-
Notifications
You must be signed in to change notification settings - Fork 1
/
data_module.py
499 lines (429 loc) · 18.4 KB
/
data_module.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
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
"""`DataModule` for training parametric models, generating and benchmarking CF explanations."""
# AUTOGENERATED! DO NOT EDIT! File to edit: ../nbs/01_data.ipynb.
# %% ../nbs/01_data.ipynb 3
from __future__ import annotations
from .utils import load_json, validate_configs, get_config, save_pytree, load_pytree, get_config
from .base import *
from .data_utils import *
from .import_essentials import *
import jax
from jax import numpy as jnp, random as jrand, lax, Array
import pandas as pd
import numpy as np
from pathlib import Path
import json, os, shutil
from urllib.request import urlretrieve
from pydantic.fields import ModelField, Field
from typing import List, Dict, Union, Optional, Tuple, Callable, Any, Iterable
import warnings
from pandas.testing import assert_frame_equal
# %% auto 0
__all__ = ['BaseDataModule', 'DataModuleConfig', 'features2config', 'features2pandas', 'dataframe2features', 'dataframe2labels',
'DataModule', 'download_data_module_files', 'load_data']
# %% ../nbs/01_data.ipynb 6
class BaseDataModule(BaseModule):
"""DataModule Interface"""
def _prepare(self, *args, **kwargs):
"""Prepare data for training"""
raise NotImplementedError
def apply_constraints(self, x: Array, cf: Array, hard: bool = False, **kwargs) -> Array:
raise NotImplementedError
def compute_reg_loss(self, x: Array, cf: Array, hard: bool = False, **kwargs) -> float:
raise NotImplementedError
# %% ../nbs/01_data.ipynb 7
class DataModuleInfoMixin:
"""This base class exposes some attributes of DataModule
at the base level for easy access.
"""
@property
def data(self) -> pd.DataFrame:
return self._data
@property
def xs(self) -> Array:
return self._features.transformed_data
@property
def ys(self) -> Array:
return self._label.transformed_data
@property
def features(self) -> FeaturesList:
return self._features
@property
def label(self) -> FeaturesList:
return self._label
@property
def dataset(self) -> Tuple[Array, Array]:
return (self.xs, self.ys)
@property
def train_indices(self) -> List[int]:
return self.config.train_indices
@property
def test_indices(self) -> List[int]:
return self.config.test_indices
# %% ../nbs/01_data.ipynb 10
class DataModuleConfig(BaseConfig):
"""Configurator of `DataModule`."""
data_dir: str = Field(None, description="The directory of dataset.")
data_name: str = Field(None, description="The name of `DataModule`.")
continous_cols: List[str] = Field([], description="Continuous features/columns in the data.")
discret_cols: List[str] = Field([], description="Categorical features/columns in the data.")
imutable_cols: List[str] = Field([], description="Immutable features/columns in the data.")
continuous_transformation: Optional[str] = Field('minmax', description="Transformation for continuous features. `None` indicates unknown.")
discret_transformation: Optional[str] = Field('ohe', description="Transformation for categorical features. `None` indicates unknown.")
sample_frac: Optional[float] = Field(
None, description="Sample fraction of the data. Default to use the entire data.", ge=0., le=1.0
)
train_indices: List[int] = Field([], description="Indices of training data.")
test_indices: List[int] = Field([], description="Indices of testing data.")
def shuffle(self, data: Array, test_size: float, seed: int = None):
"""Shuffle data with a seed."""
if seed is None:
seed = get_config().global_seed
key = jrand.PRNGKey(seed)
total_length = data.shape[0]
train_length = int((1 - test_size) * total_length)
if len(self.train_indices) == 0:
self.train_indices = jrand.permutation(key, total_length)[:train_length].tolist()
if len(self.test_indices) == 0:
self.test_indices = jrand.permutation(key, total_length)[train_length:].tolist()
# %% ../nbs/01_data.ipynb 13
def features2config(
features: FeaturesList, # FeaturesList to be converted
name: str, # Name of the data used for `DataModuleConfig`
return_dict: bool = False # Whether to return a dict or `DataModuleConfig`
) -> Union[DataModuleConfig, Dict]: # Return configs
"""Get `DataModuleConfig` from `FeaturesList`."""
cont, cats, immu = [], [], []
cont_transformation, cat_transformation = None, None
for f in features:
if f.is_categorical:
cats.append(f.name)
else:
cont.append(f.name)
if f.is_immutable:
immu.append(f.name)
configs_dict = {
"data_dir": ".",
"data_name": name,
"continous_cols": cont,
"discret_cols": cats,
"imutable_cols": immu,
"continuous_transformation": cont_transformation,
"discret_transformation": cat_transformation,
}
if return_dict:
return configs_dict
return DataModuleConfig(**configs_dict)
# %% ../nbs/01_data.ipynb 15
def features2pandas(
features: FeaturesList, # FeaturesList to be converted
labels: FeaturesList # labels to be converted
) -> pd.DataFrame: # Return pandas dataframe
"""Convert `FeaturesList` to pandas dataframe."""
feats_df = features.to_pandas()
labels_df = labels.to_pandas()
df = pd.concat([feats_df, labels_df], axis=1)
return df
# %% ../nbs/01_data.ipynb 18
def to_feature(col: str, data: pd.DataFrame, config: DataModuleConfig, transformation: str):
return Feature(
name=col, data=data[col].to_numpy().reshape(-1, 1),
transformation=transformation,
is_immutable=col in config.imutable_cols
)
# %% ../nbs/01_data.ipynb 19
def dataframe2features(
data: pd.DataFrame,
config: DataModuleConfig,
) -> FeaturesList:
"""Convert pandas dataframe of features to `FeaturesList`."""
cont_features = [to_feature(col, data, config, config.continuous_transformation) for col in config.continous_cols]
cat_features = [to_feature(col, data, config, config.discret_transformation) for col in config.discret_cols]
features = cont_features + cat_features
return FeaturesList(features)
def dataframe2labels(
data: pd.DataFrame,
config: DataModuleConfig,
) -> FeaturesList:
"""Convert pandas dataframe of labels to `FeaturesList`."""
label_cols = set(data.columns) - set(config.continous_cols) - set(config.discret_cols)
labels = [to_feature(col, data, config, 'identity') for col in label_cols]
return FeaturesList(labels)
# %% ../nbs/01_data.ipynb 21
class DataModule(BaseDataModule, DataModuleInfoMixin):
"""DataModule for tabular data."""
def __init__(
self,
features: FeaturesList,
label: FeaturesList,
config: DataModuleConfig = None,
data: pd.DataFrame = None,
**kwargs
):
self._prepare(features, label)
if config is None:
name = kwargs.pop('name', 'DataModule')
config = features2config(features, name)
config.shuffle(self.xs, test_size=0.25)
self._data = features2pandas(features, label) if data is None else data
super().__init__(config, name=config.data_name)
def _prepare(self, features, label):
if features is not None and label is not None:
self._features = FeaturesList(features)
self._label = FeaturesList(label)
elif features is None:
raise ValueError("Features cannot be None.")
elif label is None:
raise ValueError("Label cannot be None.")
def save(
self,
path: str # Path to the directory to save `DataModule`
):
"""Save `DataModule` to a directory."""
path = Path(path)
if not path.exists():
path.mkdir(parents=True)
self._features.save(path / 'features')
self._label.save(path / 'label')
if self._data is not None:
self._data.to_csv(path / 'data.csv', index=False)
with open(path / "config.json", "w") as f:
json.dump(self.config.dict(), f)
@classmethod
def load_from_path(
cls,
path: str, # Path to the directory to load `DataModule`
config: Dict|DataModuleConfig = None # Configs of `DataModule`. This argument is ignored.
) -> DataModule: # Initialized `DataModule` from path
"""Load `DataModule` from a directory."""
if config is not None:
warnings.warn("Passing `config` will have no effect.")
path = Path(path)
config = DataModuleConfig.load_from_json(path / 'config.json')
# config = validate_configs(config, DataModuleConfig)
features = FeaturesList.load_from_path(path / 'features')
label = FeaturesList.load_from_path(path / 'label')
data = pd.read_csv(path / 'data.csv')
return cls(features=features, label=label, config=config, data=data)
@classmethod
def from_path(cls, path, config: DataModuleConfig = None):
"""Alias of `load_from_path`"""
return cls.load_from_path(path, config)
@classmethod
def from_config(
cls,
config: Dict|DataModuleConfig, # Configs of `DataModule`
data: pd.DataFrame=None # Passed in pd.Dataframe
) -> DataModule: # Initialized `DataModule` from configs and data
config = validate_configs(config, DataModuleConfig)
if data is None:
data = pd.read_csv(config.data_dir)
if not isinstance(data, pd.DataFrame):
raise ValueError("`data` should be a pandas DataFrame.")
features = dataframe2features(data, config)
label = dataframe2labels(data, config)
return cls(features=features, label=label, config=config, data=data)
@classmethod
def from_numpy(
cls,
xs: np.ndarray, # Input data
ys: np.ndarray, # Labels
name: str = None, # Name of `DataModule`
transformation='minmax'
) -> DataModule: # Initialized `DataModule` from numpy arrays
"""Create `DataModule` from numpy arrays. Note that the `xs` are treated as continuous features."""
features = FeaturesList([Feature(f"feature_{i}", xs[:, i].reshape(-1, 1), transformation=transformation) for i in range(xs.shape[1])])
labels = FeaturesList([Feature(f"label", ys.reshape(-1, 1), transformation='identity')])
return cls(features=features, label=labels, name=name)
@classmethod
def from_features(
cls,
features: FeaturesList, # Features of `DataModule`
label: FeaturesList, # Labels of `DataModule`
name: str = None # Name of `DataModule`
) -> DataModule: # Initialized `DataModule` from features and labels
"""Create `DataModule` from `FeaturesList`."""
return cls(features=features, label=label, name=name)
def _get_data(self, indices):
if isinstance(indices, list):
indices = jnp.array(indices)
return (self.xs[indices], self.ys[indices])
def __getitem__(self, name: str):
if name == 'train':
return self._get_data(self.config.train_indices)
elif name in ['valid', 'test']:
return self._get_data(self.config.test_indices)
else:
raise ValueError(f"Unknown data name: {name}. Should be one of ['train', 'valid', 'test']")
def set_transformations(
self,
feature_names_to_transformation: Dict[str, Union[str, Dict, BaseTransformation]], # Dict[feature_name, Transformation]
) -> DataModule:
"""Reset transformations for features."""
self._features = self._features.set_transformations(feature_names_to_transformation)
return self
def sample(
self,
size: float | int, # Size of the sample. If float, should be 0<=size<=1.
stage: str = 'train', # Stage of data to sample from. Should be one of ['train', 'valid', 'test']
key: jrand.PRNGKey = None # Random key.
) -> Tuple[Array, Array]: # Sampled data
"""Sample data from `DataModule`."""
key = jrand.PRNGKey(get_config().global_seed) if key is None else key
xs, ys = self[stage]
indices = jnp.arange(xs.shape[0])
if isinstance(size, float) and 0 <= size <= 1:
size = int(size * indices.shape[0])
elif isinstance(size, int):
size = min(size, indices.shape[0])
else:
raise ValueError(f"`size` should be a floating number 0<=size<=1, or an integer,"
f" but got size={size}.")
indices = jrand.permutation(key, indices)[:size]
return xs[indices], ys[indices]
def transform(
self,
data: pd.DataFrame | Dict[str, Array] # Data to be transformed
) -> Array: # Transformed data
"""Transform data to `jax.Array`."""
# TODO: test this function
if isinstance(data, pd.DataFrame):
data_dict = {k: np.array(v).reshape(-1, 1) for k, v in data.iloc[:, :-1].to_dict(orient='list').items()}
return self._features.transform(data_dict)
elif isinstance(data, dict):
data = jax.tree_util.tree_map(lambda x: np.array(x).reshape(-1, 1), data)
return self._features.transform(data)
else:
raise ValueError("data should be a pandas DataFrame or `Dict[str, jax.Array]`.")
def inverse_transform(
self,
data: Array, # Data to be inverse transformed
return_type: str = 'pandas' # Type of the returned data. Should be one of ['pandas', 'dict']
) -> pd.DataFrame: # Inverse transformed data
"""Inverse transform data to `pandas.DataFrame`."""
# TODO: test this function
inversed = self._features.inverse_transform(data)
if return_type == 'pandas':
return inversed
elif return_type == 'dict':
raise NotImplementedError
else:
raise ValueError(f"Unknown return type: {return_type}. Should be one of ['pandas', 'dict']")
def apply_constraints(
self,
xs: Array, # Input data
cfs: Array, # Counterfactuals to be constrained
hard: bool = False, # Whether to apply hard constraints or not
rng_key: jrand.PRNGKey = None, # Random key
**kwargs
) -> Array: # Constrained counterfactuals
"""Apply constraints to counterfactuals."""
return self._features.apply_constraints(xs, cfs, hard=hard, rng_key=rng_key, **kwargs)
def compute_reg_loss(
self,
xs: Array, # Input data
cfs: Array, # Counterfactuals to be constrained
hard: bool = False # Whether to apply hard constraints or not
) -> float:
"""Compute regularization loss."""
return self._features.compute_reg_loss(xs, cfs, hard)
__ALL__ = [
'load_from_path',
'from_config',
'from_features',
'from_numpy',
'save',
'transform',
'inverse_transform',
'apply_constraints',
'compute_reg_loss',
'set_transformations',
'sample'
]
# %% ../nbs/01_data.ipynb 22
def dm_equals(dm1: DataModule, dm2: DataModule):
# data_equals = np.allclose(dm1.data.to_numpy(), dm2.data.to_numpy())
assert_frame_equal(dm1.data, dm2.data)
xs_equals = np.allclose(dm1.xs, dm2.xs)
ys_equals = np.allclose(dm1.ys, dm2.ys)
train_indices_equals = np.array_equal(dm1.train_indices, dm2.train_indices)
test_indices_equals = np.array_equal(dm1.test_indices, dm2.test_indices)
# print(f"data_equals: {data_equals}, xs_equals: {xs_equals}, ys_equals: {ys_equals}, train_indices_equals: {train_indices_equals}, test_indices_equals: {test_indices_equals}")
return (
xs_equals and ys_equals and
train_indices_equals and test_indices_equals
)
# %% ../nbs/01_data.ipynb 30
DEFAULT_DATA = [
'adult',
'heloc',
'oulad',
'credit',
'cancer',
'student_performance',
'titanic',
'german',
'spam',
'ozone',
'qsar',
'bioresponse',
'churn',
'road',
'dummy'
]
DEFAULT_DATA_CONFIGS = {
data: {
'data': f"{data}/data", 'model': f"{data}/model",
} for data in DEFAULT_DATA
}
# %% ../nbs/01_data.ipynb 35
def _validate_dataname(data_name: str):
if data_name not in DEFAULT_DATA:
raise ValueError(f'`data_name` must be one of {DEFAULT_DATA}, '
f'but got data_name={data_name}.')
# %% ../nbs/01_data.ipynb 36
def download_data_module_files(
data_name: str, # The name of data
data_parent_dir: Path, # The directory to save data.
download_original_data: bool = False, # Download original data or not
):
files = [
"features/data.npy", "features/treedef.json",
"label/data.npy", "label/treedef.json",
"config.json"
]
if download_original_data:
files.append("data.csv")
for f in files:
url = f"https://huggingface.co/datasets/birkhoffg/ReLax-Assets/resolve/main/{data_name}/data/{f}"
f_path = data_parent_dir / f'{data_name}/data' / f
os.makedirs(f_path.parent, exist_ok=True)
if not f_path.is_file(): urlretrieve(url, f_path)
def load_data(
data_name: str, # The name of data
return_config: bool = False, # Deprecated
data_configs: dict = None, # Data configs to override default configuration
) -> DataModule | Tuple[DataModule, DataModuleConfig]: # Return `DataModule` or (`DataModule`, `DataModuleConfig`)
"""High-level util function for loading `data` and `data_config`."""
_validate_dataname(data_name)
# create new dir
data_parent_dir = Path(os.getcwd()) / "relax-assets"
if not data_parent_dir.exists():
os.makedirs(data_parent_dir, exist_ok=True)
# download files
download_data_module_files(
data_name, data_parent_dir,
download_original_data=True
)
if return_config:
warnings.warn("`return_config` is deprecated since v0.2. "
"Please access `config` from `DataModule`.", DeprecationWarning)
# read and override config
# comment them for now since we cannot garantee the override configs are valid
# conf_path = data_parent_dir / f'{data_name}/data/config.json'
# config = load_json(conf_path)
# if not (data_configs is None):
# config.update(data_configs)
# config = DataModuleConfig(**config)
data_dir = data_parent_dir / f'{data_name}/data'
data_module = DataModule.load_from_path(data_dir, config=data_configs)
return data_module