tp_apprentissage/real_world_synthese.py

import os
import csv
from sklearn.cluster import KMeans, AgglomerativeClustering, DBSCAN
from sklearn.metrics import davies_bouldin_score, silhouette_score
from hdbscan import HDBSCAN
from graphs import bars_plot, clustering_plot
from scipy.io import arff
from sklearn.decomposition import PCA
from random import seed

# For making all the process reproductible
seed(10)

# Chargement des données contenues dans le fichier
data = arff.loadarff(f="real-world/iris.arff")[0]
print(type(data))
# Extraction des points contenus dans la donnée IRIS et les labels sous forme de texte
iris_data = [[c[0],c[1],c[2],c[3]] for c in data]
labels = [c[4].decode("utf-8") for c in data]
# Transformation des labels textuels en entiers
labels_as_integers = []
for i in range(len(labels)):
    # Iris-setosa     ==> 0
    # Iris-virginica  ==> 1
    # Iris-versicolor ==> 2
    if labels[i]=="Iris-setosa": labels_as_integers +=[0]
    elif labels[i]=="Iris-virginica": labels_as_integers +=[1]
    else: labels_as_integers += [2]

print(labels)
print(labels_as_integers)

# Réduction de dimension
modele_pca = PCA(n_components=2)
iris_dimension_reduced = modele_pca.fit_transform(X=iris_data)

modele = KMeans(n_clusters=3, init="k-means++")
predictions = modele.fit_predict(X=iris_dimension_reduced)
clustering_plot(iris_dimension_reduced[:,0],iris_dimension_reduced[:,1],predictions,
                fig_title="Iris clustering with KMeans", xlabel="1ere composante IRIS",
                ylabel="2eme composante IRIS", output="real_synthese/iris_kmeans.png")

linkages = ["single", "average", "complete", "ward"]
for linkage in linkages:
    modele = AgglomerativeClustering(n_clusters=3, linkage=linkage)
    predictions = modele.fit_predict(X=iris_dimension_reduced)
    clustering_plot(iris_dimension_reduced[:, 0], iris_dimension_reduced[:, 1], predictions,
                    fig_title="Iris clustering with AgglomerativeClustering and linkage=%s"%(linkage), xlabel="1ere composante IRIS",
                    ylabel="2eme composante IRIS", output="real_synthese/iris_agglclst_with_linkage_%s.png"%(linkage))

modele = DBSCAN(eps=0.19,min_samples=5)
predictions = modele.fit_predict(X=iris_dimension_reduced)
clustering_plot(iris_dimension_reduced[:,0],iris_dimension_reduced[:,1],predictions,
                fig_title="Iris clustering with DBSCANs", xlabel="1ere composante IRIS",
                ylabel="2eme composante IRIS", output="real_synthese/iris_dbscan.png")

modele = HDBSCAN(cluster_selection_epsilon=0.19,min_samples=5)
predictions = modele.fit_predict(X=iris_dimension_reduced)
clustering_plot(iris_dimension_reduced[:,0],iris_dimension_reduced[:,1],predictions,
                fig_title="Iris clustering with HDBSCAN", xlabel="1ere composante IRIS",
                ylabel="2eme composante IRIS", output="real_synthese/iris_hdbscan.png")