逻辑回归（二）

image.png
image.png

# 模型训练
log_model = LogisticRegression()
log_model.fit(new_x, y)

from sklearn.metrics import accuracy_score

# 查看准确率
predict_y = log_model.predict(new_x)
accuracy_score(y, predict_y)   #准确率0.8135593220338984

# 模型系数
the0 = log_model.intercept_
the1, the2, the3, the4, the5 = log_model.coef_[0]
print(the0, the1, the2, the3, the4, the5)

# 对x1排序
x1_new = x1.sort_values()

# 求出新的边界
a = the4
b = the2 + the5 * x1_new
c = the0 + the1*x1_new + the3*x1_new* x1_new
# 新的边界
x2_new_boundray = (-b + np.sqrt(b*b -4*a*c)) / (2*a)
x2_new_boundray2 = (-b - np.sqrt(b*b -4*a*c)) / (2*a)

# 添加mask
mask = df["pass"] == 1

plt.figure()
passed = plt.scatter(df["test1"][mask], df["test2"][mask])
failed = plt.scatter(df["test1"][~mask], df["test2"][~mask])
t = plt.plot(x1_new, x2_new_boundray)
t2 = plt.plot(x1_new, x2_new_boundray2)
plt.xlabel("test1")
plt.ylabel("test2")
plt.legend((passed, failed), ("passed", "failed"))
plt.show()

image.png

# 封装边界函数
def f(x):
    # 求出新的边界
    a = the4
    b = the2 + the5 * x
    c = the0 + the1*x + the3*x* x
    # 新的边界
    boundray1 = (-b + np.sqrt(b*b -4*a*c)) / (2*a)
    boundray2 = (-b - np.sqrt(b*b -4*a*c)) / (2*a)
    return boundray1, boundray2
def boundray(model, x):
    """
    :param model: trained model
    :param x: sorted x1
    :return: boundray1 boundray2  boundray line
    """
    # 模型系数
    the0 = model.intercept_
    the1, the2, the3, the4, the5 = model.coef_[0]
    # 求出新的边界
    a = the4
    b = the2 + the5 * x
    c = the0 + the1*x + the3*x* x
    # 新的边界
    boundray1 = (-b + np.sqrt(b*b -4*a*c)) / (2*a)
    boundray1 = (-b - np.sqrt(b*b -4*a*c)) / (2*a)
    return boundray1, boundray2

x2_new_boundray1 = []
x2_new_boundray2 = []
for x in x1_new:
    x2_new_boundray1.append(boundray(log_model, x)[0])
    x2_new_boundray2.append(boundray(log_model, x)[1])

image.png

#生成数据画边界
x1_range = np.array([-0.9 + x/10000 for x in range(19000)])
x2_new_boundray1 = []
x2_new_boundray2 = []
for x in x1_range:
    x2_new_boundray1.append(f(x)[0])
    x2_new_boundray2.append(f(x)[1])

image.png