app.py

import numpy as np
import gradio as gr
import matplotlib.pyplot as plt
from sklearn.manifold import TSNE
import base64
from io import BytesIO

# Function to create a tensor from input vectors and plot them
def plot_vectors(file):
    try:
        # Load the numpy array from the uploaded file
        tensor = np.load(file.name)
        
        # Check if the tensor is 2D
        if len(tensor.shape) != 2:
            raise ValueError("The file must contain a 2D array of 1D vectors.")
        
        # Apply t-SNE for dimensionality reduction
        tsne = TSNE(n_components=2, random_state=42)
        reduced_vectors = tsne.fit_transform(tensor)
        
        # Extract x and y components from the reduced vectors
        x = reduced_vectors[:, 0]
        y = reduced_vectors[:, 1]
        
        # Create the plot
        plt.figure(figsize=(8, 8))
        plt.scatter(x, y, c='blue', marker='o', label='Vectors')
        plt.xlabel('X component')
        plt.ylabel('Y component')
        plt.title('2D Distribution of Vectors (t-SNE)')
        plt.legend()
        plt.grid(True)
        plt.axis('equal')  # Equal scaling on both axes
        
        # Save the plot to a buffer
        buf = BytesIO()
        plt.savefig("result.png")
        buf.seek(0)
        plt.close()
        
        # Convert the buffer to a base64-encoded string
        buf.seek(0)
        img_str = base64.b64encode(buf.getvalue()).decode('utf-8')
        
        # Return the base64-encoded image string
        return "result.png"
    except Exception as e:
        return str(e)

# Create a Gradio interface
iface = gr.Interface(
    fn=plot_vectors, 
    inputs=gr.File(label="Upload a .npy file containing a 2D array of 1D vectors"), 
    outputs=gr.Image(type="filepath", label="Visualization Result"), 
    examples=[["random_array.npy"]]  # Example file
)

# Launch the interface
iface.launch()