# -*- coding: utf-8 -*- """01_clustering_methods.ipynb Automatically generated by Colaboratory. Original file is located at https://colab.research.google.com/drive/1mqAGInsaItbKYVUlP9muYz3fpdGBWFz5 """ import numpy as np import pandas as pd import matplotlib.pyplot as plt import seaborn as sns import sklearn.cluster as cluster import colormaps as cmaps import opinionated plt.style.use("opinionated_rc") from opinionated.core import download_googlefont download_googlefont('Quicksand', add_to_cache=True) plt.rc('font', family='Quicksand') #wget https://github.com/scikit-learn-contrib/hdbscan/raw/master/notebooks/clusterable_data.npy #!wget https://github.com/mwaskom/seaborn-data/raw/master/penguins.csv import requests # URLs of the files to download clusterable_data_url = "https://github.com/scikit-learn-contrib/hdbscan/raw/master/notebooks/clusterable_data.npy" penguins_csv_url = "https://github.com/mwaskom/seaborn-data/raw/master/penguins.csv" # Function to download and save a file from a URL def download_file(url, local_filename): with requests.get(url, stream=True) as r: r.raise_for_status() with open(local_filename, 'wb') as f: for chunk in r.iter_content(chunk_size=8192): f.write(chunk) # Download the files download_file(clusterable_data_url, "clusterable_data.npy") download_file(penguins_csv_url, "penguins.csv") print("Files downloaded successfully.") hdbscan_example_data = np.load('clusterable_data.npy') penguins_dataset = pd.read_csv('penguins.csv')[['bill_length_mm','bill_depth_mm','flipper_length_mm']].dropna().values from sklearn.preprocessing import StandardScaler scaler = StandardScaler() penguins_dataset_standardized = scaler.fit_transform(penguins_dataset) import gradio as gr import numpy as np import matplotlib.pyplot as plt from sklearn.datasets import make_blobs, make_moons, load_iris import seaborn as sns import pandas as pd import matplotlib.colors as mcolors from sklearn.cluster import KMeans from sklearn.cluster import AgglomerativeClustering from sklearn.mixture import GaussianMixture import hdbscan import genieclust # Pre-defined datasets blobs_X, _ = make_blobs(n_samples=300, centers=4, cluster_std=0.60, random_state=0) moons_X, _ = make_moons(n_samples=300, noise=0.05, random_state=0) # Penguins dataset (3D example) # For the purpose of this example, let's simulate the Penguins dataset with iris for simplicity iris_X, _ = load_iris(return_X_y=True) # Assuming iris_X to be a placeholder for the Penguins dataset with numerical features datasets = { "Blobs": blobs_X, "Moons": moons_X, "Penguins": penguins_dataset_standardized, # Placeholder for Penguins dataset "hDBSCAN sample": hdbscan_example_data } # Function for plotting the unclustered dataset def plot_unclustered(dataset_name): X = datasets[dataset_name] # Fetch dataset from the dictionary # Check if the dataset has more than 2 dimensions if X.shape[1] > 2: # Convert dataset to DataFrame for seaborn pairplot df = pd.DataFrame(X) fig = sns.pairplot(df, plot_kws={'color': 'grey','alpha':0.7}, diag_kws={'color': 'grey'}).fig else: fig, ax = plt.subplots(figsize=(8, 6)) ax.scatter(X[:, 0], X[:, 1], color='gray', marker='.',alpha=.7) ax.set_xlabel("Feature 1") ax.set_ylabel("Feature 2") ax.grid(True) plt.tight_layout() plt.close(fig) return fig def plot_clustered(dataset_name, clustering_method, kmeans_n_clusters, agg_n_clusters, agg_linkage, gmm_n_clusters, covariance_type, genie_n_clusters, gini_threshold, M,hdbscan_min_cluster_size, hdbscan_min_samples): X = datasets[dataset_name] # Determine the clustering method and fit the model accordingly if clustering_method == "K-Means": model = KMeans(n_clusters=kmeans_n_clusters) model.fit(X) labels = model.labels_ # For K-Means, labels are in .labels_ elif clustering_method == "Agglomerative": model = AgglomerativeClustering(n_clusters=agg_n_clusters, linkage=agg_linkage) model.fit(X) labels = model.labels_ # For Agglomerative Clustering, labels are in .labels_ elif clustering_method == "Gaussian Mixture": model = GaussianMixture(n_components=gmm_n_clusters, covariance_type=covariance_type) model.fit(X) labels = model.predict(X) # For Gaussian Mixture, use .predict() to get labels elif clustering_method == "Genie": model = genieclust.Genie(n_clusters=genie_n_clusters, gini_threshold=gini_threshold, M=M) labels = model.fit_predict(X) # GenieClust uses fit_predict directly for both fitting and label prediction elif clustering_method == "h-DBSCAN": clusterer = hdbscan.HDBSCAN(min_cluster_size=hdbscan_min_cluster_size, min_samples=hdbscan_min_samples).fit(X) labels = clusterer.labels_ n_clusters= len(np.unique([x for x in labels if x >= 0])) if n_clusters <= 10: original_cmap = cmaps.greenorange_12 colors = original_cmap([x for x in range(n_clusters)]) # Create a new listed colormap with the extracted colors new_cmap = mcolors.ListedColormap(colors) else: new_cmap = cmaps.cet_g_bw_minc cluster_colors = [new_cmap(x) if x >= 0 else (0.5, 0.5, 0.5) for x in labels] # Check if the dataset has more than 2 dimensions if X.shape[1] > 2: # Convert dataset to DataFrame for seaborn pairplot df = pd.DataFrame(X) # df['cluster'] = labels # fig = sns.pairplot(df, color = cluster_colors, cmap=new_cmap).fig # Create bins for each variable n_bins = 10 bins = {column: np.linspace(df[column].min(), df[column].max(), n_bins+1) for column in df.columns} # Create a figure and axes n = len(df.columns) fig, axes = plt.subplots(nrows=n, ncols=n, figsize=(n*2.3, n*2.3)) for i in range(n): for j in range(n): ax = axes[i, j] ax.grid(True, which='both', linestyle='--', linewidth=0.5) if i != j: ax.scatter(df[df.columns[j]], df[df.columns[i]], c=cluster_colors, alpha=0.8, marker='o',s = 10) else: # Diagonal - Stacked Bar Charts data = df[df.columns[i]] counts = np.zeros((n_bins, n_clusters)) for cluster in range(n_clusters): cluster_data = data[labels == cluster] hist, _ = np.histogram(cluster_data, bins=bins[df.columns[i]]) counts[:, cluster] = hist for cluster in range(n_clusters): ax.bar(range(n_bins), counts[:, cluster], width=1, align='center', bottom=np.sum(counts[:, :cluster], axis=1), color=cluster_colors[list(labels).index(cluster)] ) # Explicit axis lines at the bottom and left ax.spines['top'].set_visible(False) ax.spines['right'].set_visible(False) ax.spines['bottom'].set_visible(True) ax.spines['left'].set_visible(True) # Hide axis marks for inner plots and adjust label size if i < n - 1: ax.tick_params(labelbottom=False) # Hide x-axis labels for all but bottom row else: ax.tick_params(axis='x', labelsize=8) # Smaller labels for x-axis if j > 0: ax.tick_params(labelleft=False) # Hide y-axis labels for all but first column else: ax.tick_params(axis='y', labelsize=8) # Smaller labels for y-axis # Set labels for outer plots only if i == n - 1: ax.set_xlabel(df.columns[j], rotation=0, fontsize=12) if j == 0: ax.set_ylabel(df.columns[i], fontsize=12) else: fig, ax = plt.subplots(figsize=(8, 6)) ax.scatter(X[:, 0], X[:, 1], c=cluster_colors, marker='.') ax.grid(True) plt.tight_layout() plt.close(fig) return fig intro_md = """ # Cluster-algorithm-explorer _by [Max Noichl](https://homepage.univie.ac.at/maximilian.noichl/), for the clustering & data-visualization-workshop, Bremen, 2024_ Below you can test a number of clustering-algorithms on several easier and harder datasets. """ # Gradio interface setup remains the same with gr.Blocks(theme=gr.themes.Monochrome()) as demo: with gr.Column(): gr.Markdown(intro_md) with gr.Row(): with gr.Column(): gr.Markdown("# Choose your dataset:") dataset_dropdown = gr.Dropdown(label="Select a dataset", choices=list(datasets.keys()), value="Blobs") gr.Markdown("# Choose your Clustering algorithm & Parameters:") # Update the dropdown for clustering method to include "Genie" clustering_method_dropdown = gr.Dropdown(label="Select a clustering method", choices=["K-Means", "Agglomerative", "Gaussian Mixture", "Genie", "h-DBSCAN"], value="K-Means") # K-Means parameters with gr.Group(visible=True) as kmeans_params_group: kmeans_n_clusters_slider = gr.Slider(minimum=2, maximum=10, step=1, label="Number of Clusters (K-Means)", value=4) # Agglomerative Clustering parameters with gr.Group(visible=False) as agglomerative_params_group: agg_n_clusters_slider = gr.Slider(minimum=2, maximum=10, step=1, label="Number of Clusters (Agglomerative)", value=4) agg_linkage_dropdown = gr.Dropdown(label="Linkage Type", choices=["ward", "complete", "average", "single"], value="ward") # Gaussian Mixture Model parameters with gr.Group(visible=False) as gmm_params_group: gmm_n_clusters_slider = gr.Slider(minimum=2, maximum=10, step=1, label="Number of Components (GMM)", value=4) covariance_type_dropdown = gr.Dropdown(label="Covariance Type", choices=["full", "tied", "diag", "spherical"], value="full") # GenieClust parameters with gr.Group(visible=False) as genie_params_group: genie_n_clusters_slider = gr.Slider(minimum=2, maximum=10, step=1, label="Number of Clusters (Genie)", value=4) gini_threshold_slider = gr.Slider(minimum=0.0, maximum=1.05, step=0.05, label="Gini Threshold (Genie)", value=.3) M_slider = gr.Slider(minimum=0.5, maximum=2.0, step=0.1, label="M Parameter (Genie)", value=1.0) with gr.Group(visible=False) as hdbscan_params_group: hdbscan_min_cluster_size = gr.Slider(minimum=2, maximum=200, step=1, label="Minimal Cluster Size (hDBSCAN)", value=10) hdbscan_min_samples = gr.Slider(minimum=2, maximum=200, step=1, label="Min. Samples (hDBSCAN)", value=10) # Update the function that changes visible parameter groups based on selected clustering method def update_method_params(clustering_method): return { kmeans_params_group: gr.Group(visible=clustering_method == "K-Means"), agglomerative_params_group: gr.Group(visible=clustering_method == "Agglomerative"), gmm_params_group: gr.Group(visible=clustering_method == "Gaussian Mixture"), genie_params_group: gr.Group(visible=clustering_method == "Genie"), hdbscan_params_group: gr.Group(visible=clustering_method == "h-DBSCAN"), } clustering_method_dropdown.change(update_method_params, inputs=[clustering_method_dropdown], outputs=[kmeans_params_group, agglomerative_params_group, gmm_params_group, genie_params_group,hdbscan_params_group]) button = gr.Button("Run Clustering!") with gr.Column(): unclustered_plot_output = gr.Plot(label=None) clustered_plot_output = gr.Plot(label=None) dataset_dropdown.change(plot_unclustered, inputs=[dataset_dropdown], outputs=[unclustered_plot_output]) demo.load(plot_unclustered, inputs=[dataset_dropdown], outputs=[unclustered_plot_output]) # Update the button click event to include new parameters for GenieClust button.click( plot_clustered, inputs=[ dataset_dropdown, clustering_method_dropdown, kmeans_n_clusters_slider, agg_n_clusters_slider, agg_linkage_dropdown, gmm_n_clusters_slider, covariance_type_dropdown, genie_n_clusters_slider, # Add Genie parameters gini_threshold_slider, M_slider, hdbscan_min_cluster_size, hdbscan_min_samples ], outputs=[clustered_plot_output] ) if __name__ == "__main__": demo.launch(debug=True)