问题遇到的现象和发生背景
在使用keras搭建简单模型时，利用pandas将数据读取并清洗后，采用np.array()转化为np.array格式导入keras网络（只有简单的全连接层，采用sigmoid作为输出函数），但是在调用fit方法时，选择的batchsize为40，epochs为10，出现迭代信息不连续的情况，从40跳跃到1000，接着就是2000，不应该从40到80再到120嘛？见下图：
问题相关代码，请勿粘贴截图
import numpy as np
import pandas as pd
import keras as k
original_data = pd.read_csv("E:\Desktop\stroke_data.csv")
def data_process(original_data):
data = original_data.dropna(axis=0, how='any')
data = data.drop(columns='id')
data = data.reset_index(drop=True)
gender = np.zeros(data.shape[0])
for i in range(gender.shape[0]):
if data.loc[i, 'gender'] == 'Male':
gender[i] = 1
else:
gender[i] = 0
data = data.drop(columns='gender')
data.insert(0, 'gender', gender)
ever_married = np.zeros(data.shape[0])
for i in range(ever_married.shape[0]):
if data['ever_married'][i] == 'Yes':
ever_married[i] = 1
else:
ever_married[i] = 0
data = data.drop(columns='ever_married')
data.insert(4, 'ever_married', ever_married)
Residence_type = np.zeros(data.shape[0])
for i in range(Residence_type.shape[0]):
if data['Residence_type'][i] == 'Urban':
Residence_type[i] = 1
else:
Residence_type[i] = 0
data = data.drop(columns='Residence_type')
data.insert(6, 'Residence_type', Residence_type)
data['heart_disease'] = data['heart_disease'].apply(lambda x: x * 10)
data['hypertension'] = data['hypertension'].apply(lambda x: x * 10)
work_type = np.zeros(data.shape[0])
for i in range(work_type.shape[0]):
if data['work_type'][i] == 'Self-employed':
work_type[i] = 5
elif data['work_type'][i] == 'Private':
work_type[i] = 4
elif data['work_type'][i] == 'Self-children':
work_type[i] = 3
elif data['work_type'][i] == 'Govt_job':
work_type[i] = 2
else:
work_type[i] = 1
data = data.drop(columns='work_type')
data.insert(5, 'work_type', work_type)
rows = [x for x in data.index if data.loc[x]['smoking_status'] == 'Unknown']
data = data.drop(rows, axis=0)
data = data.reset_index(drop=True)
smoking_status = np.zeros(data.shape[0])
for i in range(smoking_status.shape[0]):
if data['smoking_status'][i] == 'never smoked':
smoking_status[i] = 0
elif data['smoking_status'][i] == 'formerly smoked':
smoking_status[i] = 10
else:
smoking_status[i] = 20
data = data.drop(columns='smoking_status')
data.insert(9, 'smoking_status', smoking_status)
bmi_max = data.bmi.max()
bmi_min = data.bmi.min()
data['bmi'] = data['bmi'].apply(lambda x: (x - bmi_min) / ((bmi_max - bmi_min) / 10))
age_max = data.age.max()
age_min = data.age.min()
data['age'] = data['age'].apply(lambda x: (x - age_min) / ((age_max - age_min) / 10))
glucose_max = data.avg_glucose_level.max()
glucose_min = data.avg_glucose_level.min()
data['avg_glucose_level'] = data['avg_glucose_level'].apply(
lambda x: (x - glucose_min) / ((glucose_max - glucose_min) / 10))
return data
processed_data = data_process(original_data)
data = np.array(processed_data.iloc[:, :-1])
labels = np.expand_dims(np.array(processed_data.iloc[:, -1]), axis=-1)
train_data = data[0:3200, :]
train_label = labels[0:3200, :]
test_data = data[3200:, :]
test_label = labels[3200:, :]
model = k.Sequential()
model.add(k.layers.Dense(8, activation='relu',input_shape=(10,)))
model.add(k.layers.Dense(6, activation='relu'))
model.add(k.layers.Dense(4, activation='relu'))
model.add(k.layers.Dense(1, activation='sigmoid'))
model.compile(optimizer='adam',
loss='binary_crossentropy',
metrics=['acc'])
model.fit(train_data, train_label, epochs=10, batch_size=40, verbose=1, validation_data=(test_data, test_label))
model.evaluate(test_data, test_label)
我想要达到的结果
我想知道这是什么原因造成的，以及怎么修复？
