Create app.py

app.py

@@ -0,0 +1,71 @@
+import gradio as gr
+import matplotlib.pyplot as plt
+from matplotlib import ticker
+from sklearn import manifold, datasets
+from mpl_toolkits.mplot3d import Axes3D
+
+
+def compare_manifold_learning(methods, n_samples, n_neighbors, n_components, perplexity):
+    S_points, S_color = datasets.make_s_curve(n_samples, random_state=0)
+    transformed_data = []
+
+    for method in methods:
+        manifold_method = {
+            "LLE Standard": manifold.LocallyLinearEmbedding(method="standard", n_neighbors=n_neighbors, n_components=n_components, eigen_solver="auto", random_state=0),
+            "LLE LTSA": manifold.LocallyLinearEmbedding(method="ltsa", n_neighbors=n_neighbors, n_components=n_components, eigen_solver="auto", random_state=0),
+            "LLE Hessian": manifold.LocallyLinearEmbedding(method="hessian", n_neighbors=n_neighbors, n_components=n_components, eigen_solver="auto", random_state=0),
+            "LLE Modified": manifold.LocallyLinearEmbedding(method="modified", n_neighbors=n_neighbors, n_components=n_components, eigen_solver="auto", random_state=0),
+            "Isomap": manifold.Isomap(n_neighbors=n_neighbors, n_components=n_components, p=1),
+            "MDS": manifold.MDS(n_components=n_components, max_iter=50, n_init=4, random_state=0, normalized_stress=False),
+            "Spectral Embedding": manifold.SpectralEmbedding(n_components=n_components, n_neighbors=n_neighbors),
+            "t-SNE": manifold.TSNE(n_components=n_components, perplexity=perplexity, init="random", n_iter=250, random_state=0)
+        }[method]
+        S_transformed = manifold_method.fit_transform(S_points)
+        transformed_data.append(S_transformed)
+
+    fig, axs = plt.subplots(1, len(transformed_data), figsize=(6 * len(transformed_data), 6))
+    fig.suptitle("Manifold Learning Comparison", fontsize=16)
+
+    for ax, method, data in zip(axs, methods, transformed_data):
+        ax.scatter(data[:, 0], data[:, 1], c=S_color, cmap=plt.cm.Spectral)
+        ax.set_title(f"Method: {method}")
+        ax.axis("tight")
+        ax.axis("off")
+        ax.xaxis.set_major_locator(ticker.NullLocator())
+        ax.yaxis.set_major_locator(ticker.NullLocator())
+
+    plt.tight_layout()
+    plt.savefig("plot.png")
+    plt.close()
+
+    return "plot.png"
+
+method_options = [
+    "LLE Standard",
+    "LLE LTSA",
+    "LLE Hessian",
+    "LLE Modified",
+    "Isomap",
+    "MDS",
+    "Spectral Embedding",
+    "t-SNE"
+]
+
+inputs = [
+    gr.components.CheckboxGroup(method_options, label="Manifold Learning Methods"),
+    gr.inputs.Slider(default=1500, label="Number of Samples", maximum=5000),
+    gr.inputs.Slider(default=12, label="Number of Neighbors"),
+    gr.inputs.Slider(default=2, label="Number of Components"),
+    gr.inputs.Slider(default=30, label="Perplexity (for t-SNE)")
+]
+
+gr.Interface(
+    fn=compare_manifold_learning,
+    inputs=inputs,
+    outputs="image",
+    examples=[
+        [method_options, 1500, 12, 2, 30]
+    ],
+    title="Manifold Learning Comparison",
+    description="Compare manifold learning methods on the S-curve dataset."
+).launch()