Copyright (c) 2015-2017 Sebastian Raschka
Note that the optional watermark extension is a small IPython notebook plugin that I developed to make the code reproducible. You can just skip the following line(s).
In [1]:
%load_ext watermark
%watermark -a 'Sebastian Raschka' -u -d -v -p numpy,pandas,matplotlib,scipy,sklearn
The use of watermark
is optional. You can install this IPython extension via "pip install watermark
". For more information, please see: https://github.com/rasbt/watermark.
In [2]:
from IPython.display import Image
%matplotlib inline
In [3]:
from sklearn.datasets import make_blobs
X, y = make_blobs(n_samples=150,
n_features=2,
centers=3,
cluster_std=0.5,
shuffle=True,
random_state=0)
In [4]:
import matplotlib.pyplot as plt
plt.scatter(X[:, 0], X[:, 1], c='white', marker='o', s=50)
plt.grid()
plt.tight_layout()
#plt.savefig('./figures/spheres.png', dpi=300)
plt.show()
In [5]:
from sklearn.cluster import KMeans
km = KMeans(n_clusters=3,
init='random',
n_init=10,
max_iter=300,
tol=1e-04,
random_state=0)
y_km = km.fit_predict(X)
plt.scatter(X[y_km == 0, 0],
X[y_km == 0, 1],
s=50,
c='lightgreen',
marker='s',
label='cluster 1')
plt.scatter(X[y_km == 1, 0],
X[y_km == 1, 1],
s=50,
c='orange',
marker='o',
label='cluster 2')
plt.scatter(X[y_km == 2, 0],
X[y_km == 2, 1],
s=50,
c='lightblue',
marker='v',
label='cluster 3')
plt.scatter(km.cluster_centers_[:, 0],
km.cluster_centers_[:, 1],
s=250,
marker='*',
c='red',
label='centroids')
plt.legend()
plt.grid()
plt.tight_layout()
#plt.savefig('./figures/centroids.png', dpi=300)
plt.show()
...
...
In [6]:
print('Distortion: %.2f' % km.inertia_)
In [7]:
distortions = []
for i in range(1, 11):
km = KMeans(n_clusters=i,
init='k-means++',
n_init=10,
max_iter=300,
random_state=0)
km.fit(X)
distortions.append(km.inertia_)
plt.plot(range(1, 11), distortions, marker='o')
plt.xlabel('Number of clusters')
plt.ylabel('Distortion')
plt.tight_layout()
#plt.savefig('./figures/elbow.png', dpi=300)
plt.show()
In [8]:
import numpy as np
from matplotlib import cm
from sklearn.metrics import silhouette_samples
km = KMeans(n_clusters=3,
init='k-means++',
n_init=10,
max_iter=300,
tol=1e-04,
random_state=0)
y_km = km.fit_predict(X)
cluster_labels = np.unique(y_km)
n_clusters = cluster_labels.shape[0]
silhouette_vals = silhouette_samples(X, y_km, metric='euclidean')
y_ax_lower, y_ax_upper = 0, 0
yticks = []
for i, c in enumerate(cluster_labels):
c_silhouette_vals = silhouette_vals[y_km == c]
c_silhouette_vals.sort()
y_ax_upper += len(c_silhouette_vals)
color = cm.jet(float(i) / n_clusters)
plt.barh(range(y_ax_lower, y_ax_upper), c_silhouette_vals, height=1.0,
edgecolor='none', color=color)
yticks.append((y_ax_lower + y_ax_upper) / 2.)
y_ax_lower += len(c_silhouette_vals)
silhouette_avg = np.mean(silhouette_vals)
plt.axvline(silhouette_avg, color="red", linestyle="--")
plt.yticks(yticks, cluster_labels + 1)
plt.ylabel('Cluster')
plt.xlabel('Silhouette coefficient')
plt.tight_layout()
# plt.savefig('./figures/silhouette.png', dpi=300)
plt.show()
Comparison to "bad" clustering:
In [9]:
km = KMeans(n_clusters=2,
init='k-means++',
n_init=10,
max_iter=300,
tol=1e-04,
random_state=0)
y_km = km.fit_predict(X)
plt.scatter(X[y_km == 0, 0],
X[y_km == 0, 1],
s=50,
c='lightgreen',
marker='s',
label='cluster 1')
plt.scatter(X[y_km == 1, 0],
X[y_km == 1, 1],
s=50,
c='orange',
marker='o',
label='cluster 2')
plt.scatter(km.cluster_centers_[:, 0], km.cluster_centers_[:, 1],
s=250, marker='*', c='red', label='centroids')
plt.legend()
plt.grid()
plt.tight_layout()
#plt.savefig('./figures/centroids_bad.png', dpi=300)
plt.show()
In [10]:
cluster_labels = np.unique(y_km)
n_clusters = cluster_labels.shape[0]
silhouette_vals = silhouette_samples(X, y_km, metric='euclidean')
y_ax_lower, y_ax_upper = 0, 0
yticks = []
for i, c in enumerate(cluster_labels):
c_silhouette_vals = silhouette_vals[y_km == c]
c_silhouette_vals.sort()
y_ax_upper += len(c_silhouette_vals)
color = cm.jet(float(i) / n_clusters)
plt.barh(range(y_ax_lower, y_ax_upper), c_silhouette_vals, height=1.0,
edgecolor='none', color=color)
yticks.append((y_ax_lower + y_ax_upper) / 2.)
y_ax_lower += len(c_silhouette_vals)
silhouette_avg = np.mean(silhouette_vals)
plt.axvline(silhouette_avg, color="red", linestyle="--")
plt.yticks(yticks, cluster_labels + 1)
plt.ylabel('Cluster')
plt.xlabel('Silhouette coefficient')
plt.tight_layout()
# plt.savefig('./figures/silhouette_bad.png', dpi=300)
plt.show()
In [11]:
Image(filename='./images/11_05.png', width=400)
Out[11]:
In [12]:
import pandas as pd
import numpy as np
np.random.seed(123)
variables = ['X', 'Y', 'Z']
labels = ['ID_0', 'ID_1', 'ID_2', 'ID_3', 'ID_4']
X = np.random.random_sample([5, 3])*10
df = pd.DataFrame(X, columns=variables, index=labels)
df
Out[12]:
In [13]:
from scipy.spatial.distance import pdist, squareform
row_dist = pd.DataFrame(squareform(pdist(df, metric='euclidean')),
columns=labels,
index=labels)
row_dist
Out[13]:
We can either pass a condensed distance matrix (upper triangular) from the pdist
function, or we can pass the "original" data array and define the metric='euclidean'
argument in linkage
. However, we should not pass the squareform distance matrix, which would yield different distance values although the overall clustering could be the same.
In [14]:
# 1. incorrect approach: Squareform distance matrix
from scipy.cluster.hierarchy import linkage
row_clusters = linkage(row_dist, method='complete', metric='euclidean')
pd.DataFrame(row_clusters,
columns=['row label 1', 'row label 2',
'distance', 'no. of items in clust.'],
index=['cluster %d' % (i + 1)
for i in range(row_clusters.shape[0])])
Out[14]:
In [15]:
# 2. correct approach: Condensed distance matrix
row_clusters = linkage(pdist(df, metric='euclidean'), method='complete')
pd.DataFrame(row_clusters,
columns=['row label 1', 'row label 2',
'distance', 'no. of items in clust.'],
index=['cluster %d' % (i + 1)
for i in range(row_clusters.shape[0])])
Out[15]:
In [16]:
# 3. correct approach: Input sample matrix
row_clusters = linkage(df.values, method='complete', metric='euclidean')
pd.DataFrame(row_clusters,
columns=['row label 1', 'row label 2',
'distance', 'no. of items in clust.'],
index=['cluster %d' % (i + 1)
for i in range(row_clusters.shape[0])])
Out[16]:
In [17]:
from scipy.cluster.hierarchy import dendrogram
# make dendrogram black (part 1/2)
# from scipy.cluster.hierarchy import set_link_color_palette
# set_link_color_palette(['black'])
row_dendr = dendrogram(row_clusters,
labels=labels,
# make dendrogram black (part 2/2)
# color_threshold=np.inf
)
plt.tight_layout()
plt.ylabel('Euclidean distance')
#plt.savefig('./figures/dendrogram.png', dpi=300,
# bbox_inches='tight')
plt.show()
In [18]:
# plot row dendrogram
fig = plt.figure(figsize=(8, 8), facecolor='white')
axd = fig.add_axes([0.09, 0.1, 0.2, 0.6])
# note: for matplotlib < v1.5.1, please use orientation='right'
row_dendr = dendrogram(row_clusters, orientation='left')
# reorder data with respect to clustering
df_rowclust = df.ix[row_dendr['leaves'][::-1]]
axd.set_xticks([])
axd.set_yticks([])
# remove axes spines from dendrogram
for i in axd.spines.values():
i.set_visible(False)
# plot heatmap
axm = fig.add_axes([0.23, 0.1, 0.6, 0.6]) # x-pos, y-pos, width, height
cax = axm.matshow(df_rowclust, interpolation='nearest', cmap='hot_r')
fig.colorbar(cax)
axm.set_xticklabels([''] + list(df_rowclust.columns))
axm.set_yticklabels([''] + list(df_rowclust.index))
# plt.savefig('./figures/heatmap.png', dpi=300)
plt.show()
In [19]:
from sklearn.cluster import AgglomerativeClustering
ac = AgglomerativeClustering(n_clusters=2,
affinity='euclidean',
linkage='complete')
labels = ac.fit_predict(X)
print('Cluster labels: %s' % labels)
In [20]:
Image(filename='./images/11_11.png', width=500)
Out[20]:
In [21]:
from sklearn.datasets import make_moons
X, y = make_moons(n_samples=200, noise=0.05, random_state=0)
plt.scatter(X[:, 0], X[:, 1])
plt.tight_layout()
# plt.savefig('./figures/moons.png', dpi=300)
plt.show()
K-means and hierarchical clustering:
In [22]:
f, (ax1, ax2) = plt.subplots(1, 2, figsize=(8, 3))
km = KMeans(n_clusters=2, random_state=0)
y_km = km.fit_predict(X)
ax1.scatter(X[y_km == 0, 0], X[y_km == 0, 1],
c='lightblue', marker='o', s=40, label='cluster 1')
ax1.scatter(X[y_km == 1, 0], X[y_km == 1, 1],
c='red', marker='s', s=40, label='cluster 2')
ax1.set_title('K-means clustering')
ac = AgglomerativeClustering(n_clusters=2,
affinity='euclidean',
linkage='complete')
y_ac = ac.fit_predict(X)
ax2.scatter(X[y_ac == 0, 0], X[y_ac == 0, 1], c='lightblue',
marker='o', s=40, label='cluster 1')
ax2.scatter(X[y_ac == 1, 0], X[y_ac == 1, 1], c='red',
marker='s', s=40, label='cluster 2')
ax2.set_title('Agglomerative clustering')
plt.legend()
plt.tight_layout()
#plt.savefig('./figures/kmeans_and_ac.png', dpi=300)
plt.show()
Density-based clustering:
In [23]:
from sklearn.cluster import DBSCAN
db = DBSCAN(eps=0.2, min_samples=5, metric='euclidean')
y_db = db.fit_predict(X)
plt.scatter(X[y_db == 0, 0], X[y_db == 0, 1],
c='lightblue', marker='o', s=40,
label='cluster 1')
plt.scatter(X[y_db == 1, 0], X[y_db == 1, 1],
c='red', marker='s', s=40,
label='cluster 2')
plt.legend()
plt.tight_layout()
#plt.savefig('./figures/moons_dbscan.png', dpi=300)
plt.show()
...