def test_cluster_map_centroid_getitem():
nb_clusters = 11
indices = list(range(len(data)))
np.random.shuffle(indices) # None trivial ordering
advanced_indices = indices + [0, 1, 2, -1, -2, -3]
cluster_map = ClusterMapCentroid()
clusters = []
for i in range(nb_clusters):
centroid = np.zeros_like(features)
cluster = ClusterCentroid(centroid)
cluster.id = cluster_map.add_cluster(cluster)
clusters.append(cluster)
# Test indexing
for i in advanced_indices:
assert_equal(cluster_map[i], clusters[i])
# Test advanced indexing
assert_arrays_equal(cluster_map[advanced_indices],
[clusters[i] for i in advanced_indices])
# Test index out of bounds
assert_raises(IndexError, cluster_map.__getitem__, len(clusters))
assert_raises(IndexError, cluster_map.__getitem__, -len(clusters) - 1)
# Test slicing and negative indexing
assert_equal(cluster_map[-1], clusters[-1])
assert_array_equal(cluster_map[::2], clusters[::2])
assert_arrays_equal(cluster_map[::-1], clusters[::-1])
assert_arrays_equal(cluster_map[:-1], clusters[:-1])
assert_arrays_equal(cluster_map[1:], clusters[1:])
def test_cluster_map_centroid_remove_cluster():
clusters = ClusterMapCentroid()
centroid1 = np.random.rand(*features_shape).astype(dtype)
cluster1 = ClusterCentroid(centroid1, indices=[1])
clusters.add_cluster(cluster1)
centroid2 = np.random.rand(*features_shape).astype(dtype)
cluster2 = ClusterCentroid(centroid2, indices=[1, 2])
clusters.add_cluster(cluster2)
centroid3 = np.random.rand(*features_shape).astype(dtype)
cluster3 = ClusterCentroid(centroid3, indices=[1, 2, 3])
clusters.add_cluster(cluster3)
assert_equal(len(clusters), 3)
clusters.remove_cluster(cluster2)
assert_equal(len(clusters), 2)
assert_array_equal(list(itertools.chain(*clusters)),
list(itertools.chain(*[cluster1, cluster3])))
assert_array_equal(clusters.centroids, np.array([centroid1, centroid3]))
assert_equal(clusters[0], cluster1)
assert_equal(clusters[1], cluster3)
clusters.remove_cluster(cluster3)
assert_equal(len(clusters), 1)
assert_array_equal(list(itertools.chain(*clusters)), list(cluster1))
assert_array_equal(clusters.centroids, np.array([centroid1]))
assert_equal(clusters[0], cluster1)
clusters.remove_cluster(cluster1)
assert_equal(len(clusters), 0)
assert_array_equal(list(itertools.chain(*clusters)), [])
assert_array_equal(clusters.centroids, [])
def test_cluster_map_centroid_getitem():
nb_clusters = 11
indices = list(range(len(data)))
np.random.shuffle(indices) # None trivial ordering
advanced_indices = indices + [0, 1, 2, -1, -2, -3]
cluster_map = ClusterMapCentroid()
clusters = []
for i in range(nb_clusters):
centroid = np.zeros_like(features)
cluster = ClusterCentroid(centroid)
cluster.id = cluster_map.add_cluster(cluster)
clusters.append(cluster)
# Test indexing
for i in advanced_indices:
assert_equal(cluster_map[i], clusters[i])
# Test advanced indexing
assert_array_equal(cluster_map[advanced_indices], [clusters[i] for i in advanced_indices])
# Test index out of bounds
assert_raises(IndexError, cluster_map.__getitem__, len(clusters))
assert_raises(IndexError, cluster_map.__getitem__, -len(clusters)-1)
# Test slicing and negative indexing
assert_equal(cluster_map[-1], clusters[-1])
assert_array_equal(cluster_map[::2], clusters[::2])
assert_arrays_equal(cluster_map[::-1], clusters[::-1])
assert_arrays_equal(cluster_map[:-1], clusters[:-1])
assert_arrays_equal(cluster_map[1:], clusters[1:])
def test_cluster_map_centroid_add_cluster():
clusters = ClusterMapCentroid()
centroids = []
for i in range(3):
cluster = ClusterCentroid(centroid=np.zeros_like(features))
centroids.append(np.zeros_like(features))
for id_data in range(2*i):
centroids[-1] = (centroids[-1]*id_data + (id_data+1)*features) / (id_data+1)
cluster.assign(id_data, (id_data+1)*features)
cluster.update()
clusters.add_cluster(cluster)
assert_array_equal(cluster.centroid, centroids[-1])
assert_equal(type(cluster), ClusterCentroid)
assert_equal(cluster, clusters[-1])
assert_equal(type(clusters.centroids), list)
assert_array_equal(list(itertools.chain(*clusters.centroids)), list(itertools.chain(*centroids)))
# Check adding features of different sizes (shorter and longer)
features_shape_short = (1, features_shape[1]-3)
features_too_short = np.ones(features_shape_short, dtype=dtype)
assert_raises(ValueError, cluster.assign, 123, features_too_short)
features_shape_long = (1, features_shape[1]+3)
features_too_long = np.ones(features_shape_long, dtype=dtype)
assert_raises(ValueError, cluster.assign, 123, features_too_long)
def test_cluster_centroid_getitem():
indices = list(range(len(data)))
np.random.shuffle(indices) # None trivial ordering
advanced_indices = indices + [0, 1, 2, -1, -2, -3]
# Test without specifying refdata in ClusterCentroid
centroid = np.zeros(features_shape)
cluster = ClusterCentroid(centroid)
for idx in indices:
cluster.assign(idx, (idx + 1) * features)
# Test indexing
for i in advanced_indices:
assert_equal(cluster[i], indices[i])
# Test advanced indexing
assert_array_equal(cluster[advanced_indices],
[indices[i] for i in advanced_indices])
# Test index out of bounds
assert_raises(IndexError, cluster.__getitem__, len(cluster))
assert_raises(IndexError, cluster.__getitem__, -len(cluster) - 1)
# Test slicing and negative indexing
assert_equal(cluster[-1], indices[-1])
assert_array_equal(cluster[::2], indices[::2])
assert_arrays_equal(cluster[::-1], indices[::-1])
assert_arrays_equal(cluster[:-1], indices[:-1])
assert_arrays_equal(cluster[1:], indices[1:])
# Test with specifying refdata in ClusterCentroid
cluster.refdata = data
# Test indexing
for i in advanced_indices:
assert_array_equal(cluster[i], data[indices[i]])
# Test advanced indexing
assert_arrays_equal(cluster[advanced_indices],
[data[indices[i]] for i in advanced_indices])
# Test index out of bounds
assert_raises(IndexError, cluster.__getitem__, len(cluster))
assert_raises(IndexError, cluster.__getitem__, -len(cluster) - 1)
# Test slicing and negative indexing
assert_array_equal(cluster[-1], data[indices[-1]])
assert_arrays_equal(cluster[::2], [data[i] for i in indices[::2]])
assert_arrays_equal(cluster[::-1], [data[i] for i in indices[::-1]])
assert_arrays_equal(cluster[:-1], [data[i] for i in indices[:-1]])
assert_arrays_equal(cluster[1:], [data[i] for i in indices[1:]])
def test_cluster_centroid_getitem():
indices = list(range(len(data)))
np.random.shuffle(indices) # None trivial ordering
advanced_indices = indices + [0, 1, 2, -1, -2, -3]
# Test without specifying refdata in ClusterCentroid
centroid = np.zeros(features_shape)
cluster = ClusterCentroid(centroid)
for idx in indices:
cluster.assign(idx, (idx+1)*features)
# Test indexing
for i in advanced_indices:
assert_equal(cluster[i], indices[i])
# Test advanced indexing
assert_array_equal(cluster[advanced_indices],
[indices[i] for i in advanced_indices])
# Test index out of bounds
assert_raises(IndexError, cluster.__getitem__, len(cluster))
assert_raises(IndexError, cluster.__getitem__, -len(cluster)-1)
# Test slicing and negative indexing
assert_equal(cluster[-1], indices[-1])
assert_array_equal(cluster[::2], indices[::2])
assert_arrays_equal(cluster[::-1], indices[::-1])
assert_arrays_equal(cluster[:-1], indices[:-1])
assert_arrays_equal(cluster[1:], indices[1:])
# Test with specifying refdata in ClusterCentroid
cluster.refdata = data
# Test indexing
for i in advanced_indices:
assert_array_equal(cluster[i], data[indices[i]])
# Test advanced indexing
assert_arrays_equal(cluster[advanced_indices],
[data[indices[i]] for i in advanced_indices])
# Test index out of bounds
assert_raises(IndexError, cluster.__getitem__, len(cluster))
assert_raises(IndexError, cluster.__getitem__, -len(cluster)-1)
# Test slicing and negative indexing
assert_array_equal(cluster[-1], data[indices[-1]])
assert_arrays_equal(cluster[::2], [data[i] for i in indices[::2]])
assert_arrays_equal(cluster[::-1], [data[i] for i in indices[::-1]])
assert_arrays_equal(cluster[:-1], [data[i] for i in indices[:-1]])
assert_arrays_equal(cluster[1:], [data[i] for i in indices[1:]])
def test_cluster_map_centroid_iter():
rng = np.random.RandomState(42)
nb_clusters = 11
cluster_map = ClusterMapCentroid()
clusters = []
for i in range(nb_clusters):
new_centroid = np.zeros_like(features)
new_cluster = ClusterCentroid(new_centroid,
indices=rng.randint(0,
len(data),
size=10))
cluster_map.add_cluster(new_cluster)
clusters.append(new_cluster)
assert_true(
all([c1 is c2 for c1, c2 in zip(cluster_map.clusters, clusters)]))
assert_array_equal(cluster_map, clusters)
assert_array_equal(cluster_map.clusters, clusters)
assert_array_equal(cluster_map, [cluster.indices for cluster in clusters])
# Set refdata
cluster_map.refdata = data
for c1, c2 in zip(cluster_map, clusters):
assert_arrays_equal(c1, [data[i] for i in c2.indices])
def test_cluster_centroid_iter():
indices = list(range(len(data)))
np.random.shuffle(indices) # None trivial ordering
# Test without specifying refdata in ClusterCentroid
centroid = np.zeros(features_shape)
cluster = ClusterCentroid(centroid)
for idx in indices:
cluster.assign(idx, (idx + 1) * features)
assert_array_equal(cluster.indices, indices)
assert_array_equal(list(cluster), indices)
# Test with specifying refdata in ClusterCentroid
cluster.refdata = data
assert_arrays_equal(list(cluster), [data[i] for i in indices])
def test_cluster_centroid_iter():
indices = list(range(len(data)))
np.random.shuffle(indices) # None trivial ordering
# Test without specifying refdata in ClusterCentroid
centroid = np.zeros(features_shape)
cluster = ClusterCentroid(centroid)
for idx in indices:
cluster.assign(idx, (idx+1)*features)
assert_array_equal(cluster.indices, indices)
assert_array_equal(list(cluster), indices)
# Test with specifying refdata in ClusterCentroid
cluster.refdata = data
assert_arrays_equal(list(cluster), [data[i] for i in indices])
def test_cluster_centroid_assign():
centroid = np.zeros(features_shape)
cluster = ClusterCentroid(centroid)
indices = []
centroid = np.zeros(features_shape, dtype=dtype)
for idx in range(1, 10):
cluster.assign(idx, (idx+1) * features)
cluster.update()
indices.append(idx)
centroid = (centroid * (idx-1) + (idx+1) * features) / idx
assert_equal(len(cluster), idx)
assert_equal(type(cluster.indices), list)
assert_array_equal(cluster.indices, indices)
assert_equal(type(cluster.centroid), np.ndarray)
assert_array_equal(cluster.centroid, centroid)
def test_cluster_centroid_attributes_and_constructor():
centroid = np.zeros(features_shape)
cluster = ClusterCentroid(centroid)
assert_equal(type(cluster), ClusterCentroid)
assert_equal(cluster.id, 0)
assert_array_equal(cluster.indices, [])
assert_array_equal(cluster.centroid, np.zeros(features_shape))
assert_equal(len(cluster), 0)
# Duplicate
assert_equal(cluster, ClusterCentroid(centroid))
assert_false(cluster != ClusterCentroid(centroid))
assert_false(cluster == ClusterCentroid(centroid + 1))
# Invalid comparison
assert_raises(TypeError, cluster.__cmp__, cluster)
def test_cluster_map_centroid_add_cluster():
clusters = ClusterMapCentroid()
centroids = []
for i in range(3):
cluster = ClusterCentroid(centroid=np.zeros_like(features))
centroids.append(np.zeros_like(features))
for id_data in range(2 * i):
centroids[-1] = ((centroids[-1] * id_data +
(id_data + 1) * features) / (id_data + 1))
cluster.assign(id_data, (id_data + 1) * features)
cluster.update()
clusters.add_cluster(cluster)
assert_array_equal(cluster.centroid, centroids[-1])
assert_equal(type(cluster), ClusterCentroid)
assert_equal(cluster, clusters[-1])
assert_equal(type(clusters.centroids), list)
assert_array_equal(list(itertools.chain(*clusters.centroids)),
list(itertools.chain(*centroids)))
# Check adding features of different sizes (shorter and longer)
features_shape_short = (1, features_shape[1] - 3)
features_too_short = np.ones(features_shape_short, dtype=dtype)
assert_raises(ValueError, cluster.assign, 123, features_too_short)
features_shape_long = (1, features_shape[1] + 3)
features_too_long = np.ones(features_shape_long, dtype=dtype)
assert_raises(ValueError, cluster.assign, 123, features_too_long)
def test_cluster_centroid_assign():
centroid = np.zeros(features_shape)
cluster = ClusterCentroid(centroid)
indices = []
centroid = np.zeros(features_shape, dtype=dtype)
for idx in range(1, 10):
cluster.assign(idx, (idx + 1) * features)
cluster.update()
indices.append(idx)
centroid = (centroid * (idx - 1) + (idx + 1) * features) / idx
assert_equal(len(cluster), idx)
assert_equal(type(cluster.indices), list)
assert_array_equal(cluster.indices, indices)
assert_equal(type(cluster.centroid), np.ndarray)
assert_array_equal(cluster.centroid, centroid)
def test_cluster_map_centroid_comparison_with_int():
clusters1_indices = range(10)
clusters2_indices = range(10, 15)
clusters3_indices = [15]
# Build a test ClusterMapCentroid
centroid = np.zeros_like(features)
cluster1 = ClusterCentroid(centroid.copy())
for i in clusters1_indices:
cluster1.assign(i, features)
cluster2 = ClusterCentroid(centroid.copy())
for i in clusters2_indices:
cluster2.assign(i, features)
cluster3 = ClusterCentroid(centroid.copy())
for i in clusters3_indices:
cluster3.assign(i, features)
# Update centroids
cluster1.update()
cluster2.update()
cluster3.update()
clusters = ClusterMapCentroid()
clusters.add_cluster(cluster1)
clusters.add_cluster(cluster2)
clusters.add_cluster(cluster3)
subset = clusters < 5
assert_equal(subset.sum(), 1)
assert_array_equal(list(clusters[subset][0]), clusters3_indices)
subset = clusters <= 5
assert_equal(subset.sum(), 2)
assert_array_equal(list(clusters[subset][0]), clusters2_indices)
assert_array_equal(list(clusters[subset][1]), clusters3_indices)
subset = clusters == 5
assert_equal(subset.sum(), 1)
assert_array_equal(list(clusters[subset][0]), clusters2_indices)
subset = clusters != 5
assert_equal(subset.sum(), 2)
assert_array_equal(list(clusters[subset][0]), clusters1_indices)
assert_array_equal(list(clusters[subset][1]), clusters3_indices)
subset = clusters > 5
assert_equal(subset.sum(), 1)
assert_array_equal(list(clusters[subset][0]), clusters1_indices)
subset = clusters >= 5
assert_equal(subset.sum(), 2)
assert_array_equal(list(clusters[subset][0]), clusters1_indices)
assert_array_equal(list(clusters[subset][1]), clusters2_indices)
def test_cluster_map_centroid_comparison_with_int():
clusters1_indices = range(10)
clusters2_indices = range(10, 15)
clusters3_indices = [15]
# Build a test ClusterMapCentroid
centroid = np.zeros_like(features)
cluster1 = ClusterCentroid(centroid.copy())
for i in clusters1_indices:
cluster1.assign(i, features)
cluster2 = ClusterCentroid(centroid.copy())
for i in clusters2_indices:
cluster2.assign(i, features)
cluster3 = ClusterCentroid(centroid.copy())
for i in clusters3_indices:
cluster3.assign(i, features)
# Update centroids
cluster1.update()
cluster2.update()
cluster3.update()
clusters = ClusterMapCentroid()
clusters.add_cluster(cluster1)
clusters.add_cluster(cluster2)
clusters.add_cluster(cluster3)
subset = clusters < 5
assert_equal(subset.sum(), 1)
assert_array_equal(list(clusters[subset][0]), clusters3_indices)
subset = clusters <= 5
assert_equal(subset.sum(), 2)
assert_array_equal(list(clusters[subset][0]), clusters2_indices)
assert_array_equal(list(clusters[subset][1]), clusters3_indices)
subset = clusters == 5
assert_equal(subset.sum(), 1)
assert_array_equal(list(clusters[subset][0]), clusters2_indices)
subset = clusters != 5
assert_equal(subset.sum(), 2)
assert_array_equal(list(clusters[subset][0]), clusters1_indices)
assert_array_equal(list(clusters[subset][1]), clusters3_indices)
subset = clusters > 5
assert_equal(subset.sum(), 1)
assert_array_equal(list(clusters[subset][0]), clusters1_indices)
subset = clusters >= 5
assert_equal(subset.sum(), 2)
assert_array_equal(list(clusters[subset][0]), clusters1_indices)
assert_array_equal(list(clusters[subset][1]), clusters2_indices)