Code Editor for app.py

import streamlit as st
import numpy as np
import pandas as pd
import hdbscan
from sklearn.preprocessing import StandardScaler
import time

st.title("HDBSCAN Clustering Interface")

# Get query parameters from URL
query_params = st.experimental_get_query_params()

# Default values for parameters
default_min_cluster_size = 5
default_min_samples = 5
default_cluster_selection_epsilon = 0.0
default_n_samples = 1000

# Read parameters from query string or use defaults
min_cluster_size = int(query_params.get('min_cluster_size', [default_min_cluster_size])[0])
min_samples = int(query_params.get('min_samples', [default_min_samples])[0])
cluster_selection_epsilon = float(query_params.get('epsilon', [default_cluster_selection_epsilon])[0])
n_samples = int(query_params.get('n_samples', [default_n_samples])[0])

# Display current parameters
st.write("Current Parameters:")
st.write({
    'min_cluster_size': min_cluster_size,
    'min_samples': min_samples,
    'cluster_selection_epsilon': cluster_selection_epsilon,
    'n_samples': n_samples
})

# Generate sample data
@st.cache_data
def generate_sample_data(n_samples):
    np.random.seed(42)
    # Generate three clusters
    cluster1 = np.random.normal(0, 1, (n_samples // 3, 2))
    cluster2 = np.random.normal(4, 1.5, (n_samples // 3, 2))
    cluster3 = np.random.normal(-4, 1, (n_samples // 3, 2))
    X = np.vstack([cluster1, cluster2, cluster3])
    return X

# Generate and scale the data
X = generate_sample_data(n_samples)
scaler = StandardScaler()
X_scaled = scaler.fit_transform(X)

# Perform clustering
start_time = time.time()
clusterer = hdbscan.HDBSCAN(
    min_cluster_size=min_cluster_size,
    min_samples=min_samples,
    cluster_selection_epsilon=cluster_selection_epsilon
)
cluster_labels = clusterer.fit_predict(X_scaled)
end_time = time.time()

# Create DataFrame for visualization
df = pd.DataFrame(X, columns=['x', 'y'])
df['cluster'] = cluster_labels

# Display results
st.write(f"Time taken: {end_time - start_time:.2f} seconds")
st.write(f"Number of clusters found: {len(np.unique(cluster_labels[cluster_labels != -1]))}")
st.write(f"Number of noise points: {np.sum(cluster_labels == -1)}")

# Plot the results
st.write("Clustering Results:")
fig = plt.figure(figsize=(10, 6))
plt.scatter(df[df['cluster'] == -1]['x'], 
           df[df['cluster'] == -1]['y'],
           c='gray', label='Noise', alpha=0.5)

for label in np.unique(cluster_labels[cluster_labels != -1]):
    mask = df['cluster'] == label
    plt.scatter(df[mask]['x'], 
               df[mask]['y'],
               label=f'Cluster {label}',
               alpha=0.7)

plt.legend()
plt.title('HDBSCAN Clustering Results')
st.pyplot(fig)

# Instructions for URL parameters
st.markdown("""
### How to use URL parameters

You can modify the clustering by adding the following parameters to the URL:
- `min_cluster_size`: Minimum size of clusters (default: 5)
- `min_samples`: Number of samples in a neighborhood (default: 5)
- `epsilon`: Cluster selection epsilon (default: 0.0)
- `n_samples`: Number of samples to generate (default: 1000)

Example URL:
```
?min_cluster_size=10&min_samples=5&epsilon=0.5&n_samples=2000
```
""")