import gradio as gr import numpy as np import matplotlib.pyplot as plt from sklearn.ensemble import GradientBoostingClassifier from sklearn.model_selection import KFold from sklearn.model_selection import train_test_split from sklearn.metrics import log_loss from scipy.special import expit theme = gr.themes.Monochrome( primary_hue="indigo", secondary_hue="blue", neutral_hue="slate", ) model_card = f""" ## Description The **Out-of-bag (OOB)** method is a useful technique for estimating the optimal number of boosting iterations. This method is similar to cross-validation, but it does not require repeated model fitting and can be computed on-the-fly. **OOB** estimates are only applicable to Stochastic Gradient Boosting (i.e., subsample < 1.0). They are calculated from the improvement in loss based on examples not included in the bootstrap sample (i.e., out-of-bag examples). The **OOB** estimator provides a conservative estimate of the true test loss, but is still a reasonable approximation for a small number of trees. This demo shows the negative OOB improvements' cumulative sum as a function of the boosting iteration. ## Dataset Simulation data """ def do_train(n_samples, n_splits, random_seed): # Generate data (adapted from G. Ridgeway's gbm example) random_state = np.random.RandomState(random_seed) x1 = random_state.uniform(size=n_samples) x2 = random_state.uniform(size=n_samples) x3 = random_state.randint(0, 4, size=n_samples) p = expit(np.sin(3 * x1) - 4 * x2 + x3) y = random_state.binomial(1, p, size=n_samples) X = np.c_[x1, x2, x3] X = X.astype(np.float32) X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.5, random_state=random_seed) # Fit classifier with out-of-bag estimates params = { "n_estimators": 1200, "max_depth": 3, "subsample": 0.5, "learning_rate": 0.01, "min_samples_leaf": 1, "random_state": random_seed, } clf = GradientBoostingClassifier(**params) clf.fit(X_train, y_train) train_acc = clf.score(X_train, y_train) test_acc = clf.score(X_test, y_test) text = f"Train set accuracy: {train_acc*100:.2f}%. Test set accuracy: {test_acc*100:.2f}%" n_estimators = params["n_estimators"] x = np.arange(n_estimators) + 1 def heldout_score(clf, X_test, y_test): """compute deviance scores on ``X_test`` and ``y_test``.""" score = np.zeros((n_estimators,), dtype=np.float64) for i, y_proba in enumerate(clf.staged_predict_proba(X_test)): score[i] = 2 * log_loss(y_test, y_proba[:, 1]) return score def cv_estimate(n_splits): cv = KFold(n_splits=n_splits) cv_clf = GradientBoostingClassifier(**params) val_scores = np.zeros((n_estimators,), dtype=np.float64) for train, test in cv.split(X_train, y_train): cv_clf.fit(X_train[train], y_train[train]) val_scores += heldout_score(cv_clf, X_train[test], y_train[test]) val_scores /= n_splits return val_scores # Estimate best n_splits using cross-validation cv_score = cv_estimate(n_splits) # Compute best n_splits for test data test_score = heldout_score(clf, X_test, y_test) # negative cumulative sum of oob improvements cumsum = -np.cumsum(clf.oob_improvement_) # min loss according to OOB oob_best_iter = x[np.argmin(cumsum)] # min loss according to test (normalize such that first loss is 0) test_score -= test_score[0] test_best_iter = x[np.argmin(test_score)] # min loss according to cv (normalize such that first loss is 0) cv_score -= cv_score[0] cv_best_iter = x[np.argmin(cv_score)] # color brew for the three curves oob_color = list(map(lambda x: x / 256.0, (190, 174, 212))) test_color = list(map(lambda x: x / 256.0, (127, 201, 127))) cv_color = list(map(lambda x: x / 256.0, (253, 192, 134))) # line type for the three curves oob_line = "dashed" test_line = "solid" cv_line = "dashdot" # plot curves and vertical lines for best iterations fig, ax = plt.subplots(figsize=(8, 6)) ax.plot(x, cumsum, label="OOB loss", color=oob_color, linestyle=oob_line) ax.plot(x, test_score, label="Test loss", color=test_color, linestyle=test_line) ax.plot(x, cv_score, label="CV loss", color=cv_color, linestyle=cv_line) ax.axvline(x=oob_best_iter, color=oob_color, linestyle=oob_line) ax.axvline(x=test_best_iter, color=test_color, linestyle=test_line) ax.axvline(x=cv_best_iter, color=cv_color, linestyle=cv_line) # add three vertical lines to xticks xticks = plt.xticks() xticks_pos = np.array( xticks[0].tolist() + [oob_best_iter, cv_best_iter, test_best_iter] ) xticks_label = np.array(list(map(lambda t: int(t), xticks[0])) + ["OOB", "CV", "Test"]) ind = np.argsort(xticks_pos) xticks_pos = xticks_pos[ind] xticks_label = xticks_label[ind] ax.set_xticks(xticks_pos, xticks_label, rotation=90) ax.legend(loc="upper center") ax.set_ylabel("normalized loss") ax.set_xlabel("number of iterations") return fig, text with gr.Blocks(theme=theme) as demo: gr.Markdown('''

Gradient Boosting Out-of-Bag estimates

''') gr.Markdown(model_card) gr.Markdown("Author: Vu Minh Chien. Based on the example from scikit-learn") n_samples = gr.Slider(minimum=500, maximum=5000, step=500, value=500, label="Number of samples") n_splits = gr.Slider(minimum=2, maximum=10, step=1, value=3, label="Number of cross validation folds") random_seed = gr.Slider(minimum=0, maximum=2000, step=1, value=0, label="Random seed") with gr.Row(): with gr.Column(): plot = gr.Plot() with gr.Column(): result = gr.Textbox(label="Resusts") n_samples.change(fn=do_train, inputs=[n_samples, n_splits, random_seed], outputs=[plot, result]) n_splits.change(fn=do_train, inputs=[n_samples, n_splits, random_seed], outputs=[plot, result]) random_seed.change(fn=do_train, inputs=[n_samples, n_splits, random_seed], outputs=[plot, result]) demo.launch()