app.py

import gradio as gr
import time

import numpy as np
import matplotlib.pyplot as plt

from sklearn import ensemble
from sklearn import datasets
from sklearn.model_selection import train_test_split

theme = gr.themes.Monochrome(
    primary_hue="indigo",
    secondary_hue="blue",
    neutral_hue="slate",
)
model_card = f"""
## Description

**Gradient boosting** is a machine learning technique that combines several regression trees to create a powerful model in an iterative manner. 
**Early stopping** is a technique used in **gradient boosting** to determine the least number of iterations required to create a model that generalizes well to new data. 
It involves specifying a validation set and using it to evaluate the model after each stage of tree building. 
The process is continued until the model's scores do not improve for a specified number of stages. 
Using early stopping can significantly reduce training time, memory usage, and prediction latency while achieving almost the same accuracy as a model built without early stopping using many more estimators.
You can play around with different ``number of samples`` and ``number of new estimators`` to see the effect

## Dataset

Iris dataset, Classification dataset, Hastie dataset
"""


def do_train(n_samples, n_estimators, progress=gr.Progress()):

    data_list = [
      datasets.load_iris(return_X_y=True),
      datasets.make_classification(n_samples=n_samples, random_state=0),
      datasets.make_hastie_10_2(n_samples=n_samples, random_state=0),
    ]
    names = ["Iris Data", "Classification Data", "Hastie Data"]

    n_gb = []
    score_gb = []
    time_gb = []
    n_gbes = []
    score_gbes = []
    time_gbes = []

    for X, y in progress.tqdm(data_list):
        X_train, X_test, y_train, y_test = train_test_split(
            X, y, test_size=0.2, random_state=0
        )
        # We specify that if the scores don't improve by at least 0.01 for the last
        # 10 stages, stop fitting additional stages
        gbes = ensemble.GradientBoostingClassifier(
            n_estimators=n_estimators,
            validation_fraction=0.2,
            n_iter_no_change=5,
            tol=0.01,
            random_state=0,
        )
        gb = ensemble.GradientBoostingClassifier(n_estimators=n_estimators, random_state=0)
        start = time.time()
        gb.fit(X_train, y_train)
        time_gb.append(time.time() - start)

        start = time.time()
        gbes.fit(X_train, y_train)
        time_gbes.append(time.time() - start)

        score_gb.append(gb.score(X_test, y_test))
        score_gbes.append(gbes.score(X_test, y_test))

        n_gb.append(gb.n_estimators_)
        n_gbes.append(gbes.n_estimators_)

    bar_width = 0.2
    n = len(data_list)
    index = np.arange(0, n * bar_width, bar_width) * 2.5
    index = index[0:n]

    fig1, axes1 = plt.subplots(figsize=(9, 5))

    bar1 = axes1.bar(
        index, score_gb, bar_width, label="Without early stopping", color="crimson"
    )
    bar2 = axes1.bar(
        index + bar_width, score_gbes, bar_width, label="With early stopping", color="coral"
    )
    axes1.set_xticks(index + bar_width, names);
    axes1.set_yticks(np.arange(0, 1.3, 0.1));

    def autolabel(ax, rects, n_estimators):
        """
        Attach a text label above each bar displaying n_estimators of each model
        """
        for i, rect in enumerate(rects):
            ax.text(
                rect.get_x() + rect.get_width() / 2.0,
                1.05 * rect.get_height(),
                "n_est=%d" % n_estimators[i],
                ha="center",
                va="bottom",
            )
    autolabel(axes1, bar1, n_gb)
    autolabel(axes1, bar2, n_gbes)
    plt.xlabel("Datasets")
    plt.ylabel("Test score")

    axes1.set_xlabel("Datasets")
    axes1.set_ylabel("Test score")
    axes1.set_ylim([0, 1.3])
    axes1.legend(loc="best")
    axes1.grid(True)


    fig2, axes2 = plt.subplots(figsize=(9, 5))

    bar1 = axes2.bar(
        index, time_gb, bar_width, label="Without early stopping", color="crimson"
    )
    bar2 = axes2.bar(
        index + bar_width, time_gbes, bar_width, label="With early stopping", color="coral"
    )

    max_y = np.amax(np.maximum(time_gb, time_gbes))

    axes2.set_xticks(index + bar_width, names)
    axes2.set_yticks(np.linspace(0, 1.3 * max_y, 13))

    autolabel(axes2, bar1, n_gb)
    autolabel(axes2, bar2, n_gbes)

    axes2.set_ylim([0, 1.3 * max_y])
    axes2.legend(loc="best")
    axes2.grid(True)

    axes2.set_xlabel("Datasets")
    axes2.set_ylabel("Fit Time")


    return fig1, fig2



with gr.Blocks(theme=theme) as demo:
    gr.Markdown('''
            <div>
            <h1 style='text-align: center'>Early stopping of Gradient Boosting</h1>
            </div>
        ''')
    gr.Markdown(model_card)
    gr.Markdown("Author: <a href=\"https://huggingface.co/vumichien\">Vu Minh Chien</a>. Based on the example from <a href=\"https://scikit-learn.org/stable/auto_examples/ensemble/plot_gradient_boosting_early_stopping.html#sphx-glr-auto-examples-ensemble-plot-gradient-boosting-early-stopping-py\">scikit-learn</a>")
    n_samples = gr.Slider(minimum=500, maximum=10000, step=500, value=1000, label="Number of samples")
    n_estimators = gr.Slider(minimum=50, maximum=300, step=50, value=100, label="Number of estimators")
    with gr.Row():
        with gr.Column():
            plot1 = gr.Plot(label="Test score")
        with gr.Column():
            plot2 = gr.Plot(label="Running time")

    n_samples.change(fn=do_train, inputs=[n_samples, n_estimators], outputs=[plot1, plot2])
    n_estimators.change(fn=do_train, inputs=[n_samples, n_estimators], outputs=[plot1, plot2])

demo.launch(enable_queue=True)