sample vectors to dense classifier"
for y, x_list in zip(labels, samples):
x = numpy.array(x_list, dtype=type)
classifier.add_sample(float(y), x)
print "Pickling dense classifier"
cPickle.dump(classifier, open('test', 'wb'))
classifier = cPickle.load(open('test', 'rb'))
print "Training dense classifier"
classifier.train(gamma=1, C=10, eps=1e-1)
print "Predicting with dense classifier"
print classifier.predict(samples[0])
print "Creating 0/1 classifier"
classifier01 = algo.svm_01(n_dims=n_dims, seed=42)
print "Adding sample vectors to 0/1 classifier"
for y, x_list in zip(labels, samples):
x = numpy.array(x_list, dtype=type)
classifier01.add_sample(float(y), x)
print "Training 0/1 classifier"
classifier01.train(gamma=1. / 3., C=100, eps=1e-1)
print "Pickling 0/1 classifier"
cPickle.dump(classifier01, open('test', 'wb'))
classifier01 = cPickle.load(open('test', 'rb'))
print "Predicting with 0/1 classifier"
print classifier01.predict(numpy.array(samples[0], dtype=type))
def persistence():
print "Persistence"
numpy.random.seed(42)
n_dims = 2
n_class = 12
size = 100
labels = numpy.random.random_integers(0, 256, size)
samples = numpy.zeros((size, n_dims), dtype=type)
print "Generating data"
for i in range(0, size):
t = 6.28 * numpy.random.random_sample()
samples[i][0] = 2 * labels[i] + 1.5 * numpy.cos(t)
samples[i][1] = 2 * labels[i] + 1.5 * numpy.sin(t)
print "Creating dense classifier"
classifier = algo.svm_dense(0, n_dims = n_dims, seed=42)
print "Adding sample vectors to dense classifier"
for y, x_list in zip(labels, samples):
x = numpy.array(x_list, dtype=type)
classifier.add_sample(float(y), x)
print "Pickling dense classifier"
cPickle.dump(classifier, open('test', 'wb'))
classifier = cPickle.load(open('test', 'rb'))
print "Training dense classifier"
classifier.train(gamma = 1, C = 10, eps=1e-1)
print "Predicting with dense classifier"
print classifier.predict(samples[0])
print "Creating 0/1 classifier"
classifier01 = algo.svm_01(n_dims = n_dims, seed=42)
print "Adding sample vectors to 0/1 classifier"
for y, x_list in zip(labels, samples):
x = numpy.array(x_list, dtype=type)
classifier01.add_sample(float(y), x)
print "Training 0/1 classifier"
classifier01.train(gamma = 1./3., C = 100, eps=1e-1)
print "Pickling 0/1 classifier"
cPickle.dump(classifier01, open('test', 'wb'))
classifier01 = cPickle.load(open('test', 'rb'))
print "Predicting with 0/1 classifier"
print classifier01.predict(numpy.array(samples[0], dtype=type))
def testScalability(self):
# Multiple of 8 for sse.
nDims = 32
nClass = 3
size = 1000
labels = _RGEN.random_integers(0, nClass - 1, size)
samples = np.random.random(size=(size, nDims)).astype(_DTYPE)
classifier = svm_dense(0, nDims, seed=_SEED)
for y, xList in zip(labels, samples):
x = np.array(xList, dtype=_DTYPE)
classifier.add_sample(float(y), x)
print "training"
classifier.train(gamma=1.0 / 3.0, C=100, eps=1e-1)
print "cross validation"
classifier.cross_validate(2, gamma=0.5, C=10, eps=1e-3)
def testScalability(self):
# Multiple of 8 for sse.
nDims = 32
nClass = 3
size = 1000
labels = _RGEN.random_integers(0, nClass - 1, size)
samples = np.random.random(size=(size, nDims)).astype(_DTYPE)
classifier = svm_dense(0, nDims, seed=_SEED)
for y, xList in zip(labels, samples):
x = np.array(xList, dtype=_DTYPE)
classifier.add_sample(float(y), x)
print "training"
classifier.train(gamma=1.0/3.0, C=100, eps=1e-1)
print "cross validation"
classifier.cross_validate(2, gamma=0.5, C=10, eps=1e-3)
def testPersistentSize(self):
for _ in range(2):
# Multiple of 8 because of sse.
nDims = 32
nClass = _RGEN.randint(4, 8)
size = _RGEN.randint(20, 50)
labels = _RGEN.random_integers(0, nClass - 1, size)
samples = _RGEN.normal(size=(size, nDims)).astype(_DTYPE)
classifier = svm_dense(0, nDims, seed=_SEED, probability=True)
for y, xList in zip(labels, samples):
x = np.array(xList, dtype=_DTYPE)
classifier.add_sample(float(y), x)
classifier.train(gamma=1.0 / 3.0, C=100, eps=1e-1)
classifier.cross_validate(2, gamma=0.5, C=10, eps=1e-3)
s = classifier.__getstate__()
print("nDims=",
nDims,
"nClass=",
nClass,
"n_vectors=",
size,
end=' ')
print("dense:", len(s), classifier.persistent_size(), end=' ')
self.assertEqual(len(s), classifier.persistent_size())
classifier01 = svm_01(0, nDims, seed=_SEED, probability=True)
for y, xList in zip(labels, samples):
x = np.array(xList, dtype=_DTYPE)
classifier01.add_sample(float(y), x)
classifier01.train(gamma=1.0 / 3.0, C=100, eps=1e-1)
classifier01.cross_validate(2, gamma=0.5, C=10, eps=1e-3)
s = classifier01.__getstate__()
print("0/1", len(s), classifier01.persistent_size())
self.assertEqual(len(s), classifier01.persistent_size())
def cross_validation():
return
print "Cross validation"
numpy.random.seed(42)
labels = [0, 1, 1, 2, 1, 2]
samples = [[0, 0, 0], [0, 1, 0], [1, 0, 1], [1, 1, 1], [1, 1, 0], [0, 1, 1]]
classifier = algo.svm_dense(0, n_dims = 3, seed=42)
print "Adding sample vectors"
for y, x_list in zip(labels, samples):
x = numpy.array(x_list, dtype=type)
classifier.add_sample(float(y), x)
cPickle.dump(classifier, open('test', 'wb'))
classifier = cPickle.load(open('test', 'rb'))
print "Training"
classifier.train(gamma = 1./3., C = 100, eps=1e-1)
print "Cross validation =",
print classifier.cross_validate(3, gamma = .5, C = 10, eps = 1e-3)
def cross_validation():
return
print "Cross validation"
numpy.random.seed(42)
labels = [0, 1, 1, 2, 1, 2]
samples = [[0, 0, 0], [0, 1, 0], [1, 0, 1], [1, 1, 1], [1, 1, 0],
[0, 1, 1]]
classifier = algo.svm_dense(0, n_dims=3, seed=42)
print "Adding sample vectors"
for y, x_list in zip(labels, samples):
x = numpy.array(x_list, dtype=type)
classifier.add_sample(float(y), x)
cPickle.dump(classifier, open('test', 'wb'))
classifier = cPickle.load(open('test', 'rb'))
print "Training"
classifier.train(gamma=1. / 3., C=100, eps=1e-1)
print "Cross validation =",
print classifier.cross_validate(3, gamma=.5, C=10, eps=1e-3)
def testPersistentSize(self):
for _ in range(2):
# Multiple of 8 because of sse.
nDims = 32
nClass = _RGEN.randint(4, 8)
size = _RGEN.randint(20, 50)
labels = _RGEN.random_integers(0, nClass - 1, size)
samples = _RGEN.normal(size=(size, nDims)).astype(_DTYPE)
classifier = svm_dense(0, nDims, seed=_SEED, probability=True)
for y, xList in zip(labels, samples):
x = np.array(xList, dtype=_DTYPE)
classifier.add_sample(float(y), x)
classifier.train(gamma=1.0/3.0, C=100, eps=1e-1)
classifier.cross_validate(2, gamma=0.5, C=10, eps=1e-3)
s = classifier.__getstate__()
print "nDims=", nDims, "nClass=", nClass, "n_vectors=", size,
print "dense:", len(s), classifier.persistent_size(),
self.assertEqual(len(s), classifier.persistent_size())
classifier01 = svm_01(0, nDims, seed=_SEED, probability=True)
for y, xList in zip(labels, samples):
x = np.array(xList, dtype=_DTYPE)
classifier01.add_sample(float(y), x)
classifier01.train(gamma=1.0/3.0, C=100, eps=1e-1)
classifier01.cross_validate(2, gamma=0.5, C=10, eps=1e-3)
s = classifier01.__getstate__()
print "0/1", len(s), classifier01.persistent_size()
self.assertEqual(len(s), classifier01.persistent_size())
def simple():
print "Simple"
numpy.random.seed(42)
n_dims = 2
n_class = 4
size = 200
labels = numpy.random.random_integers(0, n_class - 1, size)
samples = numpy.zeros((size, n_dims), dtype=type)
do_plot = False
print "Generating data"
centers = numpy.array([[0, 0], [0, 1], [1, 0], [1, 1]])
for i in range(0, size):
t = 6.28 * numpy.random.random_sample()
samples[i][0] = 2 * centers[
labels[i]][0] + .5 * numpy.random.random() * numpy.cos(t)
samples[i][1] = 2 * centers[
labels[i]][1] + .5 * numpy.random.random() * numpy.sin(t)
classifier = algo.svm_dense(0, n_dims, probability=True, seed=42)
print "Adding sample vectors"
for y, x_list in zip(labels, samples):
x = numpy.array(x_list, dtype=type)
classifier.add_sample(float(y), x)
print "Displaying problem"
problem = classifier.get_problem()
print "Problem size:", problem.size()
print "Problem dimensionality:", problem.n_dims()
print "Problem samples:"
s = numpy.zeros((problem.size(), problem.n_dims() + 1), dtype=type)
problem.get_samples(s)
print s
if do_plot:
pylab.ion()
pylab.plot(s[s[:, 0] == 0, 1], s[s[:, 0] == 0, 2], '.', color='r')
pylab.plot(s[s[:, 0] == 1, 1], s[s[:, 0] == 1, 2], '+', color='b')
pylab.plot(s[s[:, 0] == 2, 1], s[s[:, 0] == 2, 2], '^', color='g')
pylab.plot(s[s[:, 0] == 3, 1], s[s[:, 0] == 3, 2], 'v', color='g')
print "Training"
classifier.train(gamma=1. / 3., C=100, eps=1e-1)
print "Displaying model"
model = classifier.get_model()
print "Number of support vectors:", model.size()
print "Number of classes:", model.n_class()
print "Number of dimensions: ", model.n_dims()
print "Support vectors:"
sv = numpy.zeros((model.size(), model.n_dims()), dtype=type)
model.get_support_vectors(sv)
print sv
if do_plot:
pylab.plot(sv[:, 0], sv[:, 1], 'o', color='g')
print "Support vector coefficients:"
svc = numpy.zeros((model.n_class() - 1, model.size()), dtype=type)
model.get_support_vector_coefficients(svc)
print svc
print "Hyperplanes (for linear kernel only):"
h = model.get_hyperplanes()
print h
if do_plot:
xmin = numpy.min(samples[:, 0])
xmax = numpy.max(samples[:, 0])
xstep = (xmax - xmin) / 10
X = numpy.arange(xmin, xmax, xstep)
ymin = numpy.min(samples[:, 1])
ymax = numpy.max(samples[:, 1])
ystep = (ymax - ymin) / 10
Y = numpy.arange(ymin, ymax, ystep)
points = numpy.zeros((len(X), len(Y)))
for i, x in enumerate(X):
for j, y in enumerate(Y):
proba = numpy.zeros(model.n_class(), dtype=type)
classifier.predict_probability(numpy.array([x, y]), proba)
points[i, j] = proba[0]
pylab.contour(X, Y, points)
print "Cross-validation"
print classifier.cross_validate(2, gamma=.5, C=10, eps=1e-3)
print "Predicting"
for y, x_list in zip(labels, samples):
x = numpy.array(x_list, dtype=type)
proba = numpy.zeros(model.n_class(), dtype=type)
print x, ': real=', y,
print 'p1=', classifier.predict(x),
print 'p2=', classifier.predict_probability(x, proba),
print 'proba=', proba
print "Discarding problem"
classifier.discard_problem()
print "Predicting after discarding the problem"
for y, x_list in zip(labels, samples):
x = numpy.array(x_list, dtype=type)
proba = numpy.zeros(model.n_class(), dtype=type)
print x, ': real=', y,
print 'p1=', classifier.predict(x),
print 'p2=', classifier.predict_probability(x, proba),
print 'proba=', proba
def persistence():
print("Persistence")
numpy.random.seed(42)
n_dims = 2
n_class = 12
size = 100
labels = numpy.random.random_integers(0, 256, size)
samples = numpy.zeros((size, n_dims), dtype=type)
print("Generating data")
for i in range(0, size):
t = 6.28 * numpy.random.random_sample()
samples[i][0] = 2 * labels[i] + 1.5 * numpy.cos(t)
samples[i][1] = 2 * labels[i] + 1.5 * numpy.sin(t)
print("Creating dense classifier")
classifier = algo.svm_dense(0, n_dims=n_dims, seed=42)
print("Adding sample vectors to dense classifier")
for y, x_list in zip(labels, samples):
x = numpy.array(x_list, dtype=type)
classifier.add_sample(float(y), x)
print("Serializing dense classifier")
# CHH proto is not supported
# schema = classifier.getSchema()
# with open("test", "w+b") as f:
# # Save
# proto = schema.new_message()
# classifier.write(proto)
# proto.write(f)
#
# # Load
# f.seek(0)
# proto2 = schema.read(f)
# classifier = algo.svm_dense.read(proto2)
print("Training dense classifier")
classifier.train(gamma=1, C=10, eps=1e-1)
print("Predicting with dense classifier")
print(classifier.predict(samples[0]))
print("Creating 0/1 classifier")
classifier01 = algo.svm_01(n_dims=n_dims, seed=42)
print("Adding sample vectors to 0/1 classifier")
for y, x_list in zip(labels, samples):
x = numpy.array(x_list, dtype=type)
classifier01.add_sample(float(y), x)
print("Training 0/1 classifier")
classifier01.train(gamma=1. / 3., C=100, eps=1e-1)
print("Serializing 0/1 classifier")
# CHH proto is not supported
# schema = classifier01.getSchema()
# with open("test", "w+b") as f:
# # Save
# proto = schema.new_message()
# classifier01.write(proto)
# proto.write(f)
#
# # Load
# f.seek(0)
# proto2 = schema.read(f)
# classifier01 = algo.svm_01.read(proto2)
print("Predicting with 0/1 classifier")
print(classifier01.predict(numpy.array(samples[0], dtype=type)))
def simple():
print "Simple"
numpy.random.seed(42)
n_dims = 2
n_class = 4
size = 200
labels = numpy.random.random_integers(0, n_class-1, size)
samples = numpy.zeros((size, n_dims), dtype=type)
do_plot = False
print "Generating data"
centers = numpy.array([[0,0],[0,1],[1,0],[1,1]])
for i in range(0, size):
t = 6.28 * numpy.random.random_sample()
samples[i][0] = 2 * centers[labels[i]][0] + .5*numpy.random.random() * numpy.cos(t)
samples[i][1] = 2 * centers[labels[i]][1] + .5*numpy.random.random() * numpy.sin(t)
classifier = algo.svm_dense(0, n_dims, probability=True, seed=42)
print "Adding sample vectors"
for y, x_list in zip(labels, samples):
x = numpy.array(x_list, dtype=type)
classifier.add_sample(float(y), x)
print "Displaying problem"
problem = classifier.get_problem()
print "Problem size:", problem.size()
print "Problem dimensionality:", problem.n_dims()
print "Problem samples:"
s = numpy.zeros((problem.size(), problem.n_dims()+1), dtype=type)
problem.get_samples(s)
print s
if do_plot:
pylab.ion()
pylab.plot(s[s[:,0]==0,1], s[s[:,0]==0,2], '.', color='r')
pylab.plot(s[s[:,0]==1,1], s[s[:,0]==1,2], '+', color='b')
pylab.plot(s[s[:,0]==2,1], s[s[:,0]==2,2], '^', color='g')
pylab.plot(s[s[:,0]==3,1], s[s[:,0]==3,2], 'v', color='g')
print "Training"
classifier.train(gamma = 1./3., C = 100, eps=1e-1)
print "Displaying model"
model = classifier.get_model()
print "Number of support vectors:", model.size()
print "Number of classes:", model.n_class()
print "Number of dimensions: ", model.n_dims()
print "Support vectors:"
sv = numpy.zeros((model.size(), model.n_dims()), dtype=type)
model.get_support_vectors(sv)
print sv
if do_plot:
pylab.plot(sv[:,0], sv[:,1], 'o', color='g')
print "Support vector coefficients:"
svc = numpy.zeros((model.n_class()-1, model.size()), dtype=type)
model.get_support_vector_coefficients(svc)
print svc
print "Hyperplanes (for linear kernel only):"
h = model.get_hyperplanes()
print h
if do_plot:
xmin = numpy.min(samples[:,0])
xmax = numpy.max(samples[:,0])
xstep = (xmax - xmin) / 10
X = numpy.arange(xmin, xmax, xstep)
ymin = numpy.min(samples[:,1])
ymax = numpy.max(samples[:,1])
ystep = (ymax - ymin) / 10
Y = numpy.arange(ymin, ymax, ystep)
points = numpy.zeros((len(X), len(Y)))
for i,x in enumerate(X):
for j,y in enumerate(Y):
proba = numpy.zeros(model.n_class(), dtype=type)
classifier.predict_probability(numpy.array([x,y]), proba)
points[i,j] = proba[0]
pylab.contour(X,Y,points)
print "Cross-validation"
print classifier.cross_validate(2, gamma = .5, C = 10, eps = 1e-3)
print "Predicting"
for y, x_list in zip(labels, samples):
x = numpy.array(x_list, dtype=type)
proba = numpy.zeros(model.n_class(), dtype=type)
print x, ': real=', y,
print 'p1=', classifier.predict(x),
print 'p2=', classifier.predict_probability(x, proba),
print 'proba=', proba
print "Discarding problem"
classifier.discard_problem()
print "Predicting after discarding the problem"
for y, x_list in zip(labels, samples):
x = numpy.array(x_list, dtype=type)
proba = numpy.zeros(model.n_class(), dtype=type)
print x, ': real=', y,
print 'p1=', classifier.predict(x),
print 'p2=', classifier.predict_probability(x, proba),
print 'proba=', proba
def persistence():
print "Persistence"
numpy.random.seed(42)
n_dims = 2
n_class = 12
size = 100
labels = numpy.random.random_integers(0, 256, size)
samples = numpy.zeros((size, n_dims), dtype=type)
print "Generating data"
for i in range(0, size):
t = 6.28 * numpy.random.random_sample()
samples[i][0] = 2 * labels[i] + 1.5 * numpy.cos(t)
samples[i][1] = 2 * labels[i] + 1.5 * numpy.sin(t)
print "Creating dense classifier"
classifier = algo.svm_dense(0, n_dims = n_dims, seed=42)
print "Adding sample vectors to dense classifier"
for y, x_list in zip(labels, samples):
x = numpy.array(x_list, dtype=type)
classifier.add_sample(float(y), x)
print "Serializing dense classifier"
schema = classifier.getSchema()
with open("test", "w+b") as f:
# Save
proto = schema.new_message()
classifier.write(proto)
proto.write(f)
# Load
f.seek(0)
proto2 = schema.read(f)
classifier = algo.svm_dense.read(proto2)
print "Training dense classifier"
classifier.train(gamma = 1, C = 10, eps=1e-1)
print "Predicting with dense classifier"
print classifier.predict(samples[0])
print "Creating 0/1 classifier"
classifier01 = algo.svm_01(n_dims = n_dims, seed=42)
print "Adding sample vectors to 0/1 classifier"
for y, x_list in zip(labels, samples):
x = numpy.array(x_list, dtype=type)
classifier01.add_sample(float(y), x)
print "Training 0/1 classifier"
classifier01.train(gamma = 1./3., C = 100, eps=1e-1)
print "Serializing 0/1 classifier"
schema = classifier01.getSchema()
with open("test", "w+b") as f:
# Save
proto = schema.new_message()
classifier01.write(proto)
proto.write(f)
# Load
f.seek(0)
proto2 = schema.read(f)
classifier01 = algo.svm_01.read(proto2)
print "Predicting with 0/1 classifier"
print classifier01.predict(numpy.array(samples[0], dtype=type))
