|
1
|
|
|
''' |
|
2
|
|
|
Functions for data preprocessing. |
|
3
|
|
|
|
|
4
|
|
|
:author: Andreas Kanz |
|
5
|
|
|
|
|
6
|
|
|
''' |
|
7
|
|
|
|
|
8
|
|
|
# Imports |
|
9
|
|
|
import numpy as np |
|
10
|
|
|
import pandas as pd |
|
11
|
|
|
|
|
12
|
|
|
from sklearn.base import BaseEstimator, TransformerMixin |
|
13
|
|
|
from sklearn.ensemble import ExtraTreesRegressor |
|
14
|
|
|
from sklearn.experimental import enable_iterative_imputer # noqa |
|
15
|
|
|
from sklearn.impute import (SimpleImputer, IterativeImputer) |
|
16
|
|
|
from sklearn.feature_selection import (f_classif, |
|
17
|
|
|
SelectPercentile, |
|
18
|
|
|
VarianceThreshold) |
|
19
|
|
|
from sklearn.model_selection import train_test_split |
|
20
|
|
|
from sklearn.pipeline import make_pipeline |
|
21
|
|
|
from sklearn.preprocessing import OneHotEncoder, RobustScaler |
|
22
|
|
|
|
|
23
|
|
|
from .utils import (_validate_input_int, |
|
24
|
|
|
_validate_input_range, |
|
25
|
|
|
_validate_input_sum) |
|
26
|
|
|
|
|
27
|
|
|
|
|
28
|
|
|
__all__ = ['feature_selection_pipe', |
|
29
|
|
|
'num_pipe', |
|
30
|
|
|
'cat_pipe', |
|
31
|
|
|
'train_dev_test_split'] |
|
32
|
|
|
|
|
33
|
|
|
|
|
34
|
|
|
class ColumnSelector(BaseEstimator, TransformerMixin): |
|
35
|
|
|
'''''' |
|
36
|
|
|
|
|
37
|
|
|
def __init__(self, num=True): |
|
38
|
|
|
self.num = num |
|
39
|
|
|
|
|
40
|
|
|
def fit(self, X, y=None): |
|
41
|
|
|
return self |
|
42
|
|
|
|
|
43
|
|
|
def transform(self, X, y=None): |
|
44
|
|
|
temp = X.fillna(X.mode().iloc[0]).convert_dtypes() |
|
45
|
|
|
|
|
46
|
|
|
if self.num: |
|
47
|
|
|
return X[temp.select_dtypes(include=['number']).columns.tolist()] |
|
48
|
|
|
else: |
|
49
|
|
|
return X[temp.select_dtypes(exclude=['number']).columns.tolist()] |
|
50
|
|
|
|
|
51
|
|
|
|
|
52
|
|
|
def feature_selection_pipe( |
|
53
|
|
|
var_thresh=VarianceThreshold(threshold=0.1), |
|
54
|
|
|
select_percentile=SelectPercentile(f_classif, percentile=95)): |
|
55
|
|
|
'''Preprocessing operations for feature selection.''' |
|
56
|
|
|
|
|
57
|
|
|
feature_selection_pipe = make_pipeline(var_thresh, |
|
58
|
|
|
select_percentile) |
|
59
|
|
|
return feature_selection_pipe |
|
60
|
|
|
|
|
61
|
|
|
|
|
62
|
|
|
def num_pipe(imputer=IterativeImputer(estimator=ExtraTreesRegressor( |
|
63
|
|
|
n_estimators=25, n_jobs=4, random_state=408), random_state=408), |
|
64
|
|
|
scaler=RobustScaler()): |
|
65
|
|
|
'''Standard preprocessing operations on numerical data.''' |
|
66
|
|
|
|
|
67
|
|
|
num_pipe = make_pipeline(ColumnSelector(), |
|
68
|
|
|
imputer, |
|
69
|
|
|
scaler) |
|
70
|
|
|
return num_pipe |
|
71
|
|
|
|
|
72
|
|
|
|
|
73
|
|
|
def cat_pipe(imputer=SimpleImputer(strategy='most_frequent'), |
|
74
|
|
|
scaler=OneHotEncoder(handle_unknown='ignore')): |
|
75
|
|
|
'''Set of standard preprocessing operations on categorical data.''' |
|
76
|
|
|
|
|
77
|
|
|
cat_pipe = make_pipeline(ColumnSelector(num=False), |
|
78
|
|
|
imputer, |
|
79
|
|
|
scaler) |
|
80
|
|
|
return cat_pipe |
|
81
|
|
|
|
|
82
|
|
|
|
|
83
|
|
|
def train_dev_test_split(data, target, dev_size=0.1, test_size=0.1, stratify=None, random_state=408): |
|
84
|
|
|
''' |
|
85
|
|
|
Split a dataset and a label column into train, dev and test sets. |
|
86
|
|
|
|
|
87
|
|
|
Parameters: |
|
88
|
|
|
---------- |
|
89
|
|
|
|
|
90
|
|
|
data: 2D dataset that can be coerced into Pandas DataFrame. If a Pandas DataFrame is provided, the index/column \ |
|
91
|
|
|
information is used to label the plots. |
|
92
|
|
|
|
|
93
|
|
|
target: string, list, np.array or pd.Series, default None |
|
94
|
|
|
Specify target for correlation. E.g. label column to generate only the correlations between each feature \ |
|
95
|
|
|
and the label. |
|
96
|
|
|
|
|
97
|
|
|
dev_size: float, default 0.1 |
|
98
|
|
|
If float, should be between 0.0 and 1.0 and represent the proportion of the dataset to include in the dev \ |
|
99
|
|
|
split. |
|
100
|
|
|
|
|
101
|
|
|
test_size: float, default 0.1 |
|
102
|
|
|
If float, should be between 0.0 and 1.0 and represent the proportion of the dataset to include in the test \ |
|
103
|
|
|
split. |
|
104
|
|
|
|
|
105
|
|
|
stratify: target column, default None |
|
106
|
|
|
If not None, data is split in a stratified fashion, using the input as the class labels. |
|
107
|
|
|
|
|
108
|
|
|
random_state: integer, default 408 |
|
109
|
|
|
Random_state is the seed used by the random number generator. |
|
110
|
|
|
|
|
111
|
|
|
Returns |
|
112
|
|
|
------- |
|
113
|
|
|
tuple: Tuple containing train-dev-test split of inputs. |
|
114
|
|
|
''' |
|
115
|
|
|
|
|
116
|
|
|
# Validate Inputs |
|
117
|
|
|
_validate_input_range(dev_size, 'dev_size', 0, 1) |
|
118
|
|
|
_validate_input_range(test_size, 'test_size', 0, 1) |
|
119
|
|
|
_validate_input_int(random_state, 'random_state') |
|
120
|
|
|
_validate_input_sum(1, 'Sum of args', dev_size, test_size) |
|
121
|
|
|
|
|
122
|
|
|
target_data = [] |
|
123
|
|
|
if isinstance(target, str): |
|
124
|
|
|
target_data = data[target] |
|
125
|
|
|
data = data.drop(target, axis=1) |
|
126
|
|
|
|
|
127
|
|
|
elif isinstance(target, (list, pd.Series, np.ndarray)): |
|
128
|
|
|
target_data = pd.Series(target) |
|
129
|
|
|
target = target.name |
|
130
|
|
|
|
|
131
|
|
|
X_train, X_dev_test, y_train, y_dev_test = train_test_split(data, target_data, |
|
132
|
|
|
test_size=dev_size+test_size, |
|
133
|
|
|
random_state=random_state, |
|
134
|
|
|
stratify=stratify) |
|
135
|
|
|
|
|
136
|
|
|
if (dev_size == 0) or (test_size == 0): |
|
137
|
|
|
return X_train, X_dev_test, y_train, y_dev_test |
|
138
|
|
|
|
|
139
|
|
|
else: |
|
140
|
|
|
X_dev, X_test, y_dev, y_test = train_test_split(X_dev_test, y_dev_test, |
|
141
|
|
|
test_size=test_size/(dev_size+test_size), |
|
142
|
|
|
random_state=random_state, |
|
143
|
|
|
stratify=y_dev_test) |
|
144
|
|
|
return X_train, X_dev, X_test, y_train, y_dev, y_test |
|
145
|
|
|
|
akanz1 /
klib
