|
1
|
|
|
''' |
|
2
|
|
|
Utilities for data cleaning. |
|
3
|
|
|
|
|
4
|
|
|
:author: Andreas Kanz |
|
5
|
|
|
|
|
6
|
|
|
''' |
|
7
|
|
|
|
|
8
|
|
|
# Imports |
|
9
|
|
|
import pandas as pd |
|
10
|
|
|
|
|
11
|
|
|
from .describe import _memory_usage |
|
12
|
|
|
from .describe import _missing_vals |
|
13
|
|
|
|
|
14
|
|
|
|
|
15
|
|
|
def _validate_input_0_1(value, desc): |
|
16
|
|
|
if value < 0 or value > 1: |
|
17
|
|
|
raise ValueError(f'Input value for {desc} is {value} but should be a float in the range 0 <= {desc} <=1.') |
|
18
|
|
|
|
|
19
|
|
|
|
|
20
|
|
|
def convert_datatypes(data, category=True, cat_threshold=0.05, cat_exclude=[]): |
|
21
|
|
|
''' |
|
22
|
|
|
Converts columns to best possible dtypes using dtypes supporting pd.NA. |
|
23
|
|
|
|
|
24
|
|
|
Parameters |
|
25
|
|
|
---------- |
|
26
|
|
|
data: 2D dataset that can be coerced into Pandas DataFrame. If a Pandas DataFrame is provided, the index/column \ |
|
27
|
|
|
information is used to label the plots. |
|
28
|
|
|
|
|
29
|
|
|
category: bool, default True |
|
30
|
|
|
Change dtypes of columns with dtype "object" to "category". Set threshold using cat_threshold or exclude \ |
|
31
|
|
|
columns using cat_exclude. |
|
32
|
|
|
|
|
33
|
|
|
cat_threshold: float, default 0.05 |
|
34
|
|
|
Ratio of unique values below which categories are inferred and column dtype is changed to categorical. |
|
35
|
|
|
|
|
36
|
|
|
cat_exclude: default [] (empty list) |
|
37
|
|
|
List of columns to exclude from categorical conversion. |
|
38
|
|
|
|
|
39
|
|
|
Returns |
|
40
|
|
|
------- |
|
41
|
|
|
Pandas DataFrame. |
|
42
|
|
|
|
|
43
|
|
|
''' |
|
44
|
|
|
|
|
45
|
|
|
_validate_input_0_1(cat_threshold, 'cat_threshold') |
|
46
|
|
|
|
|
47
|
|
|
data = pd.DataFrame(data).copy() |
|
48
|
|
|
for col in data.columns: |
|
49
|
|
|
unique_vals_ratio = data[col].nunique(dropna=False) / data.shape[0] |
|
50
|
|
|
if (category and |
|
51
|
|
|
unique_vals_ratio < cat_threshold and |
|
52
|
|
|
col not in cat_exclude and |
|
53
|
|
|
data[col].dtype == 'object'): |
|
54
|
|
|
data[col] = data[col].astype('category') |
|
55
|
|
|
data[col] = data[col].convert_dtypes() |
|
56
|
|
|
|
|
57
|
|
|
return data |
|
58
|
|
|
|
|
59
|
|
|
|
|
60
|
|
|
def drop_missing(data, drop_threshold_cols=1, drop_threshold_rows=1): |
|
61
|
|
|
''' |
|
62
|
|
|
Drops completely empty columns and rows by default and optionally provides flexibility to loosen restrictions to \ |
|
63
|
|
|
drop additional columns and rows based on the fraction of NA-values. |
|
64
|
|
|
|
|
65
|
|
|
Parameters |
|
66
|
|
|
---------- |
|
67
|
|
|
data: 2D dataset that can be coerced into Pandas DataFrame. If a Pandas DataFrame is provided, the index/column \ |
|
68
|
|
|
information is used to label the plots. |
|
69
|
|
|
|
|
70
|
|
|
drop_threshold_cols: float, default 1 |
|
71
|
|
|
Drop columns with NA-ratio above the specified threshold. |
|
72
|
|
|
|
|
73
|
|
|
drop_threshold_rows: float, default 1 |
|
74
|
|
|
Drop rows with NA-ratio above the specified threshold. |
|
75
|
|
|
|
|
76
|
|
|
Returns |
|
77
|
|
|
------- |
|
78
|
|
|
Pandas DataFrame. |
|
79
|
|
|
|
|
80
|
|
|
Notes |
|
81
|
|
|
----- |
|
82
|
|
|
Columns are dropped first. Rows are dropped based on the remaining data. |
|
83
|
|
|
|
|
84
|
|
|
''' |
|
85
|
|
|
|
|
86
|
|
|
_validate_input_0_1(drop_threshold_cols, 'drop_threshold_cols') |
|
87
|
|
|
_validate_input_0_1(drop_threshold_rows, 'drop_threshold_rows') |
|
88
|
|
|
|
|
89
|
|
|
data = pd.DataFrame(data) |
|
90
|
|
|
data = data.dropna(axis=0, how='all') |
|
91
|
|
|
data = data.dropna(axis=1, how='all') |
|
92
|
|
|
data = data.drop(columns=data.loc[:, _missing_vals(data)['mv_cols_ratio'] > drop_threshold_cols].columns) |
|
93
|
|
|
data_cleaned = data.drop(index=data.loc[_missing_vals(data)['mv_rows_ratio'] > drop_threshold_rows, :].index) |
|
94
|
|
|
|
|
95
|
|
|
return data_cleaned |
|
96
|
|
|
|
|
97
|
|
|
|
|
98
|
|
|
def data_cleaning(data, drop_threshold_cols=0.95, drop_threshold_rows=0.95, category=True, |
|
99
|
|
|
cat_threshold=0.03, cat_exclude=[], show='changes'): |
|
100
|
|
|
''' |
|
101
|
|
|
Perform initial data cleaning tasks on a dataset, such as dropping empty rows and columns and optimizing the \ |
|
102
|
|
|
datatypes. |
|
103
|
|
|
|
|
104
|
|
|
Parameters |
|
105
|
|
|
---------- |
|
106
|
|
|
data: 2D dataset that can be coerced into Pandas DataFrame. If a Pandas DataFrame is provided, the index/column \ |
|
107
|
|
|
information is used to label the plots. |
|
108
|
|
|
|
|
109
|
|
|
drop_threshold_cols: float, default 0.95 |
|
110
|
|
|
Drop columns with NA-ratio above the specified threshold. |
|
111
|
|
|
|
|
112
|
|
|
drop_threshold_rows: float, default 0.95 |
|
113
|
|
|
Drop rows with NA-ratio above the specified threshold. |
|
114
|
|
|
|
|
115
|
|
|
category: bool, default True |
|
116
|
|
|
Change dtypes of columns to "category". Set threshold using cat_threshold. |
|
117
|
|
|
|
|
118
|
|
|
cat_threshold: float, default 0.03 |
|
119
|
|
|
Ratio of unique values below which categories are inferred and column dtype is changed to categorical. |
|
120
|
|
|
|
|
121
|
|
|
cat_exclude: default [] (empty list) |
|
122
|
|
|
List of columns to exclude from categorical conversion. |
|
123
|
|
|
|
|
124
|
|
|
show: {'all', 'changes', None} default 'all' |
|
125
|
|
|
Specify verbosity of the output. |
|
126
|
|
|
* 'all': Print information about the data before and after cleaning as well as information about changes. |
|
127
|
|
|
* 'changes': Print out differences in the data before and after cleaning. |
|
128
|
|
|
* None: no information about the data is printed. |
|
129
|
|
|
|
|
130
|
|
|
Returns |
|
131
|
|
|
------- |
|
132
|
|
|
Pandas DataFrame. |
|
133
|
|
|
|
|
134
|
|
|
See Also |
|
135
|
|
|
-------- |
|
136
|
|
|
convert_datatypes: Converts columns to best possible dtypes. |
|
137
|
|
|
drop_missing : Flexibly drops columns and rows. |
|
138
|
|
|
_memory_usage: Gives the total memory usage in kilobytes. |
|
139
|
|
|
_missing_vals: Metrics about missing values in the dataset. |
|
140
|
|
|
|
|
141
|
|
|
|
|
142
|
|
|
Notes |
|
143
|
|
|
----- |
|
144
|
|
|
The category dtype is not grouped in the summary, unless it contains exactly the same categories. |
|
145
|
|
|
|
|
146
|
|
|
''' |
|
147
|
|
|
|
|
148
|
|
|
data = pd.DataFrame(data) |
|
149
|
|
|
data_cleaned = drop_missing(data, drop_threshold_cols, drop_threshold_rows) |
|
150
|
|
|
data_cleaned = convert_datatypes(data_cleaned, category=category, cat_threshold=cat_threshold, |
|
151
|
|
|
cat_exclude=cat_exclude) |
|
152
|
|
|
|
|
153
|
|
|
if show in ['changes', 'all']: |
|
154
|
|
|
if show == 'all': |
|
155
|
|
|
print('Before data cleaning:\n') |
|
156
|
|
|
print(f'dtypes:\n{data.dtypes.value_counts()}') |
|
157
|
|
|
print(f'\nNumber of rows: {data.shape[0]}') |
|
158
|
|
|
print(f'Number of cols: {data.shape[1]}') |
|
159
|
|
|
print(f"Missing values: {_missing_vals(data)['mv_total']}") |
|
160
|
|
|
print(f'Memory usage: {_memory_usage(data)} KB') |
|
161
|
|
|
print('_______________________________________________________\n') |
|
162
|
|
|
print('After data cleaning:\n') |
|
163
|
|
|
print(f'dtypes:\n{data_cleaned.dtypes.value_counts()}') |
|
164
|
|
|
print(f'\nNumber of rows: {data_cleaned.shape[0]}') |
|
165
|
|
|
print(f'Number of cols: {data_cleaned.shape[1]}') |
|
166
|
|
|
print(f"Missing values: {_missing_vals(data_cleaned)['mv_total']}") |
|
167
|
|
|
print(f'Memory usage: {_memory_usage(data_cleaned)} KB') |
|
168
|
|
|
print('_______________________________________________________\n') |
|
169
|
|
|
|
|
170
|
|
|
print( |
|
171
|
|
|
f"Shape of cleaned data: {data_cleaned.shape} - Remaining NAs: {_missing_vals(data_cleaned)['mv_total']}") |
|
172
|
|
|
print(f'\nChanges:') |
|
173
|
|
|
print(f'Dropped rows: {data.shape[0]-data_cleaned.shape[0]}') |
|
174
|
|
|
print(f'Dropped columns: {data.shape[1]-data_cleaned.shape[1]}') |
|
175
|
|
|
print(f"Dropped missing values: {_missing_vals(data)['mv_total']-_missing_vals(data_cleaned)['mv_total']}") |
|
176
|
|
|
mem_change = _memory_usage(data)-_memory_usage(data_cleaned) |
|
177
|
|
|
print(f'Reduced memory by: {round(mem_change,2)} KB (-{round(100*mem_change/_memory_usage(data),1)}%)') |
|
178
|
|
|
|
|
179
|
|
|
return data_cleaned |
|
180
|
|
|
|
akanz1 /
klib
