-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathregmodel.py
143 lines (108 loc) · 4.12 KB
/
regmodel.py
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
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
import statsmodels.api as sm
from statsmodels.tools.eval_measures import rmse
from sklearn.model_selection import train_test_split
from sklearn.linear_model import LinearRegression, Ridge, ElasticNet
from sklearn.model_selection import RandomizedSearchCV
from sklearn.metrics import mean_absolute_error
### Reading data
iDir = 'input/'
pDir = 'plots/'
fnam = 'data.csv'
df = pd.read_csv(iDir + fnam,sep=',',header=0)
display(df.head())
display(df.tail())
### Feature inspection
fig = sns.pairplot(df,kind='reg', diag_kind='kde',corner=True,hue='group',aspect=1)
fig.savefig(pDir + 'pairplot-all.png',dpi=150)
key = df.columns[1:-1]
def plboxp(df,key,fnam):
fig, ax = plt.subplots(1,len(key),figsize=(9,3))
for i,k in enumerate(key):
sns.boxplot(data=df,y=k,x='group',hue='group',fill=False,ax=ax[i],orient='v')
fig.tight_layout()
fig.savefig(pDir + fnam + '.png',dpi=150)
plboxp(df,key,'boxplot-all')
def outlim(ser):
iqr = np.diff(np.percentile(ser,[25,75]))[0]
med = np.median(ser)
return(med-1.5*iqr,med+1.5*iqr)
dfn = pd.DataFrame()
for g in ['group1','group2']:
kg = df.group == g
dfi = df[kg]
k = True
for ke in key:
lLim,uLim = outlim(dfi[ke])
k1 = (dfi[ke]<=uLim) & (dfi[ke]>=lLim)
k = k*k1
dfn = pd.concat([dfn,dfi[k]])
print(' Size of original dataset:',df.shape[0])
print('Size of dataset wo outliers:',dfn.shape[0])
plboxp(dfn,key,'boxplot-clean')
fig = sns.pairplot(dfn,kind='reg', diag_kind='kde',corner=True,hue='group',aspect=1)
fig.savefig(pDir + 'pairplot-clean.png',dpi=150)
## Building regression model
npow = {'group1':[3.0,1.0,-.5,2.0,2.0],
'group2':[2.0,1.0,1.0,1.0,2.0]}
SEED = 1311131
y = dfn.dependent
X = pd.DataFrame()
model_name = ['LinearRegression','Ridge','ElasticNet']
model = [LinearRegression(fit_intercept=True),
Ridge(fit_intercept=True),
ElasticNet(fit_intercept=True)]
p_dist = {'ElasticNet':{ 'alpha' : np.linspace(0.1,99.9,990), 'l1_ratio' : np.linspace(0.1,0.9,9),'tol' : [1e-1,1e-4]},
'Ridge':{'alpha' : np.linspace(0.1,.9,990)},
}
col = ['#1f77b4','#ff7f0e','#2ca02c']
mar = ['s','<','o']
fig,ax = plt.subplots(1,2,figsize=(9,4))
fiq,axq = plt.subplots(1,2,figsize=(9,4))
for i,g in enumerate(['group1','group2']):
for ii,n in enumerate(npow[g]):
X[key[ii]] = dfn[key[ii]]**n
k = dfn.group == g
X_train, X_test, y_train, y_test = train_test_split(X[k], y[k], test_size = 0.3 , random_state = SEED)
vpr = []
print('\n%8s metrics'%(g))
s = [120,70,30]
for ii,m in enumerate(model):
if model_name[ii] != 'LinearRegression':
rsh = RandomizedSearchCV(estimator=m, param_distributions=p_dist[model_name[ii]])
rsh.fit(X_train, y_train)
bp = rsh.best_params_
m.set_params(**bp)
m.fit(X_train,y_train)
y_pred = m.predict(X_test)
vpr.append(y_pred)
form = '%%%is'%int(1/2*(len(model_name[ii])+8)+16)
print('%12s'%model_name[ii])
print('%16s:%8.5f'%('R^2',m.score(X_test,y_test)))
print('%16s:%8.1f'%('RMSE',rmse(y_pred,y_test)))
#ax[i].scatter(y_test,m.predict(X_test),ec=col[ii],s=s[ii],fc='white',alpha=1,label=model_name[ii])
res = np.abs(y_test-m.predict(X_test))*10
###### Correlation Plots
ax[i].scatter(y_test,y_pred,ec=col[ii],s=s[ii],fc=col[ii]+'22',label=model_name[ii])
#### Q-Q Residuals Plots
fiq = sm.qqplot(res, marker = mar[ii], markeredgecolor=col[ii],
markerfacecolor=col[ii]+'22', ax=axq[i], fit=True,
line='45', label=model_name[ii])
axq[i].set_title(g)
axq[i].legend()
axq[i].set_aspect(1)
print()
ax[0].legend()
xL = [min(y_test),max(y_test)]
ax[i].plot(xL,xL,c='tab:red')
ax[i].set_aspect(1)
ax[i].set_title(g)
dfr = pd.DataFrame(data=vpr).T
dfr.columns = model_name
dfr = pd.concat([y_test.reset_index(drop=True),dfr],axis=1)
display(dfr.head())
display(dfr.describe())
plt.show()