Update app.py

app.py

@@ -1,57 +1,104 @@
-import gradio as gr
 import pandas as pd
 import numpy as np
-import joblib
-import seaborn as sns
 import matplotlib.pyplot as plt
+import seaborn as sns
+from sklearn.impute import SimpleImputer
+from sklearn.preprocessing import StandardScaler
+from sklearn.manifold import TSNE
+import gradio as gr
+import os
+
+# --- Load Precomputed Training t-SNE Data ---
+training_tsne_path = "training_tsne.xlsx"  # Ensure this file is uploaded to Hugging Face
+
+if os.path.exists(training_tsne_path):
+    training_tsne_df = pd.read_excel(training_tsne_path)
+    print("Training t-SNE data loaded successfully!")
+else:
+    training_tsne_df = None
+    print("Error: training_tsne.xlsx not found. Please upload it.")
+
+# Feature columns used for log transformation
+columns_to_transform = ['Mn', 'Co', 'Ni', 'Cu', 'Zn', 'As', 'Se', 'Mo', 'Ag', 'Cd', 'Sb', 'Te', 'Tl', 'Pb', 'Bi']
+
+# Define global preprocessing objects (to ensure test data uses the same transformation as training)
+imputer = SimpleImputer(strategy='mean')
+scaler = StandardScaler()
+
+def process_test_data(test_file):
+    """
+    Load and preprocess test data. 
+    Assumes test data has the same feature columns as training.
+    """
+    test_df = pd.read_excel(test_file.name)
+    
+    # Ensure all required columns exist
+    missing_columns = [col for col in columns_to_transform if col not in test_df.columns]
+    if missing_columns:
+        return f"Error: Missing columns in the uploaded data: {', '.join(missing_columns)}"
+    
+    # Apply log transformation
+    test_df[columns_to_transform] = np.log1p(test_df[columns_to_transform])
+
+    # Apply the same imputer and scaler as training (fit on test data)
+    X_test_imputed = imputer.fit_transform(test_df[columns_to_transform])
+    X_test_scaled = scaler.fit_transform(X_test_imputed)
+
+    return X_test_scaled
+
+def visualize_tsne_interface(test_file):
+    """
+    Compute the t-SNE embedding for the test data,
+    load the saved training t-SNE data,
+    and overlay the test points on the training plot.
+    """
+    if training_tsne_df is None:
+        return "Error: Training t-SNE data not found. Please upload `training_tsne.xlsx`."
+
+    # Preprocess the uploaded test data
+    X_test_scaled = process_test_data(test_file)
+    if isinstance(X_test_scaled, str):
+        return X_test_scaled  # Return error message if preprocessing failed
+
+    # Compute t-SNE embedding for test data
+    tsne_test = TSNE(n_components=2, random_state=42)
+    X_tsne_test = tsne_test.fit_transform(X_test_scaled)
+
+    # Create the plot
+    plt.figure(figsize=(10, 7))
+
+    # Plot training data t-SNE with color-coded labels
+    sns.scatterplot(data=training_tsne_df,
+                    x='x',
+                    y='y',
+                    hue='label',
+                    palette=sns.color_palette("hsv", len(training_tsne_df['label'].unique())),
+                    legend='full',
+                    alpha=0.6)
+
+    # Overlay the test data points (using black 'x' markers)
+    plt.scatter(X_tsne_test[:, 0], X_tsne_test[:, 1],
+                c='black', label='Test Data', marker='x')
+
+    # Customize the plot
+    plt.title('t-SNE Visualization: Training Data with Test Data Points')
+    plt.xlabel('t-SNE Component 1')
+    plt.ylabel('t-SNE Component 2')
+    plt.legend(title='Class Labels')
+
+    # Save and return the plot image file
+    output_file = "tsne_overlay.png"
+    plt.savefig(output_file)
+    plt.close()  # Close the plot to avoid overlap on subsequent calls
+    return output_file
+
+# Create and launch the Gradio interface
+iface = gr.Interface(
+    fn=visualize_tsne_interface,
+    inputs=gr.File(label="Upload Test Data Excel File"),
+    outputs=gr.Image(type="filepath", label="t-SNE Plot"),
+    title="t-SNE Overlay Visualization",
+    description="Upload an Excel file with test data. Ensure the file includes the same feature columns used in the training data."
+)
 
-# Load saved models
-imputer = joblib.load("imputer.pkl")
-scaler = joblib.load("scaler.pkl")
-tsne = joblib.load("tsne.pkl")
-columns_to_transform = joblib.load("columns.pkl")
-
-# Function to process test data
-def process_data(file):
-    try:
-        # Load test data
-        test_df = pd.read_excel(file.name)
-
-        # Ensure correct columns
-        if set(test_df.columns) != set(columns_to_transform):
-            return "ERROR: Incorrect columns. Ensure data has only these columns: " + ", ".join(columns_to_transform), None
-        
-        # Apply log transformation
-        test_df[columns_to_transform] = test_df[columns_to_transform].apply(lambda x: np.log1p(x))
-
-        # Preprocess (impute, scale)
-        X_test = imputer.transform(test_df[columns_to_transform])
-        X_test_scaled = scaler.transform(X_test)
-
-        # Apply trained t-SNE model
-        X_test_tsne = tsne.fit_transform(X_test_scaled)
-
-        # Plot t-SNE
-        plt.figure(figsize=(8, 6))
-        plt.scatter(X_test_tsne[:, 0], X_test_tsne[:, 1], color='red', label='Test Data', alpha=0.7)
-        plt.xlabel('t-SNE Component 1')
-        plt.ylabel('t-SNE Component 2')
-        plt.legend()
-        plt.title('t-SNE Visualization')
-        plt.tight_layout()
-        tsne_fig = plt.gcf()
-
-        return tsne_fig
-
-    except Exception as e:
-        return str(e)
-
-# Gradio UI
-with gr.Blocks() as demo:
-    gr.Markdown("## Upload Geochemical Test Data")
-    file_input = gr.File(label="Upload Test Data (.xlsx)")
-    output = gr.Plot(label="t-SNE Visualization")
-    file_input.change(process_data, inputs=file_input, outputs=output)
-
-if __name__ == "__main__":
-    demo.launch(share=True)
+iface.launch()