def test_DTWKmeans_inertia_positive(self):
list_of_series = make_flat_dataset([-1.0,0,0.5],10,additive_noise_factor=0.3,level_noise_factor=0.3,lengths=[5])
num_clusters = 3
iterations = 1
random_seed = 101
clts_1 = DTWKmeans(num_clust = num_clusters, num_iter = iterations, random_seed=random_seed)
clts_1.fit(list_of_series)
intertia = clts_1._inertia(list_of_series)
assert intertia > 0
def test_random_seed(self):
list_levels = [-1.0, 0, 0.5]
list_size = [10, 5, 20]
list_lenghts = [5]
random_seed = 101
list_of_series_1 = make_flat_dataset(list_levels,
list_size,
additive_noise_factor=0.3,
level_noise_factor=0.3,
lengths=list_lenghts,
random_seed=random_seed)
list_of_series_2 = make_flat_dataset(list_levels,
list_size,
additive_noise_factor=0.3,
level_noise_factor=0.3,
lengths=list_lenghts,
random_seed=random_seed)
assert lists_of_series_are_equal(list_of_series_1, list_of_series_2)
def test_flat_unbalanced_clusters(self):
list_levels = [-1.0, 0, 0.5]
list_size = [10, 5, 20]
list_lenghts = [5]
list_of_series = make_flat_dataset(list_levels,
list_size,
additive_noise_factor=0.3,
level_noise_factor=0.3,
lengths=list_lenghts)
assert len(list_of_series) == sum(list_size)
def test_flat_equal_clusters(self):
list_levels = [-1.0, 0, 0.5]
scalar_size = 10
list_lenghts = [5]
list_of_series = make_flat_dataset(list_levels,
scalar_size,
additive_noise_factor=0.0,
level_noise_factor=0.0,
lengths=list_lenghts)
assert len(list_of_series) == scalar_size * len(list_levels)
def flat_dataset(random_seed=101):
# build the dataset around 3 levels
levels = [1.5,0,-1.5]
# with different number of elements for each cluster
sizes = [15,30,10]
# set random seed for reproduceability, you can remove the argument to allow different results for each run
list_of_series = make_flat_dataset(levels,sizes,
additive_noise_factor=0.4,level_noise_factor=0.4,
lengths=[10],random_seed=random_seed)
return list_of_series
def test_DTWKmeans_inertia_decrease_with_iteration_increase(self):
list_of_series = make_flat_dataset([-1.0,0,0.5],10,additive_noise_factor=0.3,level_noise_factor=0.3,lengths=[5])
num_clusters = 3
random_seed = 101
clts_1 = DTWKmeans(num_clust = num_clusters, num_iter=1, random_seed=random_seed)
clts_1.fit(list_of_series)
clts_2 = DTWKmeans(num_clust = num_clusters, num_iter=2, random_seed=random_seed)
clts_2.fit(list_of_series)
print (clts_1._inertia(list_of_series))
print (clts_2._inertia(list_of_series))
#assert False
assert clts_1._inertia(list_of_series) >= clts_2._inertia(list_of_series)
def test_DTWKmeans_fit_is_reproduceable_using_random_seed(self):
list_of_series = make_flat_dataset([-1.0,0,1.0],10,additive_noise_factor=0.1,level_noise_factor=0.1,lengths=[5])
num_clusters = 3
iterations = 1
random_seed = 101
clts_1 = DTWKmeans(num_clust = num_clusters, num_iter = iterations, random_seed=random_seed)
clts_1.fit(list_of_series)
df1 = pd.DataFrame(clts_1.cluster_centers_)
clts_2 = DTWKmeans(num_clust = num_clusters, num_iter = iterations, random_seed=random_seed)
clts_2.fit(list_of_series)
df2 = pd.DataFrame(clts_2.cluster_centers_)
assert np.all(df1.values==df2.values)