def clustering(fname="clustering.png"):
# Create side-by-side axes grid
_, axes = plt.subplots(ncols=2, figsize=(18,6))
X, y = make_blobs(centers=7)
# Add K-Elbow to the left
oz = KElbowVisualizer(MiniBatchKMeans(), k=(3,12), ax=axes[0])
oz.fit(X, y)
oz.finalize()
# Add SilhouetteVisualizer to the right
oz = SilhouetteVisualizer(Birch(n_clusters=5), ax=axes[1])
oz.fit(X, y)
oz.finalize()
# Save figure
path = os.path.join(FIGURES, fname)
plt.tight_layout()
plt.savefig(path)
def clustering(fname="clustering.png"):
# Create side-by-side axes grid
_, axes = plt.subplots(ncols=2, figsize=(18, 6))
X, y = make_blobs(centers=7)
# Add K-Elbow to the left
oz = KElbowVisualizer(MiniBatchKMeans(), k=(3, 12), ax=axes[0])
oz.fit(X, y)
oz.finalize()
# Add SilhouetteVisualizer to the right
oz = SilhouetteVisualizer(Birch(n_clusters=5), ax=axes[1])
oz.fit(X, y)
oz.finalize()
# Save figure
path = os.path.join(FIGURES, fname)
plt.tight_layout()
plt.savefig(path)
if len(feature_names) > 10:
k_range = range(2, len(feature_names) + 1)
else:
k_range = range(2, 2 * len(feature_names) + 1) # try bigger range for the small-d dataset
# find optimal k with elbow method
for metric in ('distortion', 'silhouette', 'calinski_harabasz'):
print("# Optimizing k for " + label + " with " + metric)
model = cluster.KMeans(precompute_distances=True, random_state=SEED, n_jobs=-1)
try:
visualizer = KElbowVisualizer(model, k=k_range, metric=metric, locate_elbow=True)
except:
visualizer = KElbowVisualizer(model, k=k_range, metric=metric, locate_elbow=False)
visualizer.fit(X)
visualizer.ax.xaxis.set_major_locator(MaxNLocator(integer=True))
visualizer.finalize()
plt.savefig(path.join(PLOT_DIR, abbrev + "_km-rp_elbow_" + metric + ".png"), bbox_inches='tight')
visualizer.show()
plt.close()
# predict best clusters
print("# Clustering " + label)
model = cluster.KMeans(n_clusters=best_k, precompute_distances=True, random_state=SEED, n_jobs=-1)
start_time = time.perf_counter()
y_pred = model.fit_predict(X)
run_time = time.perf_counter() - start_time
print(label + ": run time = " + str(run_time))
print(label + ": iterations until convergence = " + str(model.n_iter_))
df = X.assign(cluster=y_pred)
df.to_pickle(path.join(PKL_DIR, abbrev + "_km-rp.pickle"))
