generated from LufsX/Experiment-LogReg
125 lines
3.5 KiB
Python
125 lines
3.5 KiB
Python
import numpy as np
|
|
from sklearn.model_selection import KFold
|
|
from sklearn.impute import SimpleImputer
|
|
|
|
# 请在这里填写学号和姓名，不填写即成绩无效。
|
|
id = "202210423106"
|
|
name = "刘燚燊"
|
|
|
|
def load_data(filepath):
|
|
"""
|
|
从文件中加载数据，处理缺失值和标签。
|
|
|
|
Args:
|
|
- filepath (str): 数据文件路径。
|
|
|
|
Returns:
|
|
- np.array: 特征数据。
|
|
- np.array: 标签数据。
|
|
"""
|
|
with open(filepath, 'r') as f:
|
|
data = [line.strip().split() for line in f.readlines()]
|
|
|
|
# 将数据中缺失值用0填充
|
|
data = [[0 if value == '?' else float(value) for value in line] for line in data]
|
|
data = np.array(data)
|
|
|
|
# 分离特征和标签
|
|
X = data[:, :-1]
|
|
y = data[:, -1]
|
|
|
|
# 使用均值填充缺失值
|
|
imputer = SimpleImputer(strategy='mean')
|
|
X = imputer.fit_transform(X)
|
|
|
|
# 过滤掉没有类别标签的数据
|
|
valid_indices = np.isfinite(y)
|
|
X = X[valid_indices]
|
|
y = y[valid_indices]
|
|
|
|
return X, y
|
|
|
|
|
|
def train_logreg(X, y, learning_rate=0.01, iterations=1000, reg_param=0.01):
|
|
# 初始化权重
|
|
weights = np.zeros(X.shape[1])
|
|
|
|
for i in range(iterations):
|
|
# 计算线性组合和sigmoid函数
|
|
linear_model = np.dot(X, weights)
|
|
y_predicted = 1 / (1 + np.exp(-linear_model))
|
|
|
|
# 计算梯度
|
|
gradient = np.dot(X.T, (y_predicted - y)) / y.size
|
|
weights -= learning_rate * (gradient + reg_param * weights) # L2 正则化
|
|
|
|
return weights
|
|
|
|
|
|
def predict_logreg(inX, weights):
|
|
"""
|
|
对率回归预测函数。
|
|
|
|
Args:
|
|
- inX (numpy array): 输入待分类的样本。
|
|
- weights (numpy array): 训练得到的权重。
|
|
|
|
Returns:
|
|
- int: 预测的分类标签。
|
|
"""
|
|
linear_model = np.dot(inX, weights)
|
|
y_predicted = 1 / (1 + np.exp(-linear_model))
|
|
return (y_predicted > 0.5).astype(int)
|
|
|
|
|
|
def evaluate_model(X, y, learning_rate, iterations, reg_param, folds=5):
|
|
"""
|
|
使用交叉验证评估模型性能。
|
|
|
|
Args:
|
|
- X (numpy array): 特征数据。
|
|
- y (numpy array): 标签数据。
|
|
- learning_rate (float): 学习率。
|
|
- iterations (int): 训练迭代次数。
|
|
- reg_param (float): 正则化参数。
|
|
- folds (int): 交叉验证折数。
|
|
|
|
Returns:
|
|
- float: 平均准确率。
|
|
"""
|
|
kf = KFold(n_splits=folds)
|
|
accuracy_list = []
|
|
|
|
for train_index, test_index in kf.split(X):
|
|
X_train, X_test = X[train_index], X[test_index]
|
|
y_train, y_test = y[train_index], y[test_index]
|
|
|
|
weights = train_logreg(X_train, y_train, learning_rate, iterations, reg_param)
|
|
predictions = predict_logreg(X_test, weights)
|
|
|
|
accuracy = np.mean(predictions == y_test)
|
|
accuracy_list.append(accuracy)
|
|
|
|
return np.mean(accuracy_list)
|
|
|
|
|
|
if __name__ == "__main__":
|
|
# 加载训练集
|
|
X_train, y_train = load_data("./data/horse-colic.data")
|
|
|
|
# 调优超参数
|
|
best_accuracy = 0
|
|
best_params = {}
|
|
learning_rates = [0.001, 0.01, 0.1]
|
|
reg_params = [0.01, 0.1, 1.0]
|
|
|
|
for lr in learning_rates:
|
|
for reg in reg_params:
|
|
accuracy = evaluate_model(X_train, y_train, lr, 1000, reg)
|
|
print(f"尝试参数: {{'learning_rate': {lr}, 'reg_param': {reg}}}, 准确率: {accuracy:.2f}")
|
|
if accuracy > best_accuracy:
|
|
best_accuracy = accuracy
|
|
best_params = {'learning_rate': lr, 'reg_param': reg}
|
|
|
|
print(f"最佳参数: {best_params}, 最高准确率: {best_accuracy:.2f}")
|