天池大赛之工业蒸汽处理(改进版 ---- 0.1235）_天池大赛 工业蒸汽 深度学习

导包

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
%matplotlib inline
import seaborn as sns
from sklearn.linear_model import LinearRegression,Lasso,Ridge,ElasticNet,RidgeCV
from sklearn.neighbors import KNeighborsRegressor
from sklearn.ensemble import GradientBoostingRegressor,RandomForestRegressor,AdaBoostRegressor,ExtraTreesRegressor
from xgboost import XGBRegressor
from lightgbm import LGBMRegressor
#支持向量机
from sklearn.svm import SVR
from sklearn.metrics import mean_squared_error
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import MinMaxScaler,StandardScaler,PolynomialFeatures
import warnings
warnings.filterwarnings('ignore')
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17

数据聚合

train_data = pd.read_csv('./zhengqi_train.txt',sep='\t')
test_data = pd.read_csv('./zhengqi_test.txt',sep='\t')
1
2

#合并训练数据和预测数据
train_data["origin"]="train"
test_data["origin"]="test"
data_all=pd.concat([train_data,test_data],axis=0,ignore_index=True)
#View data
data_all
1
2
3
4
5
6

4813 rows × 40 columns

特征探索

#探索出去最后两列的数字属性
data_all.columns[:-2]
1
2

Index(['V0', 'V1', 'V10', 'V11', 'V12', 'V13', 'V14', 'V15', 'V16', 'V17',
       'V18', 'V19', 'V2', 'V20', 'V21', 'V22', 'V23', 'V24', 'V25', 'V26',
       'V27', 'V28', 'V29', 'V3', 'V30', 'V31', 'V32', 'V33', 'V34', 'V35',
       'V36', 'V37', 'V4', 'V5', 'V6', 'V7', 'V8', 'V9'],
      dtype='object')
1
2
3
4
5

#38个特征将一些不重要的删除
#特征分布情况，训练和测试数据特征分布不均匀，删除
for column in data_all.columns[0:-2]:
    g = sns.kdeplot(data_all[column][(data_all["origin"] == "train")], color="Red", shade = True)
    g = sns.kdeplot(data_all[column][(data_all["origin"] == "test")], ax =g, color="Blue", shade= True)
    g.set_xlabel(column)
    g.set_ylabel("Frequency")
    g = g.legend(["train","test"])
    plt.show()
1
2
3
4
5
6
7
8
9

fig = plt.figure(figsize=(10, 10))
for i in range(len(data_all.columns)-2):
    g = sns.FacetGrid(data_all, col='origin')
    g = g.map(sns.distplot, data_all.columns[i])
1
2
3
4

<Figure size 720x720 with 0 Axes>
1

#通过图示可以看出'V11','V17','V22','V5'，波动太大，删除
drop_labels = ['V11','V17','V22','V5']
data_all.drop(drop_labels,axis=1,inplace=True)
1
2
3

相关性系数corr

# 找出相关程度
plt.figure(figsize=(20, 16))  # 指定绘图对象宽度和高度
mcorr = train_data.corr()  # 相关系数矩阵，即给出了任意两个变量之间的相关系数
mask = np.zeros_like(mcorr, dtype=np.bool)  # 构造与mcorr同维数矩阵 为bool型
mask[np.triu_indices_from(mask)] = True  # 角分线右侧为True
cmap = sns.diverging_palette(220, 10, as_cmap=True)  # 返回matplotlib colormap对象
g = sns.heatmap(mcorr, mask=mask, cmap=cmap, square=True, annot=True, fmt='0.2f')  # 热力图（看两两相似度）
plt.show
1
2
3
4
5
6
7
8

<function matplotlib.pyplot.show(*args, **kw)>
1

在这里插入图片描述

# 通过相关性系数找到7个相关性不大的属性
cond = mcorr.loc['target'].abs()<0.1
drop_labels = mcorr.loc['target'][cond].index
#['V14', 'V21', 'V25', 'V26', 'V32', 'V33', 'V34']

#查看属性分布后，将分布不好的删除  （'V14', 'V21', ）
drop_labels = ['V14', 'V21']
data_all.drop(drop_labels,axis=1,inplace=True)
1
2
3
4
5
6
7
8

#删除了6个属性
data_all.shape
1
2

(4813, 34)
1

对数据进行归一化

data = data_all.iloc[:,:-2]
minmaxscale = MinMaxScaler()
data = minmaxscale.fit_transform(data)
data
1
2
3
4

array([[0.77577505, 0.723449  , 0.22174265, ..., 0.43285165, 0.66410771,
        0.73528007],
       [0.83374189, 0.77878549, 0.37388724, ..., 0.43285165, 0.7548128 ,
        0.73528007],
       [0.84023071, 0.79600421, 0.46641489, ..., 0.43285165, 0.76237156,
        0.73528007],
       ...,
       [0.31708724, 0.25289169, 0.0074184 , ..., 0.17367095, 0.10192512,
        0.64706284],
       [0.31045422, 0.24211356, 0.00323712, ..., 0.24075302, 0.1563718 ,
        0.67646858],
       [0.35948089, 0.32216088, 0.35608309, ..., 0.24897256, 0.19971655,
        0.67646858]])
1
2
3
4
5
6
7
8
9
10
11
12
13

#归一化数据
data_all_norm = pd.DataFrame(data,columns=data_all.columns[:-2])
data_all_norm
1
2
3

4813 rows × 32 columns

#将oringin和target属性merage上
data_all_norm = pd.merge(data_all_norm,data_all.iloc[:,-2:],left_index=True,right_index=True)
1
2

data_all_norm.describe()
1

def scale_minmax(data):
    return (data - data.min())/(data.max() - data.min())
1
2
'
运行
运行

#使用Box-Cox将连续数据转换的更加平滑（主要处理类似正太分布）
from scipy import stats
fcols = 6
frows = len(data_all_norm.columns[:10])
plt.figure(figsize=(4*fcols,4*frows))
i = 0

for col in data_all_norm.columns[:10]:
    dat = data_all_norm[[col, 'target']].dropna()

#     这条线就是数据分布dist：distribution（分布）
    i+=1
    plt.subplot(frows,fcols,i)
    sns.distplot(dat[col],fit = stats.norm);
    plt.title(col+' Original')
    plt.xlabel('')

#     第二个图：skew统计分析中中一个属性
#     skewness 偏斜系数，对正太分布的度量
    i+=1
    plt.subplot(frows,fcols,i)
    _=stats.probplot(dat[col], plot=plt)#画图，偏析度
    plt.title('skew='+'{:.4f}'.format(stats.skew(dat[col])))
    plt.xlabel('')
    plt.ylabel('')

#     散点图
    i+=1
    plt.subplot(frows,fcols,i)
#     plt.plot(dat[var], dat['target'],'.',alpha=0.5)
    plt.scatter(dat[col],dat['target'],alpha=0.5)
    plt.title('corr='+'{:.2f}'.format(np.corrcoef(dat[col], dat['target'])[0][1]))

#     ！！！对数据进行了处理！！！
#   数据分布图distribution
    i+=1
    plt.subplot(frows,fcols,i)
    trans_var, lambda_var = stats.boxcox(dat[col].dropna()+1)
    trans_var = scale_minmax(trans_var)      
    sns.distplot(trans_var , fit=stats.norm);
    plt.title(col+' Tramsformed')
    plt.xlabel('')

#     偏斜度
    i+=1
    plt.subplot(frows,fcols,i)
    _=stats.probplot(trans_var, plot=plt)
    plt.title('skew='+'{:.4f}'.format(stats.skew(trans_var)))
    plt.xlabel('')
    plt.ylabel('')

#     散点图
    i+=1
    plt.subplot(frows,fcols,i)
    plt.plot(trans_var, dat['target'],'.',alpha=0.5)
    plt.title('corr='+'{:.2f}'.format(np.corrcoef(trans_var,dat['target'])[0][1]))
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56

# 将数据进行Box-Cox转换
# 统计建模中常用的数据变化
# 数据更加正态化，标准化
for col in data_all_norm.columns[:-2]:
    boxcox,maxlog = stats.boxcox(data_all_norm[col] + 1)
    data_all_norm[col] = scale_minmax(boxcox)
1
2
3
4
5
6

data_all_norm
1

4813 rows × 34 columns

过滤异常值

ridge = RidgeCV(alphas=[0.0001,0.001,0.01,0.1,0.2,0.5,1,2,3,4,5,10,20,30,50])

cond = data_all_norm['origin'] == 'train'

X_train = data_all_norm[cond].iloc[:,:-2]
# 真实值
y_train = data_all_norm[cond]['target']
# 算法拟合数据和目标值的时候，不可能100%拟合
ridge.fit(X_train,y_train)
# 预测，预测值肯定会和真实值有一定的偏差，偏差特别大，当成异常值
y_ = ridge.predict(X_train)

cond = abs(y_ - y_train) > y_train.std()
print(cond.sum())
# 画图
plt.figure(figsize=(12,6))
axes = plt.subplot(1,3,1)
axes.scatter(y_train,y_)
axes.scatter(y_train[cond],y_[cond],c = 'red',s = 20)

axes = plt.subplot(1,3,2)
axes.scatter(y_train,y_train - y_)
axes.scatter(y_train[cond],(y_train - y_)[cond],c = 'red')

axes = plt.subplot(1,3,3)
# _ = axes.hist(y_train,bins = 50)
(y_train - y_).plot.hist(bins = 50,ax = axes)
(y_train - y_).loc[cond].plot.hist(bins = 50,ax = axes,color = 'r')
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28

40





<matplotlib.axes._subplots.AxesSubplot at 0x2403c0836a0>
1
2
3
4
5
6
7

index = cond[cond].index

data_all_norm.drop(index,axis = 0,inplace=True)
1
2
3

cond = data_all_norm['origin'] == 'train'
X_train = data_all_norm[cond].iloc[:,:-2]
y_train = data_all_norm[cond]['target']

cond = data_all_norm['origin'] == 'test'
X_test = data_all_norm[cond].iloc[:,:-2]
1
2
3
4
5
6

使用不同算法进行计算，最后求取平均值！！

estimators = {}
estimators['forest'] = RandomForestRegressor(n_estimators=300)
estimators['gbdt'] = GradientBoostingRegressor(n_estimators=300)
estimators['ada'] = AdaBoostRegressor(n_estimators=300)
estimators['extreme'] = ExtraTreesRegressor(n_estimators=300)
estimators['svm_rbf'] = SVR(kernel='rbf')
estimators['light'] = LGBMRegressor(n_estimators=300)
estimators['xgb'] = XGBRegressor(n_estimators=300)
1
2
3
4
5
6
7
8

#将结果存入列表中，求取平均值作为最后答案
result = []
for key,model in estimators.items():
    model.fit(X_train,y_train)
    y_ = model.predict(X_test)
    result.append(y_)
y_ = np.mean(result,axis = 0)

pd.Series(y_).to_csv('./norm.txt',index = False)
1
2
3
4
5
6
7
8
9

[19:51:26] WARNING: C:/Jenkins/workspace/xgboost-win64_release_0.90/src/objective/regression_obj.cu:152: reg:linear is now deprecated in favor of reg:squarederror.
1