朴素贝叶斯属于:概率模型、参数化模型、和生成模型
目录
1.朴素贝叶斯基本方法
2.后验概率最大化的含义
3.朴素贝叶斯算法:
朴素贝叶斯python实现4.1:
class Navie_Bayes():
def __init__(self, x_train, y_train, x_test, y_test=None):
self.x, self.y, self.xs, self.ys = x_train, y_train, x_test, y_test
def fit(self):
a, b = self.x.shape
tmp_x_0, tmp_y_0, tmp_x_1, tmp_y_1 = [], [], [], []
for i in range(a):
if self.y[i] == 1:
tmp_x_0.append(self.x[i])
tmp_y_0.append(self.y[i])
else:
tmp_x_1.append(self.x[i])
tmp_y_1.append(self.y[i])
# 求先验概率
p_ck1 = len(tmp_y_0) / a
p_ck0 = 1 - p_ck1
# 求y=1条件概率
tmp_1, tmp_2, tmp_3 = [], [], []
tmp_S, tmp_M, tmp_L = [], [], []
for j in range(len(tmp_y_0)):
if int(tmp_x_0[j][0]) == 1:
tmp_1.append(tmp_x_0[j])
elif int(tmp_x_0[j][0]) == 2:
tmp_2.append(tmp_x_0[j])
else:
tmp_3.append(tmp_x_0[j])
for v in range(len(tmp_y_0)):
if tmp_x_0[v][1] == 'S':
tmp_S.append(tmp_x_0[v])
elif tmp_x_0[v][1] == 'M':
tmp_M.append(tmp_x_0[v])
else:
tmp_L.append(tmp_x_0[v])
# 求y=-1条件概率
tmp_1_, tmp_2_, tmp_3_ = [], [], []
tmp_S_, tmp_M_, tmp_L_ = [], [], []
for j_ in range(a-len(tmp_y_0)):
if int(tmp_x_1[j_][0]) == 1:
tmp_1_.append(tmp_x_1[j_])
elif int(tmp_x_1[j_][0]) == 2:
tmp_2_.append(tmp_x_1[j_])
else:
tmp_3_.append(tmp_x_1[j_])
for v_ in range(a-len(tmp_y_0)):
if tmp_x_1[v_][1] == 'S':
tmp_S_.append(tmp_x_1[v_])
elif tmp_x_1[v_][1] == 'M':
tmp_M_.append(tmp_x_1[v_])
else:
tmp_L_.append(tmp_x_1[v_])
# 后验概率
p_y1 = p_ck1 * (len(tmp_2)/len(tmp_y_0)) * (len(tmp_S)/len(tmp_y_0))
p_y2 = p_ck0 * (len(tmp_2_)/(a-len(tmp_y_0))) * (len(tmp_S_)/(a-len(tmp_y_0)))
if p_y1 > p_y2:
print('预测结果为:',1)
return 1
else:
print('预测结果为:',-1)
return -1
def main():
x_train = np.array([[1, 'S'], [1, 'M'], [1, 'M'], [1, 'S'], [1, 'S'], [2, 'S'], [2, 'M'], [2, 'M'],
[2, 'L'], [2, 'L'], [3, 'L'], [3, 'M'], [3, 'M'], [3, 'L'], [3, 'L']])
y_train = np.array([-1, -1, 1, 1, -1, -1, -1, 1, 1, 1, 1, 1, 1, 1, -1])
x_test = np.array([2, 'S'])
navie_bayes = Navie_Bayes(x_train, y_train, x_test)
navie_bayes.fit()
结果:
朴素贝叶斯sklearn实现作业4.1
# 令s=1,m=2, l=3
x_train = np.array([[1, 1], [1, 2], [1, 2], [1, 1], [1, 1], [2, 1], [2, 2], [2, 2],
[2, 3], [2, 3], [3, 3], [3, 2], [3, 2], [3, 3], [3, 3]])
y_train = np.array([-1, -1, 1, 1, -1, -1, -1, 1, 1, 1, 1, 1, 1, 1, -1])
x_test = np.array([2, 2])
navie_bayes_gauss = GaussianNB()
navie_bayes_multi = MultinomialNB(alpha=2.0, class_prior=None, fit_prior=True)
navie_bayes_bou = BernoulliNB(alpha=2.0,binarize = 3.0,fit_prior=True)
tt = [navie_bayes_gauss, navie_bayes_multi, navie_bayes_bou]
tt_str = ['GaussianNB', 'BernoulliNB','MultinomialNB']
for items, name in zip(tt, tt_str):
items.fit(x_train, y_train)
xx = items.predict_proba(x_test.reshape(1, -1))
print('%s 模型的预测概率为:'%name, xx)
贝叶斯的优缺点:
由于贝叶斯是对条件概率分布的条件独立性假设,那么在数据集比较小的时候它会有比较好的效果,但是当数据集比较大的时候,误差会随意数据集的增大而增大,这个时候它的效果可能还不如逻辑回归等模型。
在小型数据集上,朴素贝叶斯更容易产生偏差,这可以防止其拟合噪声。