def test_random_matrix(self):
data = np.trunc(5 + 10 * np.random.rand(10, 3))
neighbors = 2
epsilon = 5.
to_dimension = 2
d = Displayer(title="Isomap algorithms comparison") \
.load(title="Canonical data set.", data=data)
start = time()
result = manifold.Isomap(neighbors, to_dimension).fit_transform(data)
elapsed = time() - start
d.load(title="SKLearn's Isomap with %i neighbors, taking %.1fs." % (neighbors, elapsed),
data=result)
start = time()
result = algorithms.Isomap(nearest_method='e', e=epsilon, n_components=to_dimension) \
.transform(data)
elapsed = time() - start
d.load(title="My E-Isomap with epsilon %i, taking %.1fs." % (epsilon, elapsed),
data=result)
start = time()
result = algorithms.Isomap(k=neighbors, n_components=to_dimension) \
.transform(data)
elapsed = time() - start
d.load(title="My K-Isomap with %i neighbors, taking %.1fs." % (neighbors, elapsed),
data=result)
d.show()
def test_display_dimensions(self):
data_dir = 'datasets/'
data_set = 'glass/glass.data'
file = os.path.join(data_dir, data_set)
print('Displaying data set {%s} in the Rn' % file)
glass = Retriever(file, delimiter=',')
# Glass has the samples' ids in the first column.
glass.split_column(0)
# Additionally, its last column represents the target feature.
glass.split_target()
data, c = glass.retrieve()
reduced_data = algorithms.Isomap(data, e=20).run()
d = Displayer(title=data_set)
# Scatter all dimensions (3-by-3), using as many graphs as necessary.
for begin in range(0, glass.features_count, 3):
end = min(glass.features_count, begin + 3)
d.load(data[:, begin:end],
color=c,
title='Dimensions: d e [%i, %i]' % (begin + 1, end))
d \
.load('Reduced glass data-set', reduced_data, c) \
.show()
def test_rendering(self):
i = algorithms.Isomap(self.test_data, n_components=2)
result = i.run()
Displayer() \
.load(result, np.random.rand(4), title='K-Isomap') \
.show()
def test_canonical_data(self):
data = np.array([
[0, 0, 0],
[1, 0, 0],
[1, 1, 0],
[1, 1, 1],
])
c = np.array([0, 0, 1, 1])
neighbors = 2
epsilon = 1.5
to_dimension = 2
d = Displayer(title="Isomap algorithms comparison") \
.load(title="Canonical data set.", data=data, color=c)
start = time()
result = manifold.Isomap(neighbors, to_dimension).fit_transform(data)
elapsed = time() - start
d.load(title="SKLearn's Isomap with %i neighbors, taking %.1fs." % (neighbors, elapsed),
data=result,
color=c)
start = time()
result = algorithms.Isomap(nearest_method='e', e=epsilon, n_components=to_dimension) \
.transform(data)
elapsed = time() - start
d.load(title="My E-Isomap with epsilon %i, taking %.1fs." % (epsilon, elapsed),
data=result,
color=c)
start = time()
result = algorithms.Isomap(k=neighbors, n_components=to_dimension) \
.transform(data)
elapsed = time() - start
d.load(title="My K-Isomap with %i neighbors, taking %.1fs." % (neighbors, elapsed),
data=result,
color=c)
d.show()
def reduce(self, evaluate=False, verbose=True):
assert self.reduction_method in ('pca', 'mds', 'isomap', 'skisomap'), \
'Error: unknown reduction method.'
print('Reducing...')
to_dimension = self.n_dimensions_()
data = self.original_data
print('\tMethod: %s' % self.reduction_method)
print('\tR^%i --> R^%i' % (data.shape[1], to_dimension))
start = time.time()
if self.reduction_method == 'pca':
self.reducer = decomposition.PCA(**self.reduction_params)
self.data = self.reducer.fit_transform(data)
elif self.reduction_method == 'mds':
self.reducer = algorithms.MDS(verbose=verbose,
**self.reduction_params)
self.data = self.reducer.transform(data)
elif self.reduction_method == 'isomap':
self.reducer = algorithms.Isomap(verbose=verbose,
**self.reduction_params)
self.data = self.reducer.transform(data)
elif self.reduction_method == 'skisomap':
self.reducer = manifold.Isomap(**self.reduction_params)
self.data = self.reducer.fit_transform(data)
end = time.time()
if evaluate:
self.evaluate()
self.displayer.load(self.data, self.target)
print('Done. (%.2f s)\n' % (end - start))