CODEWITHSUNDEEP
pythonscikit-learnscipyhierarchical-clustering
MARCO LAGALLA
MARCO LAGALLA

Reputation: 257

SciPy Dendrogram Plotting

I am playing with hierarchical documents clustering and actually my workflow is nearly this:

df = pandas.read_csv(file, delimiter='\t', index_col=0) # documents-terms matrix (very sparse)
dist_matrix = cosine_similarity(df)

linkage_matrix = ward(dist_matrix)
labels = fcluster(linkage_matrix, 5, criterion='maxclust')

Then I'm expecting to get 5 clusters, but when I plot the dendrogram

fig, ax = plt.subplots(figsize=(15, 20))  # set size
    ax = dendrogram(linkage_matrix, orientation="right")
    plt.tick_params( \
        axis='x',  # changes apply to the x-axis
        which='both',  # both major and minor ticks are affected
        bottom='off',  # ticks along the bottom edge are off
        top='off',  # ticks along the top edge are off
        labelbottom='off')

    plt.tight_layout()  # show plot with tight layout

    plt.savefig('ward_clusters.png', dpi=200)  # save figure as ward_clusters

I get the following graph

enter image description here

Based on the colors I can see 3 clusters, not 5! Am I misunderstanding the meaning of the dendrogram?

Upvotes: 1

Views: 3959

Answers (1)

llesoil
llesoil

Reputation: 101

  • First of all, if you just want to make 5 clusters, just use labels (the line with fcluster you did not use).

In labels : each point from your dataset is represented by a number. These numbers are the ids of your clusters.

  • If you want to use a dendogram, and plot 5 different clusters, then you'll have to "cut" your dendogram.

Draw a vertical line at x=5 (around 5), consider that each dendogram on the left is independent.

enter image description here

Artificially, you cut your dendogram into 5 parts (or 5 clusters).

To add some color to differentiate them, just adapt the following code (since you didn't provide your dataset, I used the iris dataset to show you one possible solution)

from scipy.cluster.hierarchy import *
from sklearn.datasets import load_iris
from sklearn.metrics.pairwise import cosine_similarity
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt

iris= load_iris()

data = iris['data']
df = pd.DataFrame(data, columns = iris['feature_names'])

# Somehow, we have something equivalent to work with now
dist_matrix = cosine_similarity(df)
linkage_matrix = ward(dist_matrix)

fig, ax = plt.subplots(figsize=(20, 10))

#here just put 5 for the color_threshold, which correspond to the position of the vertical line
ax = dendrogram(linkage_matrix, color_threshold =0.7)

plt.tick_params( \
    axis='x',
    which='both',
    bottom='off',
    top='off',
    labelbottom='off')

plt.show()

enter image description here

Upvotes: 2

Related Questions