SCDM+SLSTM_TCNN.py

import seaborn as sns
from pylab import mpl
import numpy as np
from keras.utils import np_utils
import pandas as pd

from util.K_SCDM_SLSTM_TCNN import k_SCDM_SLSTM_TCNN
from util.data_ultimately import pick_arange

import time

"""
功能：交叉验证训练盐刺激浓度识别模型（SCDM）
1。保存每一折里训练的混淆矩阵
2。保存每次实验评价指标的均值和标准差
"""

sns.set(font_scale=2)
fontsize = 24
mpl.rcParams['font.family'] = 'sans-serif'
mpl.rcParams['font.sans-serif'] = 'NSimSun,Times New Roman'

wheat = 'LD'
expr_num = 1 # 进行几次实验，试验次数过多可能会保存，原因估计尚不清楚。
# df_T = pd.read_csv(f'Datasets/{wheat}_DSCP_666.csv')
df_T = pd.read_csv(f'Datasets/666条-{wheat}-0_400mM-9fenlei.csv')
df_T = df_T/10
df_T.to_csv('test.csv')
print(f'整个数据集的数量{df_T.shape[0]}')
# num=147
# test={} # 创建字典，保存转换结果
# for i in range(df_T.shape[0]):
#     ccc = np.array(df_T.iloc[i]) # 使用iloc切片不会出现问题。处理的数据是（None,1176）
#     test[i]=pick_arange(ccc,num) # 保存转换后的数据到字典
#     del ccc # 删除每一个循环里的变量
# df_T = pd.DataFrame(test,index=None).T


#区分ab波和标签
all_ = df_T.shape[1]
ab = all_ - 1

xx = np.array(df_T.iloc[:,0:ab])
yy = np.array(df_T.iloc[:,ab:all_])
yy2 = np.squeeze(yy) # 去掉一个空维度

yy_ = np_utils.to_categorical(yy2,9) # 标签向量化
xx_ = xx.reshape(df_T.shape[0], ab, 1) # 波形扩充一个维度


i_Mean_acc, i_Mean_kappa, i_Mean_ham= [],[],[]
i_Std_acc, i_Std_kappa, i_Std_ham= [],[],[]
test={}
for i in range(expr_num):

    # save_model：是否保存模型，1：保存 0： 不保存
    result = k_SCDM_SLSTM_TCNN(inputs=xx_, targets=yy_,epochs=600,wheat=wheat,save_model=1)

    test[f'{i}'] =result

now = int(round(time.time()*1000))
localtime = time.strftime('%Y-%m-%d %H:%M:%S',time.localtime(now/1000))
pd.DataFrame(test).to_csv(f"result/SCDM_SLSTM_TCNN_{wheat}_{localtime}.csv")