sklearn-docs/MLP-Regularization

1import numpy as np2import plotly.graph_objects as go3 4from sklearn.preprocessing import StandardScaler5from sklearn.datasets import make_moons, make_circles, make_classification, make_blobs6from sklearn.neural_network import MLPClassifier7 8import gradio as gr9 10# =========================================================================11 12GRANULARITY = 0.213MARGIN = 0.514N_SAMPLES = 15015SEED = 116 17datasets = {}18X, y = make_moons(n_samples=N_SAMPLES, noise=0.2, random_state=SEED)19X = StandardScaler().fit_transform(X)20datasets["Moons"] = (X.copy(), y.copy())21 22X, y = make_circles(n_samples=N_SAMPLES, noise=0.2, factor=0.5, random_state=SEED)23X = StandardScaler().fit_transform(X)24datasets["Circles"] = (X.copy(), y.copy())25 26X, y = make_blobs(n_samples=N_SAMPLES, n_features=2, centers=4, cluster_std=2, random_state=SEED)27X = StandardScaler().fit_transform(X)28y[y==2] = 029y[y==3] = 130datasets["Blobs"] = (X.copy(), y.copy())31 32X, y =  make_classification(n_samples=N_SAMPLES, n_features=2, n_redundant=0, n_informative=2, n_clusters_per_class=1, random_state=SEED)33X += 2 * np.random.uniform(size=X.shape)34X = StandardScaler().fit_transform(X)35datasets["Linear"] = (X.copy(), y.copy())36 37# =========================================================================38 39def get_figure_dict():40    figure_dict = dict(data=[], layout={}, frames=[])41 42    play_button = dict(args=[None, {"mode": "immediate", "fromcurrent": False, "frame": {"duration": 50}, "transition": {"duration": 50}}],43                   label="Play",44                   method="animate")45 46    pause_button = dict(args=[[None], {"mode": "immediate"}],47                    label="Stop",48                    method="animate")49 50    slider = dict(steps=[], active=0, currentvalue={"prefix": "Iteration: "})51 52    figure_dict["layout"] = dict(width=600, height=600, hovermode=False, margin=dict(l=40, r=40, t=40, b=40), 53                                      title=dict(text="Decision Surface", x=0.5),54                                      sliders=[slider],55                                      updatemenus=[dict(buttons=[play_button, pause_button], direction="left", pad={"t": 85}, type="buttons", x=0.6, y=-0.05)]56                                      )57 58    return figure_dict59 60def get_decision_surface(X, model):61    x_min, x_max = X[:, 0].min() - MARGIN, X[:, 0].max() + MARGIN62    y_min, y_max = X[:, 1].min() - MARGIN, X[:, 1].max() + MARGIN63    xrange = np.arange(x_min, x_max, GRANULARITY)64    yrange = np.arange(y_min, y_max, GRANULARITY)65    x, y = np.meshgrid(xrange, yrange)66    x = x.ravel(); y = y.ravel()67    z = model.predict_proba(np.column_stack([x, y]))[:, 1]68    return x, y, z69# =========================================================================70 71def create_plot(dataset, alpha, h1, h2, seed):72    X, y = datasets[dataset]73 74    model = MLPClassifier(alpha=alpha, max_iter=2000, learning_rate_init=0.01, hidden_layer_sizes=[h1, h2], random_state=seed)75 76    figure_dict = get_figure_dict()77 78    model.partial_fit(X, y, classes=[0, 1])79    xx, yy, zz = get_decision_surface(X, model)80    figure_dict["data"] = [go.Contour(x=xx, y=yy, z=zz, opacity=0.6, showscale=False,),81                        go.Scatter(x=X[:, 0], y=X[:, 1], mode="markers", marker_color=y, marker={"colorscale": "jet", "size": 8})]82 83    prev_loss = np.inf84    tol = 3e-485    for i in range(100):86        for _ in range(3):87            model.partial_fit(X, y, classes=[0, 1])88        89        if prev_loss - model.loss_ <= tol: break90        prev_loss = model.loss_91        92        xx, yy, zz = get_decision_surface(X, model)93        figure_dict["frames"].append({"data": [go.Contour(x=xx, y=yy, z=zz, opacity=0.6, showscale=False)], "name": i})94 95        slider_step = {"args": [[i], {"mode": "immediate"}], "method": "animate", "label": i}96        figure_dict["layout"]["sliders"][0]["steps"].append(slider_step)97 98    fig = go.Figure(figure_dict)99    return fig100 101info = '''102# Effect of Regularization Parameter of Multilayer Perceptron103 104This example demonstrates the effect of varying the regularization parameter (alpha) of a multilayer perceptron on the binary classification of toy datasets, as represented by the decision surface of the classifier.105 106Higher values of alpha encourages smaller weights, thus making the model less prone to overfitting, while lower values may help against underfitting. Use the slider below to control the amount of regularization and observe how the decision surface changes with higher values.107 108The neural network is trained until the loss stops decreasing below a specific tolerance. The color of the decision surface represents the probability of observing the corresponding class.109 110Created by [@huabdul](https://huggingface.co/huabdul) based on [scikit-learn docs](https://scikit-learn.org/stable/auto_examples/neural_networks/plot_mlp_alpha.html).111'''112with gr.Blocks(analytics_enabled=False) as demo:113    with gr.Row():114        with gr.Column():115            gr.Markdown(info)116            dd_dataset = gr.Dropdown(list(datasets.keys()), value="Moons", label="Dataset", interactive=True)117            with gr.Row():118                with gr.Column(min_width=100):119                    s_alpha = gr.Slider(0, 4, value=0.1, step=0.05, label="α (regularization parameter)")120                    s_seed = gr.Slider(1, 1000, value=1, step=1, label="Seed")121                with gr.Column(min_width=100):122                    s_h1 = gr.Slider(2, 20, value=10, step=1, label="Hidden layer 1 size")123                    s_h2 = gr.Slider(2, 20, value=10, step=1, label="Hidden layer 2 size")124            submit = gr.Button("Submit")125        with gr.Column():126            plot = gr.Plot(show_label=False)127    128    submit.click(create_plot, inputs=[dd_dataset, s_alpha, s_h1, s_h2, s_seed], outputs=[plot])129    demo.load(create_plot, inputs=[dd_dataset, s_alpha, s_h1, s_h2, s_seed], outputs=[plot])130 131demo.launch()