def compare_cluster_runtime(data, n_clusters, n_components):
t0 = time()
features = data[0]
KMeans(n_clusters=n_clusters).fit_transform(features)
t1 = time() - t0
t0 = time()
mixture.GMM(n_components=n_clusters).fit(features)
t2 = time() - t0
reduced_higgs_data, t3 = pca_eval.transform(data, n_components=n_components)
reduced_higgs_data, t4 = rand_projections.transform(data, n_components=n_components)
reduced_higgs_data, t5 = ica_eval.transform(data, n_components=n_components)
reduced_higgs_data, t6 = trunk_svd.transform(data, n_components=n_components)
t0 = time()
KMeans(n_clusters=n_clusters).fit_transform(reduced_higgs_data)
t7 = time() - t0
t0 = time()
mixture.GMM(n_components=n_clusters).fit(reduced_higgs_data)
t8 = time() - t0
ser = pd.Series([t1, t2, t3, t4, t5, t6, t7, t8], index=['original Kmeans clustering',
'original GMM clustering',
'PCA', 'RCA', 'ICA', 'LSA',
'reduced Kmeans clustering',
'reduced GMM clustering'])
ser.name = 'time'
return ser
def gmm_transform(higgs_data, n_clusters, n_components):
start = time()
reduced_higgs_data, elapsed = pca_eval.transform(higgs_data, n_components=n_components)
cluster_assignment = mixture.GMM(n_components=n_clusters).fit(reduced_higgs_data).predict(reduced_higgs_data)
elapsed = time() - start
reduced_higgs_data['cluster_assighment'] = cluster_assignment
data = {'features': reduced_higgs_data, 'weights': higgs_data[1], 'labels': higgs_data[2]}
return data, elapsed
def run_higg_dimensionality_reduction(higgs_data, n_components):
pca_trns, pca_elapsed = pca_eval.transform(higgs_data, n_components=n_components)
rand_proj_trns, rand_proj_elapsed = rand_projections.transform(higgs_data, n_components=n_components)
ica_trns, ica_elapsed = ica_eval.transform(higgs_data, n_components=n_components)
lsa_trns, lsa_elapsed = trunk_svd.transform(higgs_data, n_components=n_components)
transformation_time = pd.Series([pca_elapsed, rand_proj_elapsed, ica_elapsed, lsa_elapsed],
index=['PCA', 'RCA', 'ICA', 'LSA'],
name='transformation_time')
return {'PCA': pca_trns, 'RCA': rand_proj_trns, 'ICA': ica_trns, 'LSA': lsa_trns}, transformation_time
def kmeans_transform(higgs_data, n_clusters, n_components, display=False):
start = time()
reduced_higgs_data, elapsed = pca_eval.transform(higgs_data, n_components=n_components)
cluster_data = KMeans(n_clusters=n_clusters).fit_transform(reduced_higgs_data)
elapsed = time() - start
for l in range(cluster_data.shape[1]):
reduced_higgs_data['new_feature' + str(l)] = cluster_data[:, l]
data = {'features': reduced_higgs_data, 'weights': higgs_data[1], 'labels': higgs_data[2]}
if display and n_clusters == 2:
df = pd.DataFrame.from_records(cluster_data, columns=['new_feature_' + str(n) for n in range(n_clusters)])
df['label'] = higgs_data[2].values
ax = df[df.label == 's'].plot(x='new_feature_0', y='new_feature_1',
kind='scatter', color='darkgreen', label='signal')
df[df.label == 'b'].plot(x='new_feature_0', y='new_feature_1',
kind='scatter', color='darkred', ax=ax, label='background')
return data, elapsed
def run_higg_dimensionality_reduction(higgs_data, n_components):
pca_trns, pca_elapsed = pca_eval.transform(higgs_data,
n_components=n_components)
rand_proj_trns, rand_proj_elapsed = rand_projections.transform(
higgs_data, n_components=n_components)
ica_trns, ica_elapsed = ica_eval.transform(higgs_data,
n_components=n_components)
lsa_trns, lsa_elapsed = trunk_svd.transform(higgs_data,
n_components=n_components)
transformation_time = pd.Series(
[pca_elapsed, rand_proj_elapsed, ica_elapsed, lsa_elapsed],
index=['PCA', 'RCA', 'ICA', 'LSA'],
name='transformation_time')
return {
'PCA': pca_trns,
'RCA': rand_proj_trns,
'ICA': ica_trns,
'LSA': lsa_trns
}, transformation_time