generated from LufsX/Experiment-LogReg
This commit is contained in:
+47
-40
@@ -1,6 +1,7 @@
|
||||
# function.py
|
||||
import numpy as np
|
||||
|
||||
from sklearn.preprocessing import StandardScaler
|
||||
from sklearn.metrics import accuracy_score
|
||||
"""
|
||||
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。
|
||||
@@ -26,64 +27,70 @@ name = "吴俊煯"
|
||||
|
||||
|
||||
# 这里写自己的代码实现
|
||||
def train_logreg(X, y, learning_rate=0.1, iterations=100):
|
||||
def train_logreg(X, y, learning_rate=0.01, iterations=1000, lambda_reg=0.1):
|
||||
m, n = X.shape
|
||||
weights = np.zeros(n)
|
||||
bias = 0
|
||||
|
||||
for _ in range(iterations):
|
||||
for idx, x_i in enumerate(X):
|
||||
linear_output = np.dot(x_i, weights) + bias
|
||||
y_predicted = 1 / (1 + np.exp(-linear_output))
|
||||
|
||||
# 计算误差
|
||||
error = y[idx] - y_predicted
|
||||
|
||||
# 更新权重和偏置
|
||||
weights += learning_rate * error * x_i
|
||||
bias += learning_rate * error
|
||||
linear_output = np.dot(X, weights) + bias
|
||||
y_predicted = 1 / (1 + np.exp(-linear_output))
|
||||
|
||||
# 计算梯度(包括正则化项)
|
||||
dw = (1 / m) * np.dot(X.T, (y_predicted - y)) + (lambda_reg / m) * weights
|
||||
db = (1 / m) * np.sum(y_predicted - y)
|
||||
|
||||
# 更新权重和偏置
|
||||
weights -= learning_rate * dw
|
||||
bias -= learning_rate * db
|
||||
|
||||
return weights
|
||||
return weights, bias
|
||||
|
||||
def predict_logreg(inX, weights):
|
||||
linear_output = np.dot(inX, weights)
|
||||
y_predicted = 1 / (1 + np.exp(-linear_output))
|
||||
|
||||
# 将输出转换为分类标签
|
||||
return 1 if y_predicted >= 0.5 else 0
|
||||
def predict_logreg(inX, weights, bias):
|
||||
linear_output = np.dot(inX, weights) + bias
|
||||
y_predicted = 1 if linear_output >= 0 else 0 # 简化阈值判断,因为已经通过sigmoid函数输出
|
||||
# 注意:这里为了与原始代码保持一致,使用了0.5作为阈值,但实际上由于sigmoid输出范围在(0,1),可以直接比较linear_output和0
|
||||
# return 1 if y_predicted >= 0.5 else 0 # 等价于上面的判断,但多余计算了sigmoid
|
||||
return y_predicted
|
||||
|
||||
# 这里也写自己的代码实现
|
||||
def load_data(filepath):
|
||||
data = []
|
||||
with open(filepath, 'r') as file:
|
||||
for line in file:
|
||||
# 分割行,并去除任何可能的空白字符
|
||||
parts = [part.strip() for part in line.split()]
|
||||
# 将缺失值'?'替换为0(或者你可以选择其他适当的处理方式)
|
||||
features = [float(part) if part != '?' else 1 for part in parts[:-1]]
|
||||
# 将标签从2转换为1(如果原始标签是1和2的话)
|
||||
label = int(parts[-1]) - 1 # 假设原始标签是1和2,转换为0和1
|
||||
parts = line.strip().split()
|
||||
# 将缺失值'?'替换为np.nan,然后使用0填充(或选择其他适当的填充策略)
|
||||
features = [float(part) if part != '?' else np.nan for part in parts[:-1]]
|
||||
# 填充NaN值(这里选择0,但也可以考虑使用均值、中位数等)
|
||||
features = np.nan_to_num(features, nan=0.0)
|
||||
# 将标签从2转换为1(如果原始标签是1和2的话),这里假设标签已经是0和1
|
||||
label = int(parts[-1]) # 假设标签已经是0和1,无需转换
|
||||
data.append(features + [label])
|
||||
|
||||
# 将数据转换为numpy数组
|
||||
|
||||
data_array = np.array(data)
|
||||
X = data_array[:, :-1]
|
||||
y = data_array[:, -1]
|
||||
|
||||
# 分离特征和标签
|
||||
X = data_array[:, :-1] # 所有行,除了最后一列
|
||||
y = data_array[:, -1] # 所有行的最后一列
|
||||
# 特征缩放(标准化)
|
||||
scaler = StandardScaler()
|
||||
X_scaled = scaler.fit_transform(X)
|
||||
|
||||
return X, y
|
||||
return X_scaled, y
|
||||
|
||||
if __name__ == "__main__":
|
||||
# 加载训练集
|
||||
# 加载训练集和测试集
|
||||
X_train, y_train = load_data("./data/horse-colic.data")
|
||||
|
||||
# 加载测试集
|
||||
X_test, y_test = load_data("./data/horse-colic.test")
|
||||
|
||||
|
||||
# 训练对率回归模型
|
||||
weights = train_logreg(X_train, y_train, learning_rate=0.01, iterations=1000)
|
||||
|
||||
# 测试单个数据
|
||||
y_pred = predict_logreg(X_test[11], weights)
|
||||
print(f"预测结果: {y_pred}, 实际结果: {y_test[11]}")
|
||||
weights, bias = train_logreg(X_train, y_train, learning_rate=0.01, iterations=1000, lambda_reg=0.1)
|
||||
|
||||
# 测试模型并评估性能
|
||||
y_pred = [predict_logreg(x, weights, bias) for x in X_test]
|
||||
accuracy = accuracy_score(y_test, y_pred)
|
||||
print(f"模型准确率: {accuracy}")
|
||||
|
||||
# 打印单个测试样本的预测结果和实际结果(示例)
|
||||
# 注意:索引11可能需要根据实际测试集调整,这里仅作为示例
|
||||
sample_index = 11 # 假设要测试的样本索引是11(注意:索引可能需要根据实际数据调整)
|
||||
print(f"预测结果: {y_pred[sample_index]}, 实际结果: {y_test[sample_index]}")
|
||||
Reference in New Issue
Block a user