444

function.py

@@ -1,4 +1,3 @@
-# function.py
 import numpy as np
 from sklearn.model_selection import train_test_split, GridSearchCV, cross_val_score
 from sklearn.preprocessing import StandardScaler
@@ -6,6 +5,7 @@ from sklearn.metrics import accuracy_score, roc_auc_score, f1_score
 from sklearn.ensemble import RandomForestClassifier, GradientBoostingClassifier, VotingClassifier
 from sklearn.linear_model import LogisticRegression
 from sklearn.svm import SVC
+
 """
 实验说明:
 - 自行编写实现数据加载与预测病马死亡的对率回归算法。
@@ -22,12 +22,12 @@ from sklearn.svm import SVC
 - numpy: 2.1.1
 """
 
-
 """
 请在这里填写学号和姓名，不填写即成绩无效。
 """
 id = "202210423221"
 name = "廖子豪"
+
 def load_data(filepath):
     file = open(filepath)
     fr = file.readlines()
@@ -54,7 +54,7 @@ def load_data(filepath):
     X_train = np.insert(X, 0, 1, axis=1)  # 添加偏置项
     return X_train, y_train
 
-def train_logreg(X, y, numIter=1000, learning_rate=0.06, batch_size=128, lambda_reg=0.01):
+def train_logreg(X, y, numIter=1000, learning_rate=0.06, batch_size=128, lambda_reg=0.01, penalty='l2'):
     m, n = X.shape
     weights = np.zeros(n)
     for j in range(numIter):
@@ -67,9 +67,14 @@ def train_logreg(X, y, numIter=1000, learning_rate=0.06, batch_size=128, lambda_
             h = 1 / (1 + np.exp(-z))
             error = batch_y - h
             gradient = np.dot(batch_X.T, error) / len(batch_y)
-            reg_term = 2 * lambda_reg * weights
+            if penalty == 'l1':
+                reg_term = lambda_reg * np.sign(weights)
+            elif penalty == 'l2':
+                reg_term = 2 * lambda_reg * weights
+            else:
+                raise ValueError("Unsupported penalty type. Use 'l1' or 'l2'.")
             gradient -= reg_term / len(batch_y)
-            weights += learning_rate * gradient
+            weights -= learning_rate * gradient
     return weights
 
 def predict_logreg(inX, weights):
@@ -90,7 +95,7 @@ if __name__ == "__main__":
     best_weights = None
     best_auc = 0
     for iter in range(1, 1001):
-        weights = train_logreg(X_train, y_train, numIter=iter, learning_rate=0.01, lambda_reg=0.1)
+        weights = train_logreg(X_train, y_train, numIter=iter, learning_rate=0.01, lambda_reg=0.1, penalty='l2')
         y_pred = [predict_logreg(x, weights) for x in X_val]
         auc = roc_auc_score(y_val, y_pred)
         if auc > best_auc: