def test_ax(self):
np.random.seed(0)
clf = KMeans()
cluster_labels = clf.fit_predict(self.X)
plot_silhouette(self.X, cluster_labels)
fig, ax = plt.subplots(1, 1)
out_ax = plot_silhouette(self.X, cluster_labels)
assert ax is not out_ax
out_ax = plot_silhouette(self.X, cluster_labels, ax=ax)
assert ax is out_ax
def test_array_like(self):
plot_silhouette(self.X.tolist(), self.y.tolist())
plot_silhouette(self.X.tolist(), convert_labels_into_string(self.y))
def test_cmap(self):
np.random.seed(0)
clf = KMeans()
cluster_labels = clf.fit_predict(self.X)
plot_silhouette(self.X, cluster_labels, cmap='Spectral')
plot_silhouette(self.X, cluster_labels, cmap=plt.cm.Spectral)
def test_string_classes(self):
np.random.seed(0)
clf = KMeans()
cluster_labels = clf.fit_predict(self.X)
plot_silhouette(self.X, convert_labels_into_string(cluster_labels))
def test_plot_silhouette(self):
np.random.seed(0)
clf = KMeans()
cluster_labels = clf.fit_predict(self.X)
plot_silhouette(self.X, cluster_labels)
def cluster_and_plot_pca(df,
cluster_range=np.arange(2, 9),
ClusterAlgorithm=KMeans,
cluster_kwargs=cluster_kwargs,
show=True):
'''An unsupervised learning approach to visualizing the number of clusters
that are appropriate for a given dataset `df`. If using another
ClusterAlgorithm than KMeans it must accept the kwarg n_clusters.
Data should have no missing values and all columns should have already
been transformed into numeric datatypes (i.e. converting categorical
features into one-hot encoded vectors).
Example:
from sklearn.cluster import MiniBatchKMeans
clust_kwargs = dict(random_state=77)
cluster_and_plot_pca(wine_df,
ClusterAlgorithm=MiniBatchKMeans,
cluster_kwargs=clust_kwargs)
ARGS:
df <pd.DataFrame>: DataFrame with all numeric data types and no
missing values.
KWARGS:
cluster_range <list> or <array>: List of integers signifying the number
of clusters to test in sequence.
ClusterAlgorithm <sklearn.cluster>: Currently supports KMeans
and MiniBatchKMeans.
cluster_kwargs <dict>: kwargs for ClusterAlgorithm, not including
num_clusters.
RETURNS:
None. Plots are printed to the console.
'''
pca = PCA(n_components=2, random_state=777)
X_pca = pd.DataFrame(pca.fit(df).transform(df))
for n_clusters in cluster_range:
clust_col = "clusters_"+str(n_clusters)
# perform kmeans or some other clustering algorithm
clust = ClusterAlgorithm(n_clusters=n_clusters, **cluster_kwargs)
clust.fit(df)
X_pca[clust_col] = clust.labels_
# plot PCA with segments
fig, ax = plt.subplots()
for _, x in X_pca.groupby(clust_col):
ax.scatter(x[0], x[1], label=str(_), alpha=.6)
ax.grid(alpha=.4)
sns.despine()
ax.set_title(f"Clusters={n_clusters} Projected on 2D Principal Components",
size=12)
ax.set_xlabel("PC1")
ax.set_ylabel("PC2")
ax.legend(loc="best")
# plot silhouette plot
plot_silhouette(X_pca, clust.labels_)
ax = plt.gca()
ax.grid(alpha=.4)
sns.despine()
ax.set_title(f"Silhouette Plot Clusters={n_clusters}", size=12)
if show:
plt.show()
# return the clustering algorithm for predictions
return clust
