三月：心跳信号分类预测_baseline_v3：样本中每类别进行了均分

心跳信号分类预测_baseline_v3:每个标签参数进行了均分

# 2021.05.08
# 在v2基础上 让样本均匀分布，对小样本更有利
# 本文原创 望赞鼓励，转载请说明出处.

末尾有惊喜

总结为什么？度怎么把握

---

Datawhile三月选题：心跳信号分类

---

import os
import gc
import math

import pandas as pd
import numpy as np

import lightgbm as lgb
#import xgboost as xgb
from catboost import CatBoostRegressor
from sklearn.linear_model import SGDRegressor, LinearRegression, Ridge
from sklearn.preprocessing import MinMaxScaler


from sklearn.model_selection import StratifiedKFold, KFold
from sklearn.metrics import log_loss
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import OneHotEncoder

from tqdm import tqdm
import matplotlib.pyplot as plt
import time
import warnings
warnings.filterwarnings('ignore')

#reading dataset
train = pd.read_csv("train.csv")
test = pd.read_csv("testA.csv")

train

	id	heartbeat_signals	label
0	0	0.9912297987616655,0.9435330436439665,0.764677...	0.0
1	1	0.9714822034884503,0.9289687459588268,0.572932...	0.0
2	2	1.0,0.9591487564065292,0.7013782792997189,0.23...	2.0
3	3	0.9757952826275774,0.9340884687738161,0.659636...	0.0
4	4	0.0,0.055816398940721094,0.26129357194994196,0...	2.0
...	...	...	...
99995	99995	1.0,0.677705342021188,0.22239242747868546,0.25...	0.0
99996	99996	0.9268571578157265,0.9063471198026871,0.636993...	2.0
99997	99997	0.9258351628306013,0.5873839035878395,0.633226...	3.0
99998	99998	1.0,0.9947621698382489,0.8297017704865509,0.45...	2.0
99999	99999	0.9259994004527861,0.916476635326053,0.4042900...	0.0

100000 rows × 3 columns

def reduce_mem_usage(df):
    start_mem = df.memory_usage().sum() / 1024**2 
    print('Memory usage of dataframe is {:.2f} MB'.format(start_mem))
    
    for col in df.columns:
        col_type = df[col].dtype
        
        if col_type != object:
            c_min = df[col].min()
            c_max = df[col].max()
            if str(col_type)[:3] == 'int':
                if c_min > np.iinfo(np.int8).min and c_max < np.iinfo(np.int8).max:
                    df[col] = df[col].astype(np.int8)
                elif c_min > np.iinfo(np.int16).min and c_max < np.iinfo(np.int16).max:
                    df[col] = df[col].astype(np.int16)
                elif c_min > np.iinfo(np.int32).min and c_max < np.iinfo(np.int32).max:
                    df[col] = df[col].astype(np.int32)
                elif c_min > np.iinfo(np.int64).min and c_max < np.iinfo(np.int64).max:
                    df[col] = df[col].astype(np.int64)  
            else:
                if c_min > np.finfo(np.float16).min and c_max < np.finfo(np.float16).max:
                    df[col] = df[col].astype(np.float16)
                elif c_min > np.finfo(np.float32).min and c_max < np.finfo(np.float32).max:
                    df[col] = df[col].astype(np.float32)
                else:
                    df[col] = df[col].astype(np.float64)
        else:
            df[col] = df[col].astype('category')

    end_mem = df.memory_usage().sum() / 1024**2 
    print('Memory usage after optimization is: {:.2f} MB'.format(end_mem))
    print('Decreased by {:.1f}%'.format(100 * (start_mem - end_mem) / start_mem))
    
    return df

# 简单预处理
train_list = []

for items in train.values:
    train_list.append([items[0]] + [float(i) for i in items[1].split(',')] + [items[2]])

train = pd.DataFrame(np.array(train_list))
train.columns = ['id'] + ['s_'+str(i) for i in range(len(train_list[0])-2)] + ['label']
train = reduce_mem_usage(train)

test_list=[]
for items in test.values:
    test_list.append([items[0]] + [float(i) for i in items[1].split(',')])

test = pd.DataFrame(np.array(test_list))
test.columns = ['id'] + ['s_'+str(i) for i in range(len(test_list[0])-1)]
test = reduce_mem_usage(test)

Memory usage of dataframe is 157.93 MB
Memory usage after optimization is: 39.67 MB
Decreased by 74.9%
Memory usage of dataframe is 31.43 MB
Memory usage after optimization is: 7.90 MB
Decreased by 74.9%

test

	id	s_0	s_1	s_2	s_3	s_4	s_5	s_6	s_7	s_8	...	s_195	s_196	s_197	s_198	s_199	s_200	s_201	s_202	s_203	s_204
0	100000.0	0.991699	1.000000	0.631836	0.136230	0.041412	0.102722	0.120850	0.123413	0.107910	...	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.00000
1	100001.0	0.607422	0.541504	0.340576	0.000000	0.090698	0.164917	0.195068	0.168823	0.198853	...	0.389893	0.386963	0.367188	0.364014	0.360596	0.357178	0.350586	0.350586	0.350586	0.36377
2	100002.0	0.975098	0.670898	0.686523	0.708496	0.718750	0.716797	0.720703	0.701660	0.596680	...	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.00000
3	100003.0	0.995605	0.916992	0.520996	0.000000	0.221802	0.404053	0.490479	0.527344	0.518066	...	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.00000
4	100004.0	1.000000	0.888184	0.745605	0.531738	0.380371	0.224609	0.091125	0.057648	0.003914	...	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.00000
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
19995	119995.0	1.000000	0.833008	0.634277	0.639160	0.624023	0.598145	0.613770	0.624023	0.628906	...	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.00000
19996	119996.0	1.000000	0.826172	0.452148	0.082214	0.000000	0.137085	0.201050	0.165649	0.158081	...	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.00000
19997	119997.0	0.951660	0.916504	0.667480	0.352051	0.255371	0.197388	0.173584	0.141968	0.134521	...	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.00000
19998	119998.0	0.927734	0.677246	0.242920	0.055359	0.102112	0.072266	0.021011	0.038300	0.048553	...	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.00000
19999	119999.0	0.665527	0.526855	0.516602	0.376465	0.489258	0.480713	0.459229	0.482910	0.469971	...	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.00000

20000 rows × 206 columns

train

	id	s_0	s_1	s_2	s_3	s_4	s_5	s_6	s_7	s_8	...	s_196	s_197	s_198	s_199	s_200	s_201	s_202	s_203	s_204	label
0	0.0	0.991211	0.943359	0.764648	0.618652	0.379639	0.190796	0.040222	0.026001	0.031708	...	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0
1	1.0	0.971680	0.929199	0.572754	0.178467	0.122986	0.132324	0.094421	0.089600	0.030487	...	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0
2	2.0	1.000000	0.958984	0.701172	0.231812	0.000000	0.080688	0.128418	0.187500	0.280762	...	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	2.0
3	3.0	0.975586	0.934082	0.659668	0.249878	0.237061	0.281494	0.249878	0.249878	0.241455	...	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0
4	4.0	0.000000	0.055817	0.261230	0.359863	0.433105	0.453613	0.499023	0.542969	0.616699	...	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	2.0
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
99995	99995.0	1.000000	0.677734	0.222412	0.257080	0.204712	0.054657	0.026154	0.118164	0.244873	...	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0
99996	99996.0	0.926758	0.906250	0.637207	0.415039	0.374756	0.382568	0.358887	0.341309	0.336426	...	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	2.0
99997	99997.0	0.925781	0.587402	0.633301	0.632324	0.639160	0.614258	0.599121	0.517578	0.403809	...	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	3.0
99998	99998.0	1.000000	0.994629	0.829590	0.458252	0.264160	0.240234	0.213745	0.189331	0.203857	...	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	2.0
99999	99999.0	0.925781	0.916504	0.404297	0.000000	0.262939	0.385498	0.361084	0.332764	0.339844	...	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0

100000 rows × 207 columns

train['label'].value_counts()

0.0    64327
3.0    17912
2.0    14199
1.0     3562
Name: label, dtype: int64

#存在严重的数据不均衡

# 为均等数据处理做准备
df_1 = train[train['label'] == 1]
df_2 = train[train['label'] == 2]
df_3 = train[train['label'] == 3]

# df_1
# # 经检查没问题，取得是相应label

# 对数据量不足的填充 replace=ture意思是可以重复,随机种子要一致 不然每次随出来的都不一样 样本填充到20.000
from sklearn.utils import resample

df_1_upsample = resample(df_1, n_samples = 20000, replace = True, random_state = 123)
df_2_upsample = resample(df_2, n_samples = 20000, replace = True, random_state = 123)
df_3_upsample = resample(df_3, n_samples = 20000, replace = True, random_state = 123)

#0号lebal因为样本量大 所以随机出来20，000
df_0 = train[train['label']==0].sample(n =20000, random_state=123)

#把这四个样本合并成一个新的样本
#！！！！！ignore_index=True！！！一定要加 这个不加序号是乱的 后面再运行函数cv时候会报错
train_df = pd.concat([df_0, df_1_upsample, df_2_upsample, df_3_upsample],ignore_index=True)

#新样本 四个lebal这时候均分了，余下地方和V2一致
train_df['label'].value_counts()

0.0    20000
3.0    20000
1.0    20000
2.0    20000
Name: label, dtype: int64

train_df

	id	s_0	s_1	s_2	s_3	s_4	s_5	s_6	s_7	s_8	...	s_196	s_197	s_198	s_199	s_200	s_201	s_202	s_203	s_204	label
0	12037.0	0.981445	0.986816	0.558105	0.000000	0.295898	0.424805	0.431152	0.455322	0.448975	...	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0
1	1389.0	0.986328	0.864746	0.645996	0.352783	0.087463	0.023849	0.078979	0.109924	0.112671	...	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0
2	11045.0	0.000000	0.093140	0.274414	0.442139	0.564941	0.613770	0.726074	0.769531	0.796387	...	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0
3	4849.0	1.000000	0.999023	0.883789	0.583984	0.535156	0.607910	0.593750	0.596680	0.628418	...	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0
4	40943.0	0.991211	0.941895	0.336914	0.037476	0.223389	0.366699	0.380127	0.390625	0.366699	...	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
79995	16396.0	0.904297	0.866699	0.812988	0.765137	0.686035	0.593262	0.460449	0.328125	0.197754	...	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	3.0
79996	48610.0	0.914551	0.879883	0.831543	0.785645	0.727539	0.648926	0.561035	0.456055	0.327881	...	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	3.0
79997	31521.0	0.753906	0.655762	0.537109	0.412842	0.242432	0.144897	0.051422	0.024811	0.044800	...	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	3.0
79998	69413.0	0.664062	0.611328	0.588379	0.540527	0.516113	0.466064	0.428711	0.343994	0.286865	...	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	3.0
79999	25387.0	0.857910	0.773926	0.676758	0.550293	0.393066	0.213257	0.102844	0.001230	0.000549	...	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	3.0

80000 rows × 207 columns

# #这里做相应的改变，是从新样本中取
# x_train = train.drop(['id','label'], axis=1)
# y_train = train['label']
# x_test=test.drop(['id'], axis=1)

#这里做相应的改变，是从新样本中取
x_train = train_df.drop(['id','label'], axis=1)
y_train = train_df['label']
x_test=test.drop(['id'], axis=1)

x_train

	s_0	s_1	s_2	s_3	s_4	s_5	s_6	s_7	s_8	s_9	...	s_195	s_196	s_197	s_198	s_199	s_200	s_201	s_202	s_203	s_204
0	0.981445	0.986816	0.558105	0.000000	0.295898	0.424805	0.431152	0.455322	0.448975	0.447266	...	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0
1	0.986328	0.864746	0.645996	0.352783	0.087463	0.023849	0.078979	0.109924	0.112671	0.098755	...	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0
2	0.000000	0.093140	0.274414	0.442139	0.564941	0.613770	0.726074	0.769531	0.796387	0.782715	...	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0
3	1.000000	0.999023	0.883789	0.583984	0.535156	0.607910	0.593750	0.596680	0.628418	0.619141	...	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0
4	0.991211	0.941895	0.336914	0.037476	0.223389	0.366699	0.380127	0.390625	0.366699	0.387939	...	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
79995	0.904297	0.866699	0.812988	0.765137	0.686035	0.593262	0.460449	0.328125	0.197754	0.150024	...	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0
79996	0.914551	0.879883	0.831543	0.785645	0.727539	0.648926	0.561035	0.456055	0.327881	0.238647	...	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0
79997	0.753906	0.655762	0.537109	0.412842	0.242432	0.144897	0.051422	0.024811	0.044800	0.111633	...	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0
79998	0.664062	0.611328	0.588379	0.540527	0.516113	0.466064	0.428711	0.343994	0.286865	0.125244	...	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0
79999	0.857910	0.773926	0.676758	0.550293	0.393066	0.213257	0.102844	0.001230	0.000549	0.035858	...	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0

80000 rows × 205 columns

x_test

	s_0	s_1	s_2	s_3	s_4	s_5	s_6	s_7	s_8	s_9	...	s_195	s_196	s_197	s_198	s_199	s_200	s_201	s_202	s_203	s_204
0	0.991699	1.000000	0.631836	0.136230	0.041412	0.102722	0.120850	0.123413	0.107910	0.110535	...	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.00000
1	0.607422	0.541504	0.340576	0.000000	0.090698	0.164917	0.195068	0.168823	0.198853	0.153564	...	0.389893	0.386963	0.367188	0.364014	0.360596	0.357178	0.350586	0.350586	0.350586	0.36377
2	0.975098	0.670898	0.686523	0.708496	0.718750	0.716797	0.720703	0.701660	0.596680	0.487061	...	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.00000
3	0.995605	0.916992	0.520996	0.000000	0.221802	0.404053	0.490479	0.527344	0.518066	0.545410	...	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.00000
4	1.000000	0.888184	0.745605	0.531738	0.380371	0.224609	0.091125	0.057648	0.003914	0.007820	...	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.00000
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
19995	1.000000	0.833008	0.634277	0.639160	0.624023	0.598145	0.613770	0.624023	0.628906	0.624023	...	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.00000
19996	1.000000	0.826172	0.452148	0.082214	0.000000	0.137085	0.201050	0.165649	0.158081	0.165649	...	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.00000
19997	0.951660	0.916504	0.667480	0.352051	0.255371	0.197388	0.173584	0.141968	0.134521	0.127075	...	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.00000
19998	0.927734	0.677246	0.242920	0.055359	0.102112	0.072266	0.021011	0.038300	0.048553	0.017532	...	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.00000
19999	0.665527	0.526855	0.516602	0.376465	0.489258	0.480713	0.459229	0.482910	0.469971	0.399170	...	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.00000

20000 rows × 205 columns

y_train

0        0.0
1        0.0
2        0.0
3        0.0
4        0.0
        ... 
79995    3.0
79996    3.0
79997    3.0
79998    3.0
79999    3.0
Name: label, Length: 80000, dtype: float16

# loss函数
def abs_sum(y_pre,y_tru):
    y_pre=np.array(y_pre)
    y_tru=np.array(y_tru)
    loss=sum(sum(abs(y_pre-y_tru)))
    return loss

from keras.models import Sequential
from keras.layers import Dense # for fully connected layers dense will be used
from keras.layers import Conv1D, MaxPooling1D, Flatten
from keras.optimizers import Adam

# avoid overfitting by normalizing the samples
from keras.layers.normalization import BatchNormalization

# cnn
def build_model():
    model = Sequential()
    

    #过滤器=神经元总数中的单位
    #Padding='相同'，零填充，在输入数据周围添加零像素
    model.add(Conv1D(filters = 64, kernel_size = 6, activation='relu', padding = 'same', input_shape = (205, 1))) #we pass individual values hence not 100000,187,1
    #：（none,205,64）
    
    # Normalization to avoid overfitting
    # 大概就是保持数据的敏感 ，加在全连接函数和激励函数之间：https://www.bilibili.com/video/BV1Lx411j7GT?from=search&seid=5048435414489430319
    model.add(BatchNormalization())
    # #：（none,205,64）#只是把数据重初始化一下，也就是重新分布了一下，为了使得这些数在下面的激励函数中区分的更大
    
    # Pooling 
    model.add(MaxPooling1D(pool_size=(3), strides = (2), padding = 'same'))
    # ：（none,103,64）# 因为strides是2 一下跳两步 padding=“same" 所以变化前特征总数不变 那么输出就是187/2=94

    model.add(Conv1D(filters = 64, kernel_size = 6, activation='relu', padding = 'same'))# （none, 103, 64）
    model.add(BatchNormalization())                                                      #:（none, 103, 64）
    model.add(MaxPooling1D(pool_size=(3), strides = (2), padding = 'same'))              #:（none, 52, 64）

    model.add(Conv1D( filters = 64, kernel_size = 6, activation='relu', padding = 'same'))#:（none, 52, 64）
    model.add(BatchNormalization())                                                       #:（none, 52, 64）
    model.add(MaxPooling1D(pool_size=(3), strides = (2), padding = 'same'))               #:（none, 26, 64）

    # Flatten 
    model.add(Flatten())
    #:(none, 1664) 这一步就是把多维的数据铺平 26*64 = 1664

    # Fully connected layer
    # input layer
    #Dense（用来写输出层）
    #当输入序列的长度固定时，该值为其长度。如果要在该层后接Flatten层，然后接Dense层，则必须指定该参数，否则Dense层的输出维度无法自动推断。
    #units：大于0的整数，代表该层的输出维度。
    model.add(Dense(units = 64, activation='relu'))
    
    # Hidden Layer
    model.add(Dense(units = 64, activation='relu'))
    
    # Output Layer
    model.add(Dense(units = 4, activation='softmax'))

    # loss = 'categorical_crossentropy'
    model.compile(optimizer = 'Adam', loss = 'categorical_crossentropy', metrics = ['accuracy'])
    return model

model = build_model()
# This is for one sample, i.e. one row
model.summary()

Model: "sequential"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv1d (Conv1D)              (None, 205, 64)           448       
_________________________________________________________________
batch_normalization (BatchNo (None, 205, 64)           256       
_________________________________________________________________
max_pooling1d (MaxPooling1D) (None, 103, 64)           0         
_________________________________________________________________
conv1d_1 (Conv1D)            (None, 103, 64)           24640     
_________________________________________________________________
batch_normalization_1 (Batch (None, 103, 64)           256       
_________________________________________________________________
max_pooling1d_1 (MaxPooling1 (None, 52, 64)            0         
_________________________________________________________________
conv1d_2 (Conv1D)            (None, 52, 64)            24640     
_________________________________________________________________
batch_normalization_2 (Batch (None, 52, 64)            256       
_________________________________________________________________
max_pooling1d_2 (MaxPooling1 (None, 26, 64)            0         
_________________________________________________________________
flatten (Flatten)            (None, 1664)              0         
_________________________________________________________________
dense (Dense)                (None, 64)                106560    
_________________________________________________________________
dense_1 (Dense)              (None, 64)                4160      
_________________________________________________________________
dense_2 (Dense)              (None, 4)                 260       
=================================================================
Total params: 161,476
Trainable params: 161,092
Non-trainable params: 384
_________________________________________________________________

from keras.utils.np_utils import to_categorical

def cv_model(clf, train_x, train_y, test_x, clf_name):
    folds = 4
    seed = 2021
    
    #k-交叉验证KFold
    # n_split:要划分的折数
    #shuffle: 每次都进行shuffle，测试集中折数的总和就是训练集的个数
    #random_state:随机状态 总结：对于那些本质上是随机的过程，我们有必要控制随机的状态，这样才能重复的展现相同的结果。
    #如果，对随机状态不加控制，那么实验的结果就无法固定，而是随机的显现。比喻的说一下，也不知道准不准确。
    # 一个容器中放置一定量的沙子，每次用手去抓沙子的时候，抓取的结果会受到抓取的力度、一只手抓还是两只手抓、手是干的或湿的等诸多因素的影响（将影响因素定为A={a,b,c,d,e,f,……}）。
    #固定random_state后，每次抓取沙子时的影响因素将被固定为具体的某一组，这样每次抓取的沙子就具有了相同的状态。

    
    kf = KFold(n_splits=folds, shuffle=True, random_state=seed)
    #为预测做准备
    test = np.zeros((test_x.shape[0],4))
    test_x = test_x.iloc[:, :].values
    test_x = test_x.reshape(len(test_x),test_x.shape[1],1)    

    cv_scores = []
    

    onehot_encoder = OneHotEncoder(sparse=False)
    

    for i, (train_index, test_index) in enumerate(kf.split(x_train, y_train)):
        
        print('************************************ {} ************************************'.format(str(i+1)))
        
        x_kf_train, y_kf_train, x_kf_test, y_kf_test = x_train.iloc[train_index], y_train[train_index], x_train.iloc[test_index], y_train[test_index]

        
        if clf_name == "cnn":
            x_kf_train = x_kf_train.iloc[:, :].values
            x_kf_test = x_kf_test.iloc[:, :].values
            x_kf_train = x_kf_train.reshape(len(x_kf_train),x_kf_train.shape[1],1)
            x_kf_test = x_kf_test.reshape(len(x_kf_test),x_kf_test.shape[1],1)
            
            
            y_kf_train = to_categorical(y_kf_train)
            y_kf_test = to_categorical(y_kf_test)
            
            history = model.fit(x_kf_train,y_kf_train, epochs = 15, batch_size = 32, validation_data=(x_kf_test, y_kf_test))
            x_kf_test_pred = model.predict(x_kf_test)
            test_pred = model.predict(test_x)
            
    
      
            
        
        print("y_kf_test++++++:")  
        print(y_kf_test)

        print('预测的概率矩阵为test_pred：')
        print(test_pred)
        print("abs_sum++++++:")
        score=abs_sum(y_kf_test,  x_kf_test_pred)
        cv_scores.append(score)
        print("cv_scores+++++:")
        print(cv_scores)
        
#         test += test_pred
#         test=test/kf.n_splits
        # 将将要预测的带到

            


    return test_pred

def lgb_model(x_train, y_train, x_test):
    lgb_test = cv_model(lgb, x_train, y_train, x_test, "cnn")
    return lgb_test
lgb_test = lgb_model(x_train, y_train, x_test)

************************************ 1 ************************************
Epoch 1/15
1875/1875 [==============================] - 60s 31ms/step - loss: 0.2759 - accuracy: 0.8949 - val_loss: 0.1356 - val_accuracy: 0.9588
Epoch 2/15
1875/1875 [==============================] - 59s 32ms/step - loss: 0.0875 - accuracy: 0.9688 - val_loss: 0.0670 - val_accuracy: 0.9768
Epoch 3/15
1875/1875 [==============================] - 56s 30ms/step - loss: 0.0539 - accuracy: 0.9812 - val_loss: 0.0565 - val_accuracy: 0.9797
Epoch 4/15
1875/1875 [==============================] - 56s 30ms/step - loss: 0.0407 - accuracy: 0.9865 - val_loss: 0.0454 - val_accuracy: 0.9859
Epoch 5/15
1875/1875 [==============================] - 56s 30ms/step - loss: 0.0323 - accuracy: 0.9891 - val_loss: 0.0395 - val_accuracy: 0.9867
Epoch 6/15
1875/1875 [==============================] - 57s 30ms/step - loss: 0.0233 - accuracy: 0.9924 - val_loss: 0.0342 - val_accuracy: 0.9895
Epoch 7/15
1875/1875 [==============================] - 56s 30ms/step - loss: 0.0232 - accuracy: 0.9923 - val_loss: 0.0394 - val_accuracy: 0.9868
Epoch 8/15
1875/1875 [==============================] - 57s 30ms/step - loss: 0.0151 - accuracy: 0.9951 - val_loss: 0.0416 - val_accuracy: 0.9880
Epoch 9/15
1875/1875 [==============================] - 57s 30ms/step - loss: 0.0168 - accuracy: 0.9952 - val_loss: 0.0564 - val_accuracy: 0.9827
Epoch 10/15
1875/1875 [==============================] - 55s 29ms/step - loss: 0.0196 - accuracy: 0.9939 - val_loss: 0.0274 - val_accuracy: 0.9926
Epoch 11/15
1875/1875 [==============================] - 56s 30ms/step - loss: 0.0115 - accuracy: 0.9960 - val_loss: 0.0267 - val_accuracy: 0.9938
Epoch 12/15
1875/1875 [==============================] - 56s 30ms/step - loss: 0.0087 - accuracy: 0.9973 - val_loss: 0.0472 - val_accuracy: 0.9862
Epoch 13/15
1875/1875 [==============================] - 59s 31ms/step - loss: 0.0097 - accuracy: 0.9967 - val_loss: 0.0334 - val_accuracy: 0.9923
Epoch 14/15
1875/1875 [==============================] - 57s 30ms/step - loss: 0.0091 - accuracy: 0.9970 - val_loss: 0.0478 - val_accuracy: 0.9890
Epoch 15/15
1875/1875 [==============================] - 57s 30ms/step - loss: 0.0133 - accuracy: 0.9961 - val_loss: 0.0329 - val_accuracy: 0.9912
y_kf_test++++++:
[[1. 0. 0. 0.]
 [1. 0. 0. 0.]
 [1. 0. 0. 0.]
 ...
 [0. 0. 0. 1.]
 [0. 0. 0. 1.]
 [0. 0. 0. 1.]]
预测的概率矩阵为test_pred：
[[9.9999166e-01 8.3696723e-06 7.0094308e-09 1.5220067e-08]
 [6.5147565e-10 2.9474978e-09 1.0000000e+00 3.8487034e-15]
 [3.4868474e-20 3.0091077e-13 3.5588698e-16 1.0000000e+00]
 ...
 [9.6342439e-05 3.3096262e-06 9.9989974e-01 5.9694372e-07]
 [9.9999988e-01 1.6789036e-07 1.0092740e-08 1.7682024e-11]
 [8.7801713e-01 9.7556551e-07 2.5424310e-03 1.1943953e-01]]
abs_sum++++++:
cv_scores+++++:
[417.5972366333008]
************************************ 2 ************************************
Epoch 1/15
1875/1875 [==============================] - 58s 31ms/step - loss: 0.0186 - accuracy: 0.9949 - val_loss: 0.0089 - val_accuracy: 0.9963
Epoch 2/15
1875/1875 [==============================] - 58s 31ms/step - loss: 0.0116 - accuracy: 0.9965 - val_loss: 0.0138 - val_accuracy: 0.9955
Epoch 3/15
1875/1875 [==============================] - 58s 31ms/step - loss: 0.0106 - accuracy: 0.9967 - val_loss: 0.0207 - val_accuracy: 0.9941
Epoch 4/15
1875/1875 [==============================] - 58s 31ms/step - loss: 0.0093 - accuracy: 0.9971 - val_loss: 0.0101 - val_accuracy: 0.9964
Epoch 5/15
1875/1875 [==============================] - 57s 31ms/step - loss: 0.0070 - accuracy: 0.9978 - val_loss: 0.0196 - val_accuracy: 0.9949
Epoch 6/15
1875/1875 [==============================] - 57s 31ms/step - loss: 0.0073 - accuracy: 0.9977 - val_loss: 0.0124 - val_accuracy: 0.9969
Epoch 7/15
1875/1875 [==============================] - 58s 31ms/step - loss: 0.0078 - accuracy: 0.9975 - val_loss: 0.0151 - val_accuracy: 0.9956
Epoch 8/15
1875/1875 [==============================] - 61s 32ms/step - loss: 0.0075 - accuracy: 0.9977 - val_loss: 0.0302 - val_accuracy: 0.9913
Epoch 9/15
1875/1875 [==============================] - 60s 32ms/step - loss: 0.0065 - accuracy: 0.9982 - val_loss: 0.0090 - val_accuracy: 0.9971
Epoch 10/15
1875/1875 [==============================] - 58s 31ms/step - loss: 0.0065 - accuracy: 0.9982 - val_loss: 0.0263 - val_accuracy: 0.9945
Epoch 11/15
1875/1875 [==============================] - 57s 31ms/step - loss: 0.0070 - accuracy: 0.9978 - val_loss: 0.0103 - val_accuracy: 0.9970
Epoch 12/15
1875/1875 [==============================] - 58s 31ms/step - loss: 0.0056 - accuracy: 0.9985 - val_loss: 0.0134 - val_accuracy: 0.9965
Epoch 13/15
1875/1875 [==============================] - 58s 31ms/step - loss: 0.0061 - accuracy: 0.9983 - val_loss: 0.0128 - val_accuracy: 0.9963
Epoch 14/15
1875/1875 [==============================] - 59s 31ms/step - loss: 0.0069 - accuracy: 0.9981 - val_loss: 0.0295 - val_accuracy: 0.9904
Epoch 15/15
1875/1875 [==============================] - 57s 30ms/step - loss: 0.0052 - accuracy: 0.9988 - val_loss: 0.0134 - val_accuracy: 0.9967
y_kf_test++++++:
[[1. 0. 0. 0.]
 [1. 0. 0. 0.]
 [1. 0. 0. 0.]
 ...
 [0. 0. 0. 1.]
 [0. 0. 0. 1.]
 [0. 0. 0. 1.]]
预测的概率矩阵为test_pred：
[[9.99999762e-01 2.31557891e-07 1.64789016e-12 1.57000205e-10]
 [7.08220592e-13 3.05859255e-10 1.00000000e+00 1.08148854e-23]
 [5.37503151e-19 4.81650921e-20 5.44116887e-20 1.00000000e+00]
 ...
 [2.33657775e-03 2.78295577e-03 9.94879365e-01 1.02097715e-06]
 [9.99999523e-01 4.99625287e-07 2.50645789e-11 2.91075359e-08]
 [1.00000000e+00 1.97536454e-09 4.46538850e-09 6.42643183e-09]]
abs_sum++++++:
cv_scores+++++:
[417.5972366333008, 166.70497226715088]
************************************ 3 ************************************
Epoch 1/15
1875/1875 [==============================] - 58s 31ms/step - loss: 0.0103 - accuracy: 0.9975 - val_loss: 0.0062 - val_accuracy: 0.9985
Epoch 2/15
1875/1875 [==============================] - 58s 31ms/step - loss: 0.0042 - accuracy: 0.9988 - val_loss: 0.0129 - val_accuracy: 0.9963
Epoch 3/15
1875/1875 [==============================] - 57s 30ms/step - loss: 0.0071 - accuracy: 0.9978 - val_loss: 0.0027 - val_accuracy: 0.9992
Epoch 4/15
1875/1875 [==============================] - 61s 32ms/step - loss: 0.0042 - accuracy: 0.9987 - val_loss: 0.0055 - val_accuracy: 0.9982
Epoch 5/15
1875/1875 [==============================] - 58s 31ms/step - loss: 0.0061 - accuracy: 0.9984 - val_loss: 0.0116 - val_accuracy: 0.9963
Epoch 6/15
1875/1875 [==============================] - 57s 30ms/step - loss: 0.0048 - accuracy: 0.9986 - val_loss: 0.0055 - val_accuracy: 0.9984
Epoch 7/15
1875/1875 [==============================] - 57s 31ms/step - loss: 0.0045 - accuracy: 0.9991 - val_loss: 0.0084 - val_accuracy: 0.9984
Epoch 8/15
1875/1875 [==============================] - 58s 31ms/step - loss: 0.0062 - accuracy: 0.9983 - val_loss: 0.0036 - val_accuracy: 0.9989
Epoch 9/15
1875/1875 [==============================] - 58s 31ms/step - loss: 0.0039 - accuracy: 0.9989 - val_loss: 0.0091 - val_accuracy: 0.9973
Epoch 10/15
1875/1875 [==============================] - 58s 31ms/step - loss: 0.0054 - accuracy: 0.9986 - val_loss: 0.0167 - val_accuracy: 0.9957
Epoch 11/15
1875/1875 [==============================] - 57s 31ms/step - loss: 0.0061 - accuracy: 0.9986 - val_loss: 0.0188 - val_accuracy: 0.9967
Epoch 12/15
1875/1875 [==============================] - 58s 31ms/step - loss: 0.0025 - accuracy: 0.9991 - val_loss: 0.0162 - val_accuracy: 0.9948
Epoch 13/15
1875/1875 [==============================] - 57s 30ms/step - loss: 0.0050 - accuracy: 0.9986 - val_loss: 0.0081 - val_accuracy: 0.9983
Epoch 14/15
1875/1875 [==============================] - 60s 32ms/step - loss: 0.0047 - accuracy: 0.9988 - val_loss: 0.0143 - val_accuracy: 0.9962
Epoch 15/15
1875/1875 [==============================] - 58s 31ms/step - loss: 0.0046 - accuracy: 0.9989 - val_loss: 0.0293 - val_accuracy: 0.9941
y_kf_test++++++:
[[1. 0. 0. 0.]
 [1. 0. 0. 0.]
 [1. 0. 0. 0.]
 ...
 [0. 0. 0. 1.]
 [0. 0. 0. 1.]
 [0. 0. 0. 1.]]
预测的概率矩阵为test_pred：
[[9.9999940e-01 5.5716970e-07 4.2182435e-10 7.5162426e-15]
 [3.0866704e-10 2.2008441e-08 1.0000000e+00 3.8451365e-23]
 [3.2547221e-23 1.0347285e-20 5.9116691e-22 1.0000000e+00]
 ...
 [7.4469927e-04 4.4713984e-06 9.9902654e-01 2.2425935e-04]
 [1.0000000e+00 6.6102653e-12 9.1913356e-16 7.5038865e-26]
 [9.9999976e-01 2.0411268e-07 3.4282695e-11 1.1750135e-11]]
abs_sum++++++:
cv_scores+++++:
[417.5972366333008, 166.70497226715088, 286.96989250183105]
************************************ 4 ************************************
Epoch 1/15
1875/1875 [==============================] - 57s 30ms/step - loss: 0.0064 - accuracy: 0.9982 - val_loss: 0.0071 - val_accuracy: 0.9980
Epoch 2/15
1875/1875 [==============================] - 57s 30ms/step - loss: 0.0039 - accuracy: 0.9991 - val_loss: 0.0132 - val_accuracy: 0.9973
Epoch 3/15
1875/1875 [==============================] - 57s 30ms/step - loss: 0.0060 - accuracy: 0.9983 - val_loss: 0.0273 - val_accuracy: 0.9916
Epoch 4/15
1875/1875 [==============================] - 59s 31ms/step - loss: 0.0042 - accuracy: 0.9989 - val_loss: 0.0024 - val_accuracy: 0.9991
Epoch 5/15
1875/1875 [==============================] - 57s 31ms/step - loss: 0.0030 - accuracy: 0.9992 - val_loss: 0.0026 - val_accuracy: 0.9994
Epoch 6/15
1875/1875 [==============================] - 60s 32ms/step - loss: 0.0038 - accuracy: 0.9991 - val_loss: 0.0056 - val_accuracy: 0.9982
Epoch 7/15
1875/1875 [==============================] - 62s 33ms/step - loss: 0.0036 - accuracy: 0.9991 - val_loss: 0.0026 - val_accuracy: 0.9991
Epoch 8/15
1875/1875 [==============================] - 58s 31ms/step - loss: 0.0042 - accuracy: 0.9987 - val_loss: 0.0255 - val_accuracy: 0.9923
Epoch 9/15
1875/1875 [==============================] - 56s 30ms/step - loss: 0.0042 - accuracy: 0.9990 - val_loss: 0.0107 - val_accuracy: 0.9974
Epoch 10/15
1875/1875 [==============================] - 54s 29ms/step - loss: 0.0034 - accuracy: 0.9991 - val_loss: 0.0022 - val_accuracy: 0.9992
Epoch 11/15
1875/1875 [==============================] - 53s 29ms/step - loss: 0.0039 - accuracy: 0.9989 - val_loss: 0.0222 - val_accuracy: 0.9959
Epoch 12/15
1875/1875 [==============================] - 54s 29ms/step - loss: 0.0024 - accuracy: 0.9992 - val_loss: 0.0033 - val_accuracy: 0.9990
Epoch 13/15
1875/1875 [==============================] - 55s 29ms/step - loss: 0.0033 - accuracy: 0.9991 - val_loss: 0.0153 - val_accuracy: 0.9955
Epoch 14/15
1875/1875 [==============================] - 53s 28ms/step - loss: 0.0037 - accuracy: 0.9990 - val_loss: 0.0063 - val_accuracy: 0.9982
Epoch 15/15
1875/1875 [==============================] - 55s 29ms/step - loss: 0.0027 - accuracy: 0.9991 - val_loss: 0.0042 - val_accuracy: 0.9991
y_kf_test++++++:
[[1. 0. 0. 0.]
 [1. 0. 0. 0.]
 [1. 0. 0. 0.]
 ...
 [0. 0. 0. 1.]
 [0. 0. 0. 1.]
 [0. 0. 0. 1.]]
预测的概率矩阵为test_pred：
[[1.0000000e+00 1.2987890e-12 4.3421242e-17 9.6314740e-25]
 [1.0664977e-16 1.0852620e-15 1.0000000e+00 3.7733482e-19]
 [2.0541892e-26 1.0776717e-29 6.0904610e-24 1.0000000e+00]
 ...
 [2.2533643e-10 1.9210296e-10 1.0000000e+00 6.5144643e-12]
 [1.0000000e+00 8.0374228e-11 8.2770508e-16 1.1548383e-20]
 [1.0000000e+00 3.1995117e-15 3.1809543e-14 2.6292750e-17]]
abs_sum++++++:
cv_scores+++++:
[417.5972366333008, 166.70497226715088, 286.96989250183105, 46.44265604019165]

temp=pd.DataFrame(lgb_test)
result=pd.read_csv('sample_submit.csv')
result['label_0']=temp[0]
result['label_1']=temp[1]
result['label_2']=temp[2]
result['label_3']=temp[3]
result.to_csv('submit_baseline_v3.0.csv',index=False)

submit_data=pd.read_csv('submit_baseline_v3.0.csv')

submit_data

	id	label_0	label_1	label_2	label_3
0	100000	1.000000e+00	1.298789e-12	4.342124e-17	9.631474e-25
1	100001	1.066498e-16	1.085262e-15	1.000000e+00	3.773348e-19
2	100002	2.054189e-26	1.077672e-29	6.090461e-24	1.000000e+00
3	100003	1.000000e+00	5.491751e-15	4.521685e-22	1.850977e-25
4	100004	1.000000e+00	2.030157e-10	9.926018e-13	1.117518e-13
...	...	...	...	...	...
19995	119995	1.000000e+00	2.684174e-08	8.052518e-12	5.653323e-13
19996	119996	1.000000e+00	9.495173e-12	9.100256e-13	6.368752e-20
19997	119997	2.253364e-10	1.921030e-10	1.000000e+00	6.514464e-12
19998	119998	1.000000e+00	8.037423e-11	8.277051e-16	1.154838e-20
19999	119999	1.000000e+00	3.199512e-15	3.180954e-14	2.629275e-17

20000 rows × 5 columns

for index,row in submit_data.iterrows():
    row_max = max(list(row)[1:])
    if row_max > 0.9:
        for i in range(1,5):
            if row[i]>0.9:
                submit_data.iloc[index,i] = 1
            else:
                submit_data.iloc[index,i] = 0
submit_data

	id	label_0	label_1	label_2	label_3
0	100000	1.0	0.0	0.0	0.0
1	100001	0.0	0.0	1.0	0.0
2	100002	0.0	0.0	0.0	1.0
3	100003	1.0	0.0	0.0	0.0
4	100004	1.0	0.0	0.0	0.0
...	...	...	...	...	...
19995	119995	1.0	0.0	0.0	0.0
19996	119996	1.0	0.0	0.0	0.0
19997	119997	0.0	0.0	1.0	0.0
19998	119998	1.0	0.0	0.0	0.0
19999	119999	1.0	0.0	0.0	0.0

20000 rows × 5 columns

submit_data.to_csv('submit_baseline_v3.0.1.csv',index=False)

V3分数降了一百多分！！！！