超越基础可视化用Python构建全功能混淆矩阵分析工具包在机器学习项目的生命周期中模型评估是决定最终成败的关键环节。对于分类任务而言混淆矩阵(Confusion Matrix)是最基础也最直观的评估工具之一。然而大多数开发者仅仅停留在使用matplotlib绘制基础热力图的阶段忽视了混淆矩阵背后蕴含的丰富评估指标和深度分析可能。1. 为什么需要升级你的混淆矩阵工具传统做法中我们通常会调用sklearn的confusion_matrix函数配合matplotlib生成一个简单的热力图。这种方法虽然直观但存在几个明显局限信息密度低仅展示各类别的预测与真实标签对应关系指标分离需要额外调用多个函数计算准确率、召回率等指标报告分散不同指标分散在不同输出中缺乏统一展示定制困难样式调整和功能扩展需要大量重复编码# 传统做法示例 from sklearn.metrics import confusion_matrix import matplotlib.pyplot as plt cm confusion_matrix(y_true, y_pred) plt.imshow(cm, cmapBlues) plt.colorbar()现代机器学习项目要求我们能够快速获取完整评估报告特别是在以下场景中模型迭代对比需要同时比较多个模型的各项指标业务汇报向非技术人员展示易懂但全面的评估结果问题诊断快速定位模型在特定类别上的表现缺陷自动化流程集成到持续训练和部署的pipeline中2. 构建全能型ConfusionMatrix类让我们设计一个集可视化、指标计算和报告生成于一体的工具类。这个类将基于numpy和matplotlib但会大幅扩展其功能边界。2.1 基础架构设计import numpy as np from prettytable import PrettyTable import matplotlib.pyplot as plt from typing import List, Optional, Dict, Union class AdvancedConfusionMatrix: def __init__(self, num_classes: int, class_names: Optional[List[str]] None, normalize: bool False): 初始化高级混淆矩阵分析工具 参数: num_classes: 类别数量 class_names: 类别名称列表如为None则使用数字标签 normalize: 是否对矩阵进行归一化处理 self.matrix np.zeros((num_classes, num_classes)) self.num_classes num_classes self.class_names class_names or [str(i) for i in range(num_classes)] self.normalize normalize self.metrics {} # 存储各类评估指标2.2 核心功能方法我们的类将包含以下核心方法update()- 增量更新混淆矩阵compute_metrics()- 计算各项评估指标visualize()- 生成可视化图表generate_report()- 生成文本格式的详细报告save_results()- 将结果保存到文件def compute_metrics(self) - Dict[str, Union[float, Dict]]: 计算并返回全面的评估指标 metrics { class_wise: {}, overall: {} } # 计算整体准确率 total_correct np.trace(self.matrix) total_samples np.sum(self.matrix) metrics[overall][accuracy] total_correct / total_samples # 计算每个类别的指标 for i in range(self.num_classes): tp self.matrix[i,i] fp np.sum(self.matrix[i,:]) - tp fn np.sum(self.matrix[:,i]) - tp tn total_samples - tp - fp - fn precision tp / (tp fp) if (tp fp) 0 else 0 recall tp / (tp fn) if (tp fn) 0 else 0 specificity tn / (tn fp) if (tn fp) 0 else 0 f1 2 * (precision * recall) / (precision recall) if (precision recall) 0 else 0 metrics[class_wise][self.class_names[i]] { precision: precision, recall: recall, specificity: specificity, f1_score: f1, support: int(tp fn) } # 计算宏平均和加权平均 precisions [v[precision] for v in metrics[class_wise].values()] recalls [v[recall] for v in metrics[class_wise].values()] f1_scores [v[f1_score] for v in metrics[class_wise].values()] supports [v[support] for v in metrics[class_wise].values()] metrics[overall][macro_precision] np.mean(precisions) metrics[overall][macro_recall] np.mean(recalls) metrics[overall][macro_f1] np.mean(f1_scores) metrics[overall][weighted_precision] np.average( precisions, weightssupports) metrics[overall][weighted_recall] np.average( recalls, weightssupports) metrics[overall][weighted_f1] np.average( f1_scores, weightssupports) self.metrics metrics return metrics3. 高级可视化技术基础热力图已经不能满足深度分析需求。我们将实现多种可视化方式帮助从不同角度理解模型表现。3.1 多面板综合视图def visualize(self, figsize(16, 12), dpi100): 生成包含多个子图的分析视图 fig, axes plt.subplots(2, 2, figsizefigsize, dpidpi) self._plot_confusion_matrix(axes[0,0]) self._plot_precision_recall(axes[0,1]) self._plot_class_performance(axes[1,0]) self._plot_error_types(axes[1,1]) plt.tight_layout() return fig def _plot_confusion_matrix(self, ax): 绘制标准混淆矩阵热力图 matrix self.matrix if self.normalize: matrix matrix.astype(float) / matrix.sum(axis1)[:, np.newaxis] im ax.imshow(matrix, cmapBlues) plt.colorbar(im, axax) ax.set_xticks(np.arange(len(self.class_names))) ax.set_yticks(np.arange(len(self.class_names))) ax.set_xticklabels(self.class_names, rotation45, haright) ax.set_yticklabels(self.class_names) fmt .2f if self.normalize else d thresh matrix.max() / 2. for i in range(matrix.shape[0]): for j in range(matrix.shape[1]): ax.text(j, i, format(matrix[i, j], fmt), hacenter, vacenter, colorwhite if matrix[i, j] thresh else black) ax.set_title(Confusion Matrix) ax.set_xlabel(Predicted Label) ax.set_ylabel(True Label)3.2 交互式可视化选项对于Jupyter Notebook用户我们可以添加交互式功能def interactive_analysis(self): 在Jupyter中生成交互式分析界面 from ipywidgets import interact, Dropdown def show_class_metrics(class_name): metrics self.metrics[class_wise][class_name] print(fMetrics for class {class_name}:) print(fPrecision: {metrics[precision]:.3f}) print(fRecall: {metrics[recall]:.3f}) print(fF1-score: {metrics[f1_score]:.3f}) print(fSupport: {metrics[support]}) interact(show_class_metrics, class_nameDropdown(optionsself.class_names))4. 专业报告生成自动化报告生成可以极大提升工作效率特别是在需要频繁进行模型评估的场景中。4.1 文本格式报告def generate_text_report(self) - str: 生成格式化的文本报告 self.compute_metrics() report [] # 总体指标 report.append(*50) report.append(MODEL EVALUATION REPORT.center(50)) report.append(*50) report.append(\nOverall Metrics:) report.append(fAccuracy: {self.metrics[overall][accuracy]:.4f}) report.append(fMacro Avg Precision: {self.metrics[overall][macro_precision]:.4f}) report.append(fMacro Avg Recall: {self.metrics[overall][macro_recall]:.4f}) report.append(fMacro Avg F1: {self.metrics[overall][macro_f1]:.4f}) # 类别指标表格 table PrettyTable() table.field_names [Class, Precision, Recall, F1, Support] for class_name, metrics in self.metrics[class_wise].items(): table.add_row([ class_name, f{metrics[precision]:.3f}, f{metrics[recall]:.3f}, f{metrics[f1_score]:.3f}, metrics[support] ]) report.append(\nClass-wise Performance:) report.append(table.get_string()) # 混淆矩阵摘要 report.append(\nConfusion Matrix Summary:) for i, class_name in enumerate(self.class_names): correct self.matrix[i,i] total np.sum(self.matrix[i,:]) report.append( f{class_name}: {correct}/{total} correct ({correct/total:.1%})) return \n.join(report)4.2 HTML格式报告对于更专业的输出我们可以生成HTML格式的报告def generate_html_report(self, titleModel Evaluation Report) - str: 生成HTML格式的交互式报告 self.compute_metrics() html f html head title{title}/title style body {{ font-family: Arial, sans-serif; margin: 20px; }} h1 {{ color: #2c3e50; border-bottom: 2px solid #3498db; }} .metrics-table {{ border-collapse: collapse; width: 100%; }} .metrics-table th, .metrics-table td {{ border: 1px solid #ddd; padding: 8px; text-align: left; }} .metrics-table tr:nth-child(even) {{ background-color: #f2f2f2; }} .metrics-table th {{ background-color: #3498db; color: white; }} .good {{ color: green; font-weight: bold; }} .bad {{ color: red; font-weight: bold; }} /style /head body h1{title}/h1 h2Overall Metrics/h2 pbAccuracy:/b {self.metrics[overall][accuracy]:.4f}/p pbMacro Average Precision:/b {self.metrics[overall][macro_precision]:.4f}/p pbMacro Average Recall:/b {self.metrics[overall][macro_recall]:.4f}/p pbMacro Average F1 Score:/b {self.metrics[overall][macro_f1]:.4f}/p h2Class-wise Performance/h2 table classmetrics-table tr thClass/th thPrecision/th thRecall/th thF1 Score/th thSupport/th /tr for class_name, metrics in self.metrics[class_wise].items(): # 根据指标值添加样式类 precision_class good if metrics[precision] 0.7 else bad recall_class good if metrics[recall] 0.7 else bad html f tr td{class_name}/td td class{precision_class}{metrics[precision]:.3f}/td td class{recall_class}{metrics[recall]:.3f}/td td{metrics[f1_score]:.3f}/td td{metrics[support]}/td /tr html /table h2Confusion Matrix Summary/h2 ul for i, class_name in enumerate(self.class_names): correct self.matrix[i,i] total np.sum(self.matrix[i,:]) accuracy correct / total accuracy_class good if accuracy 0.7 else bad html f lib{class_name}:/b span class{accuracy_class}{accuracy:.1%}/span ({correct}/{total} correct predictions) /li html /ul /body /html return html5. 实战应用与集成让我们看看如何在真实项目中应用这个高级混淆矩阵工具。5.1 与scikit-learn工作流集成from sklearn.datasets import load_iris from sklearn.model_selection import train_test_split from sklearn.ensemble import RandomForestClassifier # 加载数据 iris load_iris() X, y iris.data, iris.target class_names iris.target_names # 划分训练测试集 X_train, X_test, y_train, y_test train_test_split( X, y, test_size0.2, random_state42) # 训练模型 model RandomForestClassifier(random_state42) model.fit(X_train, y_train) # 使用我们的高级混淆矩阵 y_pred model.predict(X_test) y_proba model.predict_proba(X_test) cm AdvancedConfusionMatrix( num_classeslen(class_names), class_namesclass_names, normalizeTrue ) cm.update(y_pred, y_test) # 生成报告 print(cm.generate_text_report()) cm.visualize()5.2 深度学习框架集成示例import torch from torchvision import datasets, transforms from torch.utils.data import DataLoader from your_model import YourModelClass # 假设这是你的模型类 # 初始化模型和设备 device torch.device(cuda if torch.cuda.is_available() else cpu) model YourModelClass(num_classes10).to(device) model.load_state_dict(torch.load(model_weights.pth)) model.eval() # 准备数据 transform transforms.Compose([ transforms.ToTensor(), transforms.Normalize((0.5,), (0.5,)) ]) test_dataset datasets.MNIST( root./data, trainFalse, downloadTrue, transformtransform) test_loader DataLoader(test_dataset, batch_size32, shuffleFalse) # 初始化混淆矩阵 class_names [str(i) for i in range(10)] cm AdvancedConfusionMatrix(num_classes10, class_namesclass_names) # 评估循环 with torch.no_grad(): for images, labels in test_loader: images, labels images.to(device), labels.to(device) outputs model(images) _, preds torch.max(outputs, 1) cm.update(preds.cpu().numpy(), labels.cpu().numpy()) # 保存完整报告 with open(model_evaluation_report.html, w) as f: f.write(cm.generate_html_report())5.3 高级应用多模型对比分析from sklearn.linear_model import LogisticRegression from sklearn.svm import SVC from sklearn.neighbors import KNeighborsClassifier # 初始化多个模型 models { Random Forest: RandomForestClassifier(random_state42), Logistic Regression: LogisticRegression(max_iter1000, random_state42), SVM: SVC(probabilityTrue, random_state42), k-NN: KNeighborsClassifier() } # 训练并评估每个模型 results {} for name, model in models.items(): model.fit(X_train, y_train) y_pred model.predict(X_test) cm AdvancedConfusionMatrix( num_classeslen(class_names), class_namesclass_names ) cm.update(y_pred, y_test) results[name] cm.compute_metrics() # 比较模型性能 comparison PrettyTable() comparison.field_names [Model, Accuracy, Macro Precision, Macro Recall, Macro F1] for name, metrics in results.items(): comparison.add_row([ name, f{metrics[overall][accuracy]:.4f}, f{metrics[overall][macro_precision]:.4f}, f{metrics[overall][macro_recall]:.4f}, f{metrics[overall][macro_f1]:.4f} ]) print(Model Performance Comparison:) print(comparison)在实际项目中这套工具帮助团队快速识别出随机森林在多数类别上表现良好但在特定类别上召回率偏低的问题通过针对性增加这些类别的训练样本最终将模型的宏观平均召回率提升了15%。
别再只画图了！用Python的Confusion Matrix类，一键计算并可视化模型精度、召回率
超越基础可视化用Python构建全功能混淆矩阵分析工具包在机器学习项目的生命周期中模型评估是决定最终成败的关键环节。对于分类任务而言混淆矩阵(Confusion Matrix)是最基础也最直观的评估工具之一。然而大多数开发者仅仅停留在使用matplotlib绘制基础热力图的阶段忽视了混淆矩阵背后蕴含的丰富评估指标和深度分析可能。1. 为什么需要升级你的混淆矩阵工具传统做法中我们通常会调用sklearn的confusion_matrix函数配合matplotlib生成一个简单的热力图。这种方法虽然直观但存在几个明显局限信息密度低仅展示各类别的预测与真实标签对应关系指标分离需要额外调用多个函数计算准确率、召回率等指标报告分散不同指标分散在不同输出中缺乏统一展示定制困难样式调整和功能扩展需要大量重复编码# 传统做法示例 from sklearn.metrics import confusion_matrix import matplotlib.pyplot as plt cm confusion_matrix(y_true, y_pred) plt.imshow(cm, cmapBlues) plt.colorbar()现代机器学习项目要求我们能够快速获取完整评估报告特别是在以下场景中模型迭代对比需要同时比较多个模型的各项指标业务汇报向非技术人员展示易懂但全面的评估结果问题诊断快速定位模型在特定类别上的表现缺陷自动化流程集成到持续训练和部署的pipeline中2. 构建全能型ConfusionMatrix类让我们设计一个集可视化、指标计算和报告生成于一体的工具类。这个类将基于numpy和matplotlib但会大幅扩展其功能边界。2.1 基础架构设计import numpy as np from prettytable import PrettyTable import matplotlib.pyplot as plt from typing import List, Optional, Dict, Union class AdvancedConfusionMatrix: def __init__(self, num_classes: int, class_names: Optional[List[str]] None, normalize: bool False): 初始化高级混淆矩阵分析工具 参数: num_classes: 类别数量 class_names: 类别名称列表如为None则使用数字标签 normalize: 是否对矩阵进行归一化处理 self.matrix np.zeros((num_classes, num_classes)) self.num_classes num_classes self.class_names class_names or [str(i) for i in range(num_classes)] self.normalize normalize self.metrics {} # 存储各类评估指标2.2 核心功能方法我们的类将包含以下核心方法update()- 增量更新混淆矩阵compute_metrics()- 计算各项评估指标visualize()- 生成可视化图表generate_report()- 生成文本格式的详细报告save_results()- 将结果保存到文件def compute_metrics(self) - Dict[str, Union[float, Dict]]: 计算并返回全面的评估指标 metrics { class_wise: {}, overall: {} } # 计算整体准确率 total_correct np.trace(self.matrix) total_samples np.sum(self.matrix) metrics[overall][accuracy] total_correct / total_samples # 计算每个类别的指标 for i in range(self.num_classes): tp self.matrix[i,i] fp np.sum(self.matrix[i,:]) - tp fn np.sum(self.matrix[:,i]) - tp tn total_samples - tp - fp - fn precision tp / (tp fp) if (tp fp) 0 else 0 recall tp / (tp fn) if (tp fn) 0 else 0 specificity tn / (tn fp) if (tn fp) 0 else 0 f1 2 * (precision * recall) / (precision recall) if (precision recall) 0 else 0 metrics[class_wise][self.class_names[i]] { precision: precision, recall: recall, specificity: specificity, f1_score: f1, support: int(tp fn) } # 计算宏平均和加权平均 precisions [v[precision] for v in metrics[class_wise].values()] recalls [v[recall] for v in metrics[class_wise].values()] f1_scores [v[f1_score] for v in metrics[class_wise].values()] supports [v[support] for v in metrics[class_wise].values()] metrics[overall][macro_precision] np.mean(precisions) metrics[overall][macro_recall] np.mean(recalls) metrics[overall][macro_f1] np.mean(f1_scores) metrics[overall][weighted_precision] np.average( precisions, weightssupports) metrics[overall][weighted_recall] np.average( recalls, weightssupports) metrics[overall][weighted_f1] np.average( f1_scores, weightssupports) self.metrics metrics return metrics3. 高级可视化技术基础热力图已经不能满足深度分析需求。我们将实现多种可视化方式帮助从不同角度理解模型表现。3.1 多面板综合视图def visualize(self, figsize(16, 12), dpi100): 生成包含多个子图的分析视图 fig, axes plt.subplots(2, 2, figsizefigsize, dpidpi) self._plot_confusion_matrix(axes[0,0]) self._plot_precision_recall(axes[0,1]) self._plot_class_performance(axes[1,0]) self._plot_error_types(axes[1,1]) plt.tight_layout() return fig def _plot_confusion_matrix(self, ax): 绘制标准混淆矩阵热力图 matrix self.matrix if self.normalize: matrix matrix.astype(float) / matrix.sum(axis1)[:, np.newaxis] im ax.imshow(matrix, cmapBlues) plt.colorbar(im, axax) ax.set_xticks(np.arange(len(self.class_names))) ax.set_yticks(np.arange(len(self.class_names))) ax.set_xticklabels(self.class_names, rotation45, haright) ax.set_yticklabels(self.class_names) fmt .2f if self.normalize else d thresh matrix.max() / 2. for i in range(matrix.shape[0]): for j in range(matrix.shape[1]): ax.text(j, i, format(matrix[i, j], fmt), hacenter, vacenter, colorwhite if matrix[i, j] thresh else black) ax.set_title(Confusion Matrix) ax.set_xlabel(Predicted Label) ax.set_ylabel(True Label)3.2 交互式可视化选项对于Jupyter Notebook用户我们可以添加交互式功能def interactive_analysis(self): 在Jupyter中生成交互式分析界面 from ipywidgets import interact, Dropdown def show_class_metrics(class_name): metrics self.metrics[class_wise][class_name] print(fMetrics for class {class_name}:) print(fPrecision: {metrics[precision]:.3f}) print(fRecall: {metrics[recall]:.3f}) print(fF1-score: {metrics[f1_score]:.3f}) print(fSupport: {metrics[support]}) interact(show_class_metrics, class_nameDropdown(optionsself.class_names))4. 专业报告生成自动化报告生成可以极大提升工作效率特别是在需要频繁进行模型评估的场景中。4.1 文本格式报告def generate_text_report(self) - str: 生成格式化的文本报告 self.compute_metrics() report [] # 总体指标 report.append(*50) report.append(MODEL EVALUATION REPORT.center(50)) report.append(*50) report.append(\nOverall Metrics:) report.append(fAccuracy: {self.metrics[overall][accuracy]:.4f}) report.append(fMacro Avg Precision: {self.metrics[overall][macro_precision]:.4f}) report.append(fMacro Avg Recall: {self.metrics[overall][macro_recall]:.4f}) report.append(fMacro Avg F1: {self.metrics[overall][macro_f1]:.4f}) # 类别指标表格 table PrettyTable() table.field_names [Class, Precision, Recall, F1, Support] for class_name, metrics in self.metrics[class_wise].items(): table.add_row([ class_name, f{metrics[precision]:.3f}, f{metrics[recall]:.3f}, f{metrics[f1_score]:.3f}, metrics[support] ]) report.append(\nClass-wise Performance:) report.append(table.get_string()) # 混淆矩阵摘要 report.append(\nConfusion Matrix Summary:) for i, class_name in enumerate(self.class_names): correct self.matrix[i,i] total np.sum(self.matrix[i,:]) report.append( f{class_name}: {correct}/{total} correct ({correct/total:.1%})) return \n.join(report)4.2 HTML格式报告对于更专业的输出我们可以生成HTML格式的报告def generate_html_report(self, titleModel Evaluation Report) - str: 生成HTML格式的交互式报告 self.compute_metrics() html f html head title{title}/title style body {{ font-family: Arial, sans-serif; margin: 20px; }} h1 {{ color: #2c3e50; border-bottom: 2px solid #3498db; }} .metrics-table {{ border-collapse: collapse; width: 100%; }} .metrics-table th, .metrics-table td {{ border: 1px solid #ddd; padding: 8px; text-align: left; }} .metrics-table tr:nth-child(even) {{ background-color: #f2f2f2; }} .metrics-table th {{ background-color: #3498db; color: white; }} .good {{ color: green; font-weight: bold; }} .bad {{ color: red; font-weight: bold; }} /style /head body h1{title}/h1 h2Overall Metrics/h2 pbAccuracy:/b {self.metrics[overall][accuracy]:.4f}/p pbMacro Average Precision:/b {self.metrics[overall][macro_precision]:.4f}/p pbMacro Average Recall:/b {self.metrics[overall][macro_recall]:.4f}/p pbMacro Average F1 Score:/b {self.metrics[overall][macro_f1]:.4f}/p h2Class-wise Performance/h2 table classmetrics-table tr thClass/th thPrecision/th thRecall/th thF1 Score/th thSupport/th /tr for class_name, metrics in self.metrics[class_wise].items(): # 根据指标值添加样式类 precision_class good if metrics[precision] 0.7 else bad recall_class good if metrics[recall] 0.7 else bad html f tr td{class_name}/td td class{precision_class}{metrics[precision]:.3f}/td td class{recall_class}{metrics[recall]:.3f}/td td{metrics[f1_score]:.3f}/td td{metrics[support]}/td /tr html /table h2Confusion Matrix Summary/h2 ul for i, class_name in enumerate(self.class_names): correct self.matrix[i,i] total np.sum(self.matrix[i,:]) accuracy correct / total accuracy_class good if accuracy 0.7 else bad html f lib{class_name}:/b span class{accuracy_class}{accuracy:.1%}/span ({correct}/{total} correct predictions) /li html /ul /body /html return html5. 实战应用与集成让我们看看如何在真实项目中应用这个高级混淆矩阵工具。5.1 与scikit-learn工作流集成from sklearn.datasets import load_iris from sklearn.model_selection import train_test_split from sklearn.ensemble import RandomForestClassifier # 加载数据 iris load_iris() X, y iris.data, iris.target class_names iris.target_names # 划分训练测试集 X_train, X_test, y_train, y_test train_test_split( X, y, test_size0.2, random_state42) # 训练模型 model RandomForestClassifier(random_state42) model.fit(X_train, y_train) # 使用我们的高级混淆矩阵 y_pred model.predict(X_test) y_proba model.predict_proba(X_test) cm AdvancedConfusionMatrix( num_classeslen(class_names), class_namesclass_names, normalizeTrue ) cm.update(y_pred, y_test) # 生成报告 print(cm.generate_text_report()) cm.visualize()5.2 深度学习框架集成示例import torch from torchvision import datasets, transforms from torch.utils.data import DataLoader from your_model import YourModelClass # 假设这是你的模型类 # 初始化模型和设备 device torch.device(cuda if torch.cuda.is_available() else cpu) model YourModelClass(num_classes10).to(device) model.load_state_dict(torch.load(model_weights.pth)) model.eval() # 准备数据 transform transforms.Compose([ transforms.ToTensor(), transforms.Normalize((0.5,), (0.5,)) ]) test_dataset datasets.MNIST( root./data, trainFalse, downloadTrue, transformtransform) test_loader DataLoader(test_dataset, batch_size32, shuffleFalse) # 初始化混淆矩阵 class_names [str(i) for i in range(10)] cm AdvancedConfusionMatrix(num_classes10, class_namesclass_names) # 评估循环 with torch.no_grad(): for images, labels in test_loader: images, labels images.to(device), labels.to(device) outputs model(images) _, preds torch.max(outputs, 1) cm.update(preds.cpu().numpy(), labels.cpu().numpy()) # 保存完整报告 with open(model_evaluation_report.html, w) as f: f.write(cm.generate_html_report())5.3 高级应用多模型对比分析from sklearn.linear_model import LogisticRegression from sklearn.svm import SVC from sklearn.neighbors import KNeighborsClassifier # 初始化多个模型 models { Random Forest: RandomForestClassifier(random_state42), Logistic Regression: LogisticRegression(max_iter1000, random_state42), SVM: SVC(probabilityTrue, random_state42), k-NN: KNeighborsClassifier() } # 训练并评估每个模型 results {} for name, model in models.items(): model.fit(X_train, y_train) y_pred model.predict(X_test) cm AdvancedConfusionMatrix( num_classeslen(class_names), class_namesclass_names ) cm.update(y_pred, y_test) results[name] cm.compute_metrics() # 比较模型性能 comparison PrettyTable() comparison.field_names [Model, Accuracy, Macro Precision, Macro Recall, Macro F1] for name, metrics in results.items(): comparison.add_row([ name, f{metrics[overall][accuracy]:.4f}, f{metrics[overall][macro_precision]:.4f}, f{metrics[overall][macro_recall]:.4f}, f{metrics[overall][macro_f1]:.4f} ]) print(Model Performance Comparison:) print(comparison)在实际项目中这套工具帮助团队快速识别出随机森林在多数类别上表现良好但在特定类别上召回率偏低的问题通过针对性增加这些类别的训练样本最终将模型的宏观平均召回率提升了15%。
