def test_cross():
XX = []
for j in range(-10, 10):
XX.append([j, 0.])
for i in range(1, 10):
XX.append([0., i])
XX.append([0., i - 10])
XX = np.array(XX)
# np.random.shuffle(XX)
dr = DRUHG(max_ranking=200, verbose=False)
dr.fit(XX)
print(XX)
# center is an outlier
labels = dr.labels_
n_clusters = len(set(labels)) - int(-1 in labels)
print('n_clusters', n_clusters)
print('labels', len(labels), labels)
if _plot_graph:
plt.close('all')
dr.minimum_spanning_tree_.plot()
plt.savefig('test_cross.png')
assert (n_clusters == 4)
assert (np.count_nonzero(labels == -1) == 5)
def test_synthetic_outliers():
XX = pd.read_csv('druhg\\tests\\synthetic.csv', sep=',')
XX.drop(u'outlier', axis=1, inplace=True)
XX = np.array(XX)
dr = DRUHG(algorithm='slow',
max_ranking=200,
limit2=len(XX),
exclude=[1978, 1973],
verbose=False)
dr.fit(XX)
if _plot_graph:
plt.close('all')
dr.minimum_spanning_tree_.plot()
plt.savefig('test_synthetic_outliers.png')
# values, counts = np.unique(dr.labels_, return_counts=True)
# for i, v in enumerate(values):
# print (v, counts[i])
labels = dr.labels_
# labels = dr.relabel(limit1=1)
n_clusters = len(set(labels)) - int(-1 in labels)
print(labels)
print('n_clusters', n_clusters)
assert (n_clusters == 6)
def test_cube():
XX = []
size = 10
for i in range(0, size):
for j in range(0, size):
for k in range(0, size):
XX.append([i, j, k])
XX = np.array(XX)
# for i, x in enumerate(XX):
# print (i, x)
dr = DRUHG(algorithm='slow', max_ranking=200, verbose=False)
dr.fit(XX)
s = 2 * len(XX) - 2
print(dr.mst_)
# assert (0==1)
labels = dr.labels_
n_clusters = len(set(labels)) - int(-1 in labels)
print('n_clusters', n_clusters)
print('labels', labels)
assert (n_clusters == 1 + 6)
# labels = dr.relabel(limit1=1)
labels = dr.relabel(limit1=1, limit2=len(XX))
print('out')
n_clusters = len(set(labels)) - int(-1 in labels)
print('n_clusters2', n_clusters)
print('labels2', labels)
assert (n_clusters == 1 + 6 + 12)
def test_plot_dendrogram():
iris = datasets.load_iris()
XX = iris['data']
dr = DRUHG(
max_ranking=50, limit2=int(len(XX) / 2),
fix_outliers=1) #, limit1=0, limit2=int(len(XX)/2), fix_outliers=1)
dr.fit(XX)
dr.single_linkage_.plot()
def test_plot_one_dimension():
iris = datasets.load_iris()
XX = iris['data']
XX = XX.reshape(XX.size, 1)
XX = np.array(sorted(XX))
dr = DRUHG(max_ranking=50)
dr.fit(XX)
dr.minimum_spanning_tree_.plot()
def test_druhg_sparse():
row = np.array([0, 0, 1, 2, 2, 2])
col = np.array([0, 2, 2, 0, 1, 2])
data = np.array([1, 2, 3, 4, 5, 6])
sparse_X = sparse.csr_matrix((data, (row, col)), shape=(3, 3))
dr = DRUHG()
dr.fit(sparse_X)
print('sparse labels', dr.labels_)
def test_moons_and_blobs():
XX = X
dr = DRUHG(max_ranking=50, verbose=False)
dr.fit(XX)
labels = dr.labels_
n_clusters = len(set(labels)) - int(-1 in labels)
# expecting 4 clusters
print(labels)
assert (n_clusters == 4)
def test_copycat():
XX = [[0]] * 100
XX = np.array(XX)
dr = DRUHG(max_ranking=200, limit1=1, verbose=False)
dr.fit(XX)
print(dr.mst_[0], dr.mst_[1])
print('pairs', dr.mst_)
print('labels', dr.labels_)
labels = dr.labels_
assert (all(x == labels[0] for x in labels))
def test_chameleon():
XX = pd.read_csv('druhg\\tests\\chameleon.csv', sep='\t', header=None)
XX = np.array(XX)
dr = DRUHG(algorithm='slow',
max_ranking=200,
limit1=1,
limit2=len(XX),
verbose=False)
dr.fit(XX)
labels = dr.labels_
# labels = dr.relabel(limit1=1)
values, counts = np.unique(labels, return_counts=True)
n_clusters = 0
for i, c in enumerate(counts):
print(i, c, values[i])
if c > 500 and values[i] >= 0:
n_clusters += 1
print('n_clusters', n_clusters)
if _plot_graph:
plt.close('all')
dr.minimum_spanning_tree_.plot()
plt.savefig('test_cham.png')
dr = DRUHG(max_ranking=200,
limit1=1,
limit2=int(len(XX) / 4),
exclude=[],
verbose=False)
dr.fit(XX)
if _plot_graph:
plt.close('all')
dr.minimum_spanning_tree_.plot()
plt.savefig('test_cham2.png')
values, counts = np.unique(dr.labels_, return_counts=True)
for i, v in enumerate(values):
if counts[i] > 200:
print(v, counts[i])
assert (n_clusters == 6)
def test_copycats2():
XX = np.concatenate(([[0]] * 10, [[1]] * 10))
dr = DRUHG(max_ranking=200, limit1=1, verbose=False)
dr.fit(XX)
print(dr.mst_[0], dr.mst_[1])
print('pairs', dr.mst_)
print('labels', dr.labels_)
labels = dr.labels_
assert (not all(x == labels[0] for x in labels))
assert (labels[0] != labels[-1])
uniques, counts = np.unique(labels, True)
print(uniques, counts)
n_clusters = len(set(labels)) - int(-1 in labels)
assert (n_clusters == 2)
def test_copycats3(): # should fail until weights are made
XX = np.concatenate(([[0]] * 100, [[1]] * 100, [[2]] * 5))
dr = DRUHG(max_ranking=10, limit1=1, limit2=250, verbose=False)
dr.fit(XX)
print(dr.mst_[0], dr.mst_[1])
print('pairs', dr.mst_)
print('labels', dr.labels_)
labels = dr.labels_
assert (not all(x == labels[0] for x in labels))
assert (labels[0] != labels[-1])
uniques, counts = np.unique(labels, True)
print(uniques, counts)
n_clusters = len(set(labels)) - int(-1 in labels)
assert (n_clusters == 3)
def test_iris():
iris = datasets.load_iris()
XX = iris['data']
# print (XX, type(XX))
dr = DRUHG(max_ranking=50, verbose=False)
dr.fit(XX)
labels = dr.labels_
ari = adjusted_rand_score(iris['target'], labels)
print('iris ari', ari)
assert (ari >= 0.50)
# breaking biggest cluster
labels = dr.relabel(limit1=0, limit2=int(len(XX) / 2), fix_outliers=1)
ari = adjusted_rand_score(iris['target'], labels)
print('iris ari', ari)
assert (ari >= 0.85)
def test_particles():
XX = [[-0.51, 1.5], [1.51, 1.5]]
for i in range(-3, 5):
for j in range(-6, 1):
XX.append([i, j])
XX = np.array(XX)
dr = DRUHG(max_ranking=200, limit1=1, limit2=len(XX), verbose=False)
dr.fit(XX)
s = 2 * len(XX) - 2
print(dr.mst_)
print(dr.labels_)
print(dr.mst_[s - 1], dr.mst_[s - 2], XX[dr.mst_[s - 1]],
XX[dr.mst_[s - 2]])
# two points are further metrically but close reciprocally
assert (dr.mst_[s - 4] * dr.mst_[s - 3] == 0)
assert (dr.mst_[s - 4] + dr.mst_[s - 3] == 1)
def test_hdbscan_clusterable_data():
XX = np.load('druhg\\tests\\clusterable_data.npy')
dr = DRUHG(max_ranking=50, algorithm='slow', verbose=False)
dr.fit(XX)
labels = dr.labels_
uniques, counts = np.unique(labels, True)
print(uniques, counts)
n_clusters = len(set(labels)) - int(-1 in labels)
print(n_clusters)
if _plot_graph:
plt.close('all')
dr.minimum_spanning_tree_.plot()
plt.savefig('test_hdbscan_clusterable_data.png')
assert (n_clusters == 6)
def test_line():
XX = [[0., 1.], [0., 2.], [0., 3.], [0., 4.]]
XX = np.array(XX)
dr = DRUHG(max_ranking=200, limit1=1, verbose=False)
dr.fit(XX)
# starts from the middle
print(dr.mst_[0], dr.mst_[1])
print(dr.labels_)
assert (dr.mst_[0] * dr.mst_[1] == 2)
# zero clusters cause it always grows by 1
print('pairs', dr.mst_)
print('labels', dr.labels_)
labels = dr.labels_
assert (not all(x == labels[0] for x in labels))
assert (labels[0] == labels[3])
assert (labels[1] == labels[2])
def test_two_squares():
XX = []
size, scale = 6, 1
for i in range(0, size):
for j in range(0, size):
XX.append([scale * i, scale * j])
XX.append([2 * size + scale * i, scale * j])
XX = np.array(XX)
dr = DRUHG(max_ranking=200, verbose=False)
dr.fit(XX)
s = 2 * len(XX) - 2
print(dr.mst_)
print(dr.mst_[s - 1], dr.mst_[s - 2], XX[dr.mst_[s - 1]],
XX[dr.mst_[s - 2]])
labels = dr.labels_
n_clusters = len(set(labels)) - int(-1 in labels)
assert (n_clusters == 2)
def test_scaled_square():
XX = []
size, scale = 10, 0.01
for i in range(0, size):
for j in range(0, size):
XX.append([scale * i, scale * j])
XX = np.array(XX)
dr = DRUHG(max_ranking=200, limit2=len(XX), verbose=False)
dr.fit(XX)
# s = 2*len(XX) - 2
# print (dr.mst_)
# print (dr.mst_[s-1], dr.mst_[s-2], XX[dr.mst_[s-1]], XX[dr.mst_[s-2]])
labels = dr.labels_
print(labels)
n_clusters = len(set(labels)) - int(-1 in labels)
print('n_clusters', n_clusters)
assert (n_clusters == 1)
def test_bomb():
XX = [[0., 1.], [0., 2.], [0., 3.], [0., 4.], [0., 5.]]
for i in range(-3, 4):
for j in range(-6, 1):
XX.append([i, j])
XX = np.array(XX)
dr = DRUHG(algorithm='slow',
max_ranking=200,
limit1=1,
limit2=len(XX),
verbose=False)
dr.fit(XX)
s = 2 * len(XX) - 2
print(dr.mst_)
x = 12
labs = dr.labels_
# fuse is separate
print(labs)
assert (labs[0] == labs[1] == labs[2] == labs[3])
assert (np.count_nonzero(labs == labs[0]) == 4)
def test_longline():
XX = []
for i in range(0, 100):
XX.append([0., i])
XX = np.array(XX)
dr = DRUHG(max_ranking=50, limit1=1, limit2=len(XX), verbose=False)
dr.fit(XX)
# s = 2*len(XX) - 2
# starts somewhere in the middle
# and grows one by one
# that's why there are no clusters
print('pairs', dr.mst_)
print('labels', dr.labels_)
# assert (len(dr.parents_)==0)
labels = dr.labels_
assert (not all(x == labels[0] for x in labels))
assert (labels[0] == labels[len(labels) - 1])
assert (labels[1] == labels[len(labels) - 2])
assert (labels[0] != labels[1])
def test_square():
XX = []
size, scale = 10, 1
for i in range(0, size):
for j in range(0, size):
XX.append([scale * i, scale * j])
XX = np.array(XX)
dr = DRUHG(max_ranking=10, limit1=1, limit2=len(XX), verbose=False)
dr.fit(XX)
s = 2 * len(XX) - 2
print(dr.mst_)
print(dr.mst_[s - 1], dr.mst_[s - 2], XX[dr.mst_[s - 1]],
XX[dr.mst_[s - 2]])
labels = dr.labels_
n_clusters = len(set(labels)) - int(-1 in labels)
print('n_clusters', n_clusters)
print(dr.mst_)
# print (XX)
print(dr.labels_)
# assert (n_clusters==1)
# labels = dr.relabel(limit1=1, limit2=size*2)
n_clusters = len(set(labels)) - int(-1 in labels)
# print('n_clusters', n_clusters)
print('pairs', dr.mst_)
print('labels', dr.labels_)
if _plot_graph:
plt.close('all')
dr.minimum_spanning_tree_.plot()
plt.savefig('test_square.png')
assert (n_clusters >= 5)
labels = dr.labels_
assert (not all(x == labels[0] for x in labels))
assert (labels[1] != labels[size])
assert (labels[1] != labels[2 * size - 1])
assert (labels[size] != labels[2 * size - 1])
assert (labels[size] != labels[2 * size - 1])
assert (labels[1] != labels[len(labels) - 2])
assert (labels[int(size * size / 2)] >= 0)
def test_compound():
XX = pd.read_csv('druhg/tests/Compound.csv', sep=',',
header=None).drop(2, axis=1)
XX = np.array(XX)
dr = DRUHG(max_ranking=200,
limit1=3,
limit2=len(XX),
exclude=[776],
verbose=False)
dr.fit(XX)
if _plot_graph:
plt.close('all')
dr.minimum_spanning_tree_.plot()
plt.savefig('test_compound.png')
labels = dr.labels_
# labels = dr.relabel(limit1=1)
n_clusters = len(set(labels)) - int(-1 in labels)
# np.save('labels_compound', labels)
print('n_clusters', n_clusters)
assert (n_clusters == 5)
def test_2and3():
cons = 10.
XX = [[0., 0.], [1., 1.], [cons + 3., 2.], [cons + 4., 1.],
[cons + 5., 2.]]
XX = np.array(XX)
dr = DRUHG(max_ranking=200, limit1=1, limit2=1000, verbose=False)
dr.fit(XX)
# two clusters
# assert (len(dr.parents_) == 2)
print(dr.mst_)
print(dr.mst_[6] * dr.mst_[7])
# proper connection between two groups
assert (dr.mst_[6] * dr.mst_[7] == 2)
labels = dr.labels_
print('pairs', dr.mst_)
print('labels', dr.labels_)
n_clusters = len(set(labels)) - int(-1 in labels)
print('n_clusters', n_clusters)
assert (labels[0] == labels[1])
assert (not all(x == labels[0] for x in labels))
assert (labels[2] == labels[3] == labels[4])
assert (labels[0] != labels[2])
assert (n_clusters == 2)
def test_druhg_distance_matrix():
D = distance.squareform(distance.pdist(X))
D /= np.max(D)
print(D.shape)
dr = druhg(D, metric='precomputed')
print(dr)
n_clusters = len(set(dr[0])) - int(-1 in dr[0])
print(n_clusters)
assert (n_clusters == 4)
dr = DRUHG(metric="precomputed", limit1=5).fit(D)
labels = dr.labels_
print(labels)
n_clusters = len(set(labels)) - int(-1 in labels)
print(n_clusters)
if _plot_graph:
plt.close('all')
dr.minimum_spanning_tree_.plot()
plt.savefig('test_druhg_distance_matrix.png')
assert (n_clusters == 4)
def test_plot_mst():
iris = datasets.load_iris()
XX = iris['data']
dr = DRUHG(max_ranking=50)
dr.fit(XX)
dr.minimum_spanning_tree_.plot()