|
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
|
|
|
SelectFromModel, |
|
18
|
|
|
SelectPercentile, |
|
19
|
|
|
VarianceThreshold) |
|
20
|
|
|
from sklearn.linear_model import LassoCV |
|
21
|
|
|
from sklearn.model_selection import train_test_split |
|
22
|
|
|
from sklearn.pipeline import make_pipeline |
|
23
|
|
|
from sklearn.preprocessing import (OneHotEncoder, |
|
24
|
|
|
RobustScaler, |
|
25
|
|
|
MaxAbsScaler) |
|
26
|
|
|
|
|
27
|
|
|
from .utils import (_validate_input_int, |
|
28
|
|
|
_validate_input_range, |
|
29
|
|
|
_validate_input_sum) |
|
30
|
|
|
|
|
31
|
|
|
|
|
32
|
|
|
__all__ = ['feature_selection_pipe', |
|
33
|
|
|
'num_pipe', |
|
34
|
|
|
'cat_pipe', |
|
35
|
|
|
'train_dev_test_split'] |
|
36
|
|
|
|
|
37
|
|
|
|
|
38
|
|
|
class ColumnSelector(BaseEstimator, TransformerMixin): |
|
39
|
|
|
''' |
|
40
|
|
|
Determines and selects numerical and categorical columns from a dataset based on their supposed dtype. Unlike \ |
|
41
|
|
|
sklearn's make_column_selector() missing values are temporarily filled in to allow convert_dtypes() to determine \ |
|
42
|
|
|
the dtype of a column. |
|
43
|
|
|
|
|
44
|
|
|
Parameter: |
|
45
|
|
|
--------- |
|
46
|
|
|
num: default, True |
|
47
|
|
|
Select only numerica Columns. If num = False, only categorical columns are selected. |
|
48
|
|
|
|
|
49
|
|
|
Returns: |
|
50
|
|
|
------- |
|
51
|
|
|
Dataset containing only numerical or categorical data. |
|
52
|
|
|
''' |
|
53
|
|
|
|
|
54
|
|
|
def __init__(self, num=True): |
|
55
|
|
|
self.num = num |
|
56
|
|
|
|
|
57
|
|
|
def fit(self, X, y=None): |
|
58
|
|
|
return self |
|
59
|
|
|
|
|
60
|
|
|
def transform(self, X, y=None): |
|
61
|
|
|
temp = X.fillna(X.mode().iloc[0]).convert_dtypes() |
|
62
|
|
|
|
|
63
|
|
|
if self.num: |
|
64
|
|
|
return X[temp.select_dtypes(include=['number']).columns.tolist()] |
|
65
|
|
|
else: |
|
66
|
|
|
return X[temp.select_dtypes(exclude=['number']).columns.tolist()] |
|
67
|
|
|
|
|
68
|
|
|
|
|
69
|
|
|
class PipeInfo(BaseEstimator, TransformerMixin): |
|
70
|
|
|
''' |
|
71
|
|
|
Prints intermediary information about the dataset from within a pipeline. Include at any point in a Pipeline to |
|
72
|
|
|
print out the shape of the dataset at this point and to receive an indication of the progress within the pipeline. |
|
73
|
|
|
Set to 'None' to avoid printing the shape of the dataset. This parameter can also be set as a hyperparameter, \ |
|
74
|
|
|
e.g. 'pipeline__pipeinfo-1': [None] or 'pipeline__pipeinfo-1__name': ['my_custom_name']. |
|
75
|
|
|
|
|
76
|
|
|
Parameter: |
|
77
|
|
|
--------- |
|
78
|
|
|
name: string, default None |
|
79
|
|
|
Provide a name for the current step. |
|
80
|
|
|
|
|
81
|
|
|
Returns: |
|
82
|
|
|
------- |
|
83
|
|
|
Data: Data is being passed through. |
|
84
|
|
|
''' |
|
85
|
|
|
|
|
86
|
|
|
def __init__(self, name=None): |
|
87
|
|
|
self.name = name |
|
88
|
|
|
|
|
89
|
|
|
def fit(self, X, y=None): |
|
90
|
|
|
return self |
|
91
|
|
|
|
|
92
|
|
|
def transform(self, X, y=None): |
|
93
|
|
|
print(f'Step: {self.name} --- Shape: {X.shape}') |
|
94
|
|
|
return X |
|
95
|
|
|
|
|
96
|
|
|
|
|
97
|
|
|
def cat_pipe(imputer=SimpleImputer(strategy='most_frequent'), |
|
98
|
|
|
encoder=OneHotEncoder(handle_unknown='ignore'), |
|
99
|
|
|
scaler=MaxAbsScaler(), |
|
100
|
|
|
encoder_info=PipeInfo(name='after encoding')): |
|
101
|
|
|
''' |
|
102
|
|
|
Standard preprocessing operations on categorical data. |
|
103
|
|
|
|
|
104
|
|
|
Parameters: |
|
105
|
|
|
---------- |
|
106
|
|
|
imputer: default, SimpleImputer(strategy='most_frequent') |
|
107
|
|
|
|
|
108
|
|
|
encoder: default, OneHotEncoder(handle_unknown='ignore') |
|
109
|
|
|
Encode categorical features as a one-hot numeric array. |
|
110
|
|
|
|
|
111
|
|
|
scaler: default, MaxAbsScaler() |
|
112
|
|
|
Scale each feature by its maximum absolute value. MaxAbsScaler() does not shift/center the data, and thus does \ |
|
113
|
|
|
not destroy any sparsity. It is recommended to check for outliers before applying MaxAbsScaler(). |
|
114
|
|
|
|
|
115
|
|
|
encoder_info: |
|
116
|
|
|
Prints the shape of the dataset at the end of 'cat_pipe'. Set to 'None' to avoid printing the shape of \ |
|
117
|
|
|
dataset. This parameter can also be set as a hyperparameter, e.g. 'pipeline__pipeinfo-1': [None] or \ |
|
118
|
|
|
'pipeline__pipeinfo-1__name': ['my_custom_name']. |
|
119
|
|
|
|
|
120
|
|
|
Returns: |
|
121
|
|
|
------- |
|
122
|
|
|
Pipeline |
|
123
|
|
|
''' |
|
124
|
|
|
|
|
125
|
|
|
cat_pipe = make_pipeline(ColumnSelector(num=False), |
|
126
|
|
|
imputer, |
|
127
|
|
|
encoder, encoder_info, |
|
128
|
|
|
scaler) |
|
129
|
|
|
return cat_pipe |
|
130
|
|
|
|
|
131
|
|
|
|
|
132
|
|
|
def feature_selection_pipe( |
|
133
|
|
|
var_thresh=VarianceThreshold(threshold=0.1), |
|
134
|
|
|
select_from_model=SelectFromModel(LassoCV(cv=4, random_state=408), threshold="0.1*median"), |
|
135
|
|
|
select_percentile=SelectPercentile(f_classif, percentile=95), |
|
136
|
|
|
var_thresh_info=PipeInfo(name='after var_thresh'), |
|
137
|
|
|
select_from_model_info=PipeInfo(name='after select_from_model'), |
|
138
|
|
|
select_percentile_info=PipeInfo(name='after select_percentile')): |
|
139
|
|
|
''' |
|
140
|
|
|
Preprocessing operations for feature selection. |
|
141
|
|
|
|
|
142
|
|
|
Parameters: |
|
143
|
|
|
---------- |
|
144
|
|
|
var_thresh: default, VarianceThreshold(threshold=0.1) |
|
145
|
|
|
Specify a threshold to drop low variance features. |
|
146
|
|
|
|
|
147
|
|
|
select_from_model: default, SelectFromModel(LassoCV(cv=4, random_state=408), threshold="0.1*median") |
|
148
|
|
|
Specify an estimator which is used for selecting features based on importance weights. |
|
149
|
|
|
|
|
150
|
|
|
select_percentile: default, SelectPercentile(f_classif, percentile=95) |
|
151
|
|
|
Specify a score-function and a percentile value of features to keep. |
|
152
|
|
|
|
|
153
|
|
|
var_thresh_info, select_from_model_info, select_percentile_info |
|
154
|
|
|
Prints the shape of the dataset after applying the respective function. Set to 'None' to avoid printing the \ |
|
155
|
|
|
shape of dataset. This parameter can also be set as a hyperparameter, e.g. 'pipeline__pipeinfo-1': [None] \ |
|
156
|
|
|
or 'pipeline__pipeinfo-1__name': ['my_custom_name']. |
|
157
|
|
|
|
|
158
|
|
|
Returns: |
|
159
|
|
|
------- |
|
160
|
|
|
Pipeline |
|
161
|
|
|
''' |
|
162
|
|
|
|
|
163
|
|
|
feature_selection_pipe = make_pipeline(var_thresh, var_thresh_info, |
|
164
|
|
|
select_from_model, select_from_model_info, |
|
165
|
|
|
select_percentile, select_percentile_info) |
|
166
|
|
|
return feature_selection_pipe |
|
167
|
|
|
|
|
168
|
|
|
|
|
169
|
|
|
def num_pipe(imputer=IterativeImputer(estimator=ExtraTreesRegressor( |
|
170
|
|
|
n_estimators=25, n_jobs=4, random_state=408), random_state=408), |
|
171
|
|
|
scaler=RobustScaler()): |
|
172
|
|
|
''' |
|
173
|
|
|
Standard preprocessing operations on numerical data. |
|
174
|
|
|
|
|
175
|
|
|
Parameters: |
|
176
|
|
|
---------- |
|
177
|
|
|
imputer: default, IterativeImputer(estimator=ExtraTreesRegressor(n_estimators=25, n_jobs=4, random_state=408), \ |
|
178
|
|
|
random_state=408) |
|
179
|
|
|
|
|
180
|
|
|
scaler: default, RobustScaler() |
|
181
|
|
|
|
|
182
|
|
|
Returns: |
|
183
|
|
|
------- |
|
184
|
|
|
Pipeline |
|
185
|
|
|
''' |
|
186
|
|
|
|
|
187
|
|
|
num_pipe = make_pipeline(ColumnSelector(), |
|
188
|
|
|
imputer, |
|
189
|
|
|
scaler) |
|
190
|
|
|
return num_pipe |
|
191
|
|
|
|
|
192
|
|
|
|
|
193
|
|
|
def train_dev_test_split(data, target, dev_size=0.1, test_size=0.1, stratify=None, random_state=408): |
|
194
|
|
|
''' |
|
195
|
|
|
Split a dataset and a label column into train, dev and test sets. |
|
196
|
|
|
|
|
197
|
|
|
Parameters: |
|
198
|
|
|
---------- |
|
199
|
|
|
|
|
200
|
|
|
data: 2D dataset that can be coerced into Pandas DataFrame. If a Pandas DataFrame is provided, the index/column \ |
|
201
|
|
|
information is used to label the plots. |
|
202
|
|
|
|
|
203
|
|
|
target: string, list, np.array or pd.Series, default None |
|
204
|
|
|
Specify target for correlation. E.g. label column to generate only the correlations between each feature \ |
|
205
|
|
|
and the label. |
|
206
|
|
|
|
|
207
|
|
|
dev_size: float, default 0.1 |
|
208
|
|
|
If float, should be between 0.0 and 1.0 and represent the proportion of the dataset to include in the dev \ |
|
209
|
|
|
split. |
|
210
|
|
|
|
|
211
|
|
|
test_size: float, default 0.1 |
|
212
|
|
|
If float, should be between 0.0 and 1.0 and represent the proportion of the dataset to include in the test \ |
|
213
|
|
|
split. |
|
214
|
|
|
|
|
215
|
|
|
stratify: target column, default None |
|
216
|
|
|
If not None, data is split in a stratified fashion, using the input as the class labels. |
|
217
|
|
|
|
|
218
|
|
|
random_state: integer, default 408 |
|
219
|
|
|
Random_state is the seed used by the random number generator. |
|
220
|
|
|
|
|
221
|
|
|
Returns |
|
222
|
|
|
------- |
|
223
|
|
|
tuple: Tuple containing train-dev-test split of inputs. |
|
224
|
|
|
''' |
|
225
|
|
|
|
|
226
|
|
|
# Validate Inputs |
|
227
|
|
|
_validate_input_range(dev_size, 'dev_size', 0, 1) |
|
228
|
|
|
_validate_input_range(test_size, 'test_size', 0, 1) |
|
229
|
|
|
_validate_input_int(random_state, 'random_state') |
|
230
|
|
|
_validate_input_sum(1, 'Dev and test', dev_size, test_size) |
|
231
|
|
|
|
|
232
|
|
|
target_data = [] |
|
233
|
|
|
if isinstance(target, str): |
|
234
|
|
|
target_data = data[target] |
|
235
|
|
|
data = data.drop(target, axis=1) |
|
236
|
|
|
|
|
237
|
|
|
elif isinstance(target, (list, pd.Series, np.ndarray)): |
|
238
|
|
|
target_data = pd.Series(target) |
|
239
|
|
|
|
|
240
|
|
|
X_train, X_dev_test, y_train, y_dev_test = train_test_split(data, target_data, |
|
241
|
|
|
test_size=dev_size+test_size, |
|
242
|
|
|
random_state=random_state, |
|
243
|
|
|
stratify=stratify) |
|
244
|
|
|
|
|
245
|
|
|
if (dev_size == 0) or (test_size == 0): |
|
246
|
|
|
return X_train, X_dev_test, y_train, y_dev_test |
|
247
|
|
|
|
|
248
|
|
|
else: |
|
249
|
|
|
X_dev, X_test, y_dev, y_test = train_test_split(X_dev_test, y_dev_test, |
|
250
|
|
|
test_size=test_size/(dev_size+test_size), |
|
251
|
|
|
random_state=random_state, |
|
252
|
|
|
stratify=y_dev_test) |
|
253
|
|
|
return X_train, X_dev, X_test, y_train, y_dev, y_test |
|
254
|
|
|
|
akanz1 /
klib
