1、svc參數的說明
SVC參數解釋
(1)C: 目標函數的懲罰係數C,用來平衡分類間隔margin和錯分樣本的,default C = 1.0;
(2)kernel:參數選擇有RBF, Linear, Poly, Sigmoid, 默認的是"RBF";
(3)degree:if you choose 'Poly' in param 2, this is effective, degree決定了多項式的最高次冪;
(4)gamma:核函數的係數('Poly', 'RBF' and 'Sigmoid'), 默認是gamma = 1 / n_features;
(5)coef0:核函數中的獨立項,'RBF' and 'Poly'有效;
(6)probablity: 可能性估計是否使用(true or false);
(7)shrinking:是否進行啓發式;
(8)tol(default = 1e - 3): svm結束標準的精度;
(9)cache_size: 制定訓練所需要的內存(以MB爲單位);
(10)class_weight: 每個類所佔據的權重,不同的類設置不同的懲罰參數C, 缺省的話自適應;
(11)verbose: 跟多線程有關,不大明白啥意思具體;
(12)max_iter: 最大迭代次數,default = 1, if max_iter = -1, no limited;
(13)decision_function_shape : ‘ovo’ 一對一, ‘ovr’ 多對多 or None 無, default=None
(14)random_state :用於概率估計的數據重排時的僞隨機數生成器的種子。
2、交叉驗證
from sklearn.cross_validation import cross_val_score
metric = cross_val_score(clf,X,y,cv=5,scoring=‘ ‘).mean()
[‘accuracy‘,
‘adjusted_rand_score‘, ‘average_precision‘, ‘f1‘, ‘f1_macro‘, ‘f1_micro‘, ‘f1_samples‘, ‘f1_weighted‘, ‘log_loss‘, ‘mean_absolute_error‘, ‘mean_squared_error‘, ‘median_absolute_error‘, ‘precision‘, ‘precision_macro‘, ‘precision_micro‘, ‘precision_samples‘,
‘precision_weighted‘, ‘r2‘, ‘recall‘, ‘recall_macro‘, ‘recall_micro‘, ‘recall_samples‘, ‘recall_weighted‘, ‘roc_auc‘]
sklearn.metrics.accuracy_score
sklearn.metrics.average_precision_score
sklearn.metrics.f1_score f1就是F-measure
sklearn.metrics.precision_score
sklearn.metrics.recall_score
sklearn.metrics.roc_auc_score
sklearn.metrics.adjusted_rand_score
sklearn.metrics.mean_squared_error