# [ Clustering Algorithms ] ( CheatSheet )

Data Preparation

● Load Dataset: data = pd.read_csv('data.csv')

● Standardize Features: scaled_data = StandardScaler().fit_transform(data)
● Principal Component Analysis (PCA) for Dimensionality Reduction:
pca_data = PCA(n_components=2).fit_transform(scaled_data)
● Visualize Data Distribution: plt.scatter(data[:, 0], data[:, 1])
● Compute Pairwise Distance Matrix for Clustering: dist_matrix =
pairwise_distances(data, metric='euclidean')
● Handle Missing Values: data_filled =
SimpleImputer(strategy='mean').fit_transform(data)
● Create Dendrogram for Hierarchical Clustering: dendrogram =
sch.dendrogram(sch.linkage(data, method='ward'))
● Identify Optimal Number of Clusters via Elbow Method: inertia =
[KMeans(n_clusters=i).fit(data).inertia_ for i in range(1, 11)];
plt.plot(range(1, 11), inertia)
● Normalize Features for Clustering: normalized_data = normalize(data)
● Use TSNE for Visualization of High-dimensional Data: tsne_data =
TSNE(n_components=2).fit_transform(data)

K-means Clustering

● Perform K-means Clustering: kmeans = KMeans(n_clusters=3).fit(data)

● Predict Cluster Labels: labels = kmeans.predict(data)
● Visualize Clusters: plt.scatter(data[:, 0], data[:, 1], c=labels)
● Determine Centroids: centroids = kmeans.cluster_centers_
● Silhouette Score for Model Evaluation: silhouette_score(data, labels)
● Initializing K-means with Smart Start (k-means++): kmeans =
KMeans(n_clusters=3, init='k-means++').fit(data)
● Mini-Batch K-means for Large Datasets: minibatch_kmeans =
MiniBatchKMeans(n_clusters=3).fit(data)
● Find Optimal K Using the Silhouette Method: silhouette_optimal_k(data)
● Elbow Method Visualization for Optimal K: plot_elbow_method(data)
● Assign New Data Points to Existing Clusters:new_labels =
kmeans.predict(new_data)
● Iterative Training to Refine Centroids: kmeans.fit(data);
kmeans.partial_fit(more_data)

By: Waleed Mousa

 ● Visualize Cluster Centers on 2D Plot:
plt.scatter(kmeans.cluster_centers_[:, 0], kmeans.cluster_centers_[:, 1],
s=300, c='red')
● Calculate Within-cluster Sum of Squares (WSS): wss = kmeans.inertia_
● K-means Clustering with Specific Random State for Reproducibility: kmeans
= KMeans(n_clusters=3, random_state=42).fit(data)
● Use K-means for Color Quantization in Images: quantized_img =
quantize_colors(image_data, n_colors=8)

Hierarchical Clustering

● Perform Agglomerative Hierarchical Clustering: model =

AgglomerativeClustering(n_clusters=3).fit(data)
● Extract Cluster Labels: labels = model.labels_
● Plot Dendrogram Function: plot_dendrogram(model, truncate_mode='level',
p=3)
● Cophenetic Correlation Coefficient: c, coph_dists =
cophenet(sch.linkage(data, 'ward'), pdist(data))
● Agglomerative Hierarchical Clustering with Different Linkage Criteria:
model = AgglomerativeClustering(n_clusters=3,
linkage='average').fit(data)
● Scikit-learn's Convenience Function for Ward's Method: ward =
AgglomerativeClustering(n_clusters=3, linkage='ward').fit(data)
● Generate Dendrogram from Linkage Matrix: linkage_matrix =
ward(children=model.children_); dendrogram(linkage_matrix)
● Cutting the Dendrogram to Form Clusters: labels =
fcluster(linkage_matrix, t=3, criterion='maxclust')
● Creating Custom Distance Matrix for Agglomerative Clustering: model =
AgglomerativeClustering(n_clusters=3, affinity='precomputed',
linkage='complete').fit(custom_distance_matrix)
● Interactive Dendrogram Plotting with Plotly:
plot_dendrogram_plotly(linkage_matrix)
● Evaluate Model Using Davies-Bouldin Index: db_index =
davies_bouldin_score(data, model.labels_)
● Use SciPy for More Detailed Dendrogram Customization:
dendrogram(sch.linkage(data, method='ward'), color_threshold=1)
● Dynamic Thresholding for Cluster Formation in Hierarchical Clustering:
dynamic_labels = dynamic_threshold_clustering(linkage_matrix)

DBSCAN

● Apply DBSCAN Clustering: dbscan = DBSCAN(eps=0.3,

min_samples=10).fit(data)

By: Waleed Mousa

 ● Cluster Labels from DBSCAN: labels = dbscan.labels_
● Identify Core Samples: core_samples_mask = np.zeros_like(dbscan.labels_,
dtype=bool); core_samples_mask[dbscan.core_sample_indices_] = True
● Visualize DBSCAN Clusters: plt.scatter(data[:, 0], data[:, 1], c=labels)
● Adjusting eps and min_samples for Density Variation: dbscan =
DBSCAN(eps=0.5, min_samples=15).fit(data)
● Handling Noise Points Identified by DBSCAN: noise = data[dbscan.labels_
== -1]
● Cluster Core Samples Extraction and Visualization: core_samples =
data[dbscan.core_sample_indices_]
● Evaluate Clustering with Silhouette Score: silhouette_score(data,
dbscan.labels_)
● Visualize DBSCAN Clusters with Matplotlib: plot_dbscan_clusters(data,
dbscan.labels_, dbscan.core_sample_indices_)
● Optimize DBSCAN Parameters with Grid Search: optimal_params =
optimize_dbscan(data)
● Use DBSCAN for Anomaly Detection: anomalies = data[dbscan.labels_ == -1]
● Scaling Features for Improved DBSCAN Performance: scaled_data =
StandardScaler().fit_transform(data); dbscan.fit(scaled_data)
● Calculate Number of Clusters in DBSCAN: n_clusters_ =
len(set(dbscan.labels_)) - (1 if -1 in dbscan.labels_ else 0)
● Visualize Spatial Clusters with Folium for Geospatial Data:
plot_folium_map_geospatial(data, dbscan.labels_)

Advanced Clustering Algorithms

● Spectral Clustering: spectral_labels =

SpectralClustering(n_clusters=3).fit_predict(data)
● Mean Shift Clustering: meanShift = MeanShift().fit(data); labels =
meanShift.labels_
● Affinity Propagation: affinity = AffinityPropagation().fit(data);
cluster_centers_indices = affinity.cluster_centers_indices_
● Hierarchical Density-Based Spatial Clustering (HDBSCAN): hdbscan_model =
hdbscan.HDBSCAN(min_cluster_size=10).fit(data); labels =
hdbscan_model.labels_
● Gaussian Mixture Models for Soft Clustering: gmm =
GaussianMixture(n_components=3).fit(data); labels = gmm.predict(data)
● OPTICS Clustering: optics = OPTICS(min_samples=10, xi=.05,
min_cluster_size=.05).fit(data); labels = optics.labels_
● Evaluate Clustering with Adjusted Rand Index:
adjusted_rand_score(true_labels, predicted_labels)

By: Waleed Mousa

 ● Clustering Based on Graph Connectivity (Agglomerative): connectivity =
kneighbors_graph(data, n_neighbors=10, include_self=False); ward =
AgglomerativeClustering(connectivity=connectivity); labels =
ward.fit_predict(data)
● BIRCH for Large Datasets: birch = BIRCH(n_clusters=3).fit(data); labels =
birch.predict(data)
● CURE Clustering Implementation with PyCaret: from pycaret.clustering
import *; cure_model = create_model('cure', num_clusters=3, data=data)
● Using Silhouette Plots to Evaluate Clustering Quality:
plot_silhouette(data, labels)
● Cluster Validation Using the Davies-Bouldin Index:
davies_bouldin_score(data, labels)

Optimization Strategies

● Grid Search for Optimal Parameters in K-means: param_grid =

{'n_clusters': range(1, 11)}; grid_search = GridSearchCV(KMeans(),
param_grid); grid_search.fit(data)
● Using the Gap Statistic to Determine the Number of Clusters:
gap_statistic, opt_k = optimalK(data, nrefs=3, maxClusters=10)
● Auto-Scaling Features Based on Clustering Tendency: scaler =
autoscale_based_on_clustering_tendency(data)
● Parallel Coordinate Plot for Cluster Visualization:
pd.plotting.parallel_coordinates(data.assign(cluster=labels), 'cluster')
● Cluster Stability Evaluation via Bootstrapping: stability =
bootstrap_stability(data, KMeans(n_clusters=3), n_bootstraps=10)
● Elbow Method with Inertia and Silhouette Analysis Combined:
evaluate_clustering_elbow_silhouette(data, max_clusters=10)

Specialized Clustering Applications

● Temporal or Sequential Data Clustering (e.g., Time Series):

ts_cluster_labels =
TimeSeriesKMeans(n_clusters=3).fit_predict(time_series_data)
● Clustering Geospatial Data: geo_cluster_labels = DBSCAN(eps=0.1,
min_samples=5).fit_predict(geo_data[['latitude', 'longitude']])
● Image Segmentation Using Clustering: segmented_image =
KMeans(n_clusters=3).fit_predict(image_pixels)
● Text Clustering for Document Categorization: text_cluster_labels =
MiniBatchKMeans(n_clusters=5).fit_predict(tfidf_matrix)

By: Waleed Mousa

 ● Clustering for Anomaly Detection: anomaly_labels =
IsolationForest().fit_predict(data)
● Clustering in Bioinformatics (e.g., Gene Expression Data):
gene_cluster_labels =
AgglomerativeClustering().fit_predict(gene_expression_data)

Integrative Approaches and Advanced Techniques

● Consensus Clustering for Stability and Robustness: consensus_labels =

consensus_cluster(data, KMeans(), n_clusters_range=[2,10],
bootstrap_samples=100)
● Feature Learning with Clustering (e.g., Autoencoders): encoded_features
= Autoencoder().fit_transform(data); ae_cluster_labels =
KMeans(n_clusters=3).fit_predict(encoded_features)
● Cluster Ensembles for Improved Performance: ensemble_labels =
ClusterEnsembles(hyperparameters, data)
● Use of Clustering for Dimension Reduction: reduced_data =
clustering_based_dimension_reduction(data, n_clusters=10)
● Integrating Clustering with Classification for Semi-supervised Learning:
semi_labels = semi_supervised_learning_with_clustering(data,
partial_labels)
● Leveraging Graph-based Clustering for Complex Networks:
graph_cluster_labels =
SpectralClustering(n_clusters=3).fit_predict(adjacency_matrix)
● Multi-view Clustering for Integrating Different Types of Data:
multi_view_labels = MultiViewClustering().fit_predict([view1_data,
view2_data])
● Interactive Clustering for User-guided Analysis: interactive_clusters =
interactive_clustering(data, initial_guesses)
● Utilizing Clustering for Data Cleaning and Preprocessing: clean_data =
data_cleaning_with_clustering(data)
● Hierarchical Clustering for Large Datasets via BIRCH:
large_scale_cluster_labels = BIRCH(threshold=0.5,
n_clusters=None).fit_predict(large_data)
● Spatial Clustering for Location Data Optimization: location_clusters =
OPTICS(min_samples=50, xi=0.05,
min_cluster_size=0.1).fit_predict(location_data)
● Integrating Clustering with Reinforcement Learning for Dynamic
Environments: dynamic_cluster_labels =
reinforcement_learning_with_clustering(state_data)

By: Waleed Mousa

● Clustering for Recommender Systems (User or Item Clustering):
recommender_clusters =
KMeans(n_clusters=10).fit_predict(user_feature_matrix)
● Deep Clustering for Unsupervised Feature Learning: deep_cluster_labels =
DeepClustering().fit_predict(data)
● Clustering Validation in Multi-dimensional Datasets: validation_scores =
multidimensional_clustering_validation(data, cluster_labels)

By: Waleed Mousa