svc.fit(X=, y=, sample_weight=)理解这句代码

% ========== 训练阶段：统一预处理流程 ========== clc; clear; % ========== 参数配置 ========== test_dir = 'F:\dataset\test'; val_dir = 'F:\dataset\val'; save_dir = 'F:\results'; rows = 480; cols = 480; bands = 300; sg_params.window = 11; sg_params.order = 3; % ========== 加载训练数据 ========== roi_dir = 'F:\dataset_roi_masks'; [X_train, Y_train] = load_all_rois(roi_dir); if iscell(X_train), X_train = cell2mat(X_train); end X_train = double(X_train); % SG 滤波 X_train = apply_sg_filter(X_train, sg_params.window, sg_params.order); % 去除NaN nan_rows = any(isnan(X_train), 2); X_train(nan_rows, :) = []; Y_train(nan_rows, :) = []; % PCA：训练阶段只 fit 一次 [coeff, ~, latent, ~, explained, mu] = pca(X_train); cum_explained = cumsum(explained); num_pc = find(cum_explained >= 95, 1); X_train_pca = (X_train - mu) * coeff(:, 1:num_pc); num_pc = min(num_pc, 10); pca_coeff = coeff; pca_mu = mu; fprintf('选择主成分数 num_pc = %d，累计贡献率 = %.2f%%\n', num_pc, cum_explained(num_pc)); % 类别样本权重（平衡） num_0 = sum(Y_train == 0); num_1 = sum(Y_train == 1); weight_0 = length(Y_train) / (2 * num_0); weight_1 = length(Y_train) / (2 * num_1); sample_weights = zeros(size(Y_train)); sample_weights(Y_train == 0) = weight_0; sample_weights(Y_train == 1) = weight_1; % ========== SVM 训练 ========== C = 0.1; Gamma = 0.1; SVMModel = fitcsvm(X_train_pca, Y_train, ... 'KernelFunction', 'rbf', ... 'BoxConstraint', C, ... 'KernelScale', Gamma, ... 'Weights', sample_weights, ... 'Standardize', true, ... 'ClassNames', [0; 1]); fprintf('训练完成，支持向量数量: %d\n', size(SVMModel.SupportVectors, 1)); % 保存模型与参数 save('SVMModel.mat', 'SVMModel', 'pca_coeff', 'pca_mu', 'sg_params', 'num_pc'); % ========== 批量预测 ========== predict_folder(test_dir, SVMModel, rows, cols, bands, ... sg_params, pca_coeff, pca_mu, num_pc, save_dir, 'test_'); predict_folder(val_dir, SVMModel, rows, cols, bands, ... sg_params, pca_coeff, pca_mu, num_pc, save_dir, 'val_'); 分析代码