Skip to content

Fillers

cesnet_tszoo.utils.filler.filler

Filler

Bases: ABC

Base class for data fillers.

This class serves as the foundation for creating custom fillers. To implement a custom filler, this class must be subclassed and extended. Fillers are used to handle missing data in a dataset.

Example:

import numpy as np

class ForwardFiller(Filler):

    def __init__(self):
        self.last_values = {}
        self.initialized = False

    def __init_attributes(self, batch_values: np.ndarray):
        for name in batch_values.dtype.names:
            self.last_values[name] = None

        self.initialized = True

    def __fill_section(self, values: np.ndarray, missing_mask: np.ndarray, last_values: np.ndarray, name: str) -> np.ndarray:
        if last_values is not None and np.any(missing_mask[0]):
            values[0, missing_mask[0]] = last_values[missing_mask[0]]

        orig_shape = values.shape
        t = orig_shape[0]
        flat_size = int(np.prod(orig_shape[1:]))

        values_2d = values.reshape(t, flat_size)
        mask_2d = missing_mask.reshape(t, flat_size)

        mask = mask_2d.T
        values_t = values_2d.T

        idx = np.where(~mask, np.arange(mask.shape[1]), 0)
        np.maximum.accumulate(idx, axis=1, out=idx)

        values_t[mask] = values_t[np.nonzero(mask)[0], idx[mask]]
        values_t = values_t.T

        values = values_2d.reshape(orig_shape)

        self.last_values[name] = np.copy(values[-1])

        return values
Source code in cesnet_tszoo\utils\filler\filler.py 
 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
class Filler(ABC):
    """
    Base class for data fillers.

    This class serves as the foundation for creating custom fillers. To implement a custom filler, this class must be subclassed and extended.
    Fillers are used to handle missing data in a dataset.

    Example:

        import numpy as np

        class ForwardFiller(Filler):

            def __init__(self):
                self.last_values = {}
                self.initialized = False

            def __init_attributes(self, batch_values: np.ndarray):
                for name in batch_values.dtype.names:
                    self.last_values[name] = None

                self.initialized = True

            def __fill_section(self, values: np.ndarray, missing_mask: np.ndarray, last_values: np.ndarray, name: str) -> np.ndarray:
                if last_values is not None and np.any(missing_mask[0]):
                    values[0, missing_mask[0]] = last_values[missing_mask[0]]

                orig_shape = values.shape
                t = orig_shape[0]
                flat_size = int(np.prod(orig_shape[1:]))

                values_2d = values.reshape(t, flat_size)
                mask_2d = missing_mask.reshape(t, flat_size)

                mask = mask_2d.T
                values_t = values_2d.T

                idx = np.where(~mask, np.arange(mask.shape[1]), 0)
                np.maximum.accumulate(idx, axis=1, out=idx)

                values_t[mask] = values_t[np.nonzero(mask)[0], idx[mask]]
                values_t = values_t.T

                values = values_2d.reshape(orig_shape)

                self.last_values[name] = np.copy(values[-1])

                return values
            """

    @abstractmethod
    def fill(self, batch_values: np.ndarray, missing_masks: dict[str, np.ndarray], **kwargs) -> np.ndarray:
        """Fills missing data in the `batch_values`.

        This method is responsible for filling missing data within a single time series.

        Parameters:
            batch_values: A structured numpy array representing data for a single time series with shape `(times)`. Use data["base_data"] to get non matrix features excluding any identifiers. 
                          For matrix features use their name instead of base_data. 
            missing_masks: Masks of missing values in batch_values. Keys is "base_data" and names of matrix features.
            kwargs: default_values 

        Returns:
            The filled data, with the same shape and dtype as the input `(times)`.               
        """
        ...

fill abstractmethod 

fill(batch_values: ndarray, missing_masks: dict[str, ndarray], **kwargs) -> np.ndarray

Fills missing data in the batch_values.

This method is responsible for filling missing data within a single time series.

Parameters:

Name Type Description Default
batch_values ndarray

A structured numpy array representing data for a single time series with shape (times). Use data["base_data"] to get non matrix features excluding any identifiers. For matrix features use their name instead of base_data.

required
missing_masks dict[str, ndarray]

Masks of missing values in batch_values. Keys is "base_data" and names of matrix features.

 required
kwargs

default_values

{}

Returns:

Type Description
ndarray

The filled data, with the same shape and dtype as the input (times).
Source code in cesnet_tszoo\utils\filler\filler.py 
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
@abstractmethod
def fill(self, batch_values: np.ndarray, missing_masks: dict[str, np.ndarray], **kwargs) -> np.ndarray:
    """Fills missing data in the `batch_values`.

    This method is responsible for filling missing data within a single time series.

    Parameters:
        batch_values: A structured numpy array representing data for a single time series with shape `(times)`. Use data["base_data"] to get non matrix features excluding any identifiers. 
                      For matrix features use their name instead of base_data. 
        missing_masks: Masks of missing values in batch_values. Keys is "base_data" and names of matrix features.
        kwargs: default_values 

    Returns:
        The filled data, with the same shape and dtype as the input `(times)`.               
    """
    ...

MeanFiller

Bases: Filler

Fills values from total mean of all previous values.

Corresponds to enum FillerType.MEAN_FILLER or literal mean_filler.

Source code in cesnet_tszoo\utils\filler\filler.py 
 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
class MeanFiller(Filler):
    """
    Fills values from total mean of all previous values.

    Corresponds to enum [`FillerType.MEAN_FILLER`][cesnet_tszoo.utils.enums.FillerType] or literal `mean_filler`.
    """

    def __init__(self):
        self.initialized = False
        self.averages = {}
        self.total_existing_values = {}

    def __init_attributes(self, batch_values: np.ndarray):
        for name in batch_values.dtype.names:
            base_shape = batch_values[name].shape[1:]
            self.averages[name] = np.zeros((1, *base_shape), dtype=np.float64)
            self.total_existing_values[name] = np.zeros((1, *base_shape), dtype=np.float64)

        self.initialized = True

    def __fill_section(self, values: np.ndarray, missing_mask: np.ndarray, averages: np.ndarray, total_existing_values: np.ndarray, name: str) -> np.ndarray:
        existing_mask = ~missing_mask
        batch_counts = np.cumsum(existing_mask, axis=0)
        batch_sums = np.cumsum(np.where(existing_mask, values, 0.0), axis=0)

        prev_counts = total_existing_values
        total_counts = batch_counts + prev_counts

        with np.errstate(invalid="ignore", divide="ignore"):
            running_avg = ((averages / total_counts) * prev_counts) + (batch_sums / total_counts)

        running_avg = np.vstack([averages, running_avg[:-1]])

        fill_mask = missing_mask & (total_counts > 0)
        values[fill_mask] = running_avg[fill_mask]

        total_existing_values = total_counts[-1]

        valid_cols = total_existing_values > 0

        self.averages[name][0, valid_cols] = running_avg[-1, valid_cols]
        self.total_existing_values[name] = total_existing_values

        return values

    def fill(self, batch_values: np.ndarray, missing_masks: dict[str, np.ndarray], **kwargs) -> np.ndarray:

        if not self.initialized:
            self.__init_attributes(batch_values)

        for name in batch_values.dtype.names:

            values = batch_values[name].view()
            missing_mask = missing_masks[name]
            averages = self.averages[name]
            total_existing_values = self.total_existing_values[name]

            self.__fill_section(values, missing_mask, averages, total_existing_values, name)

        return batch_values

ForwardFiller

Bases: Filler

Fills missing values based on last existing value.

Corresponds to enum FillerType.FORWARD_FILLER or literal forward_filler.

Source code in cesnet_tszoo\utils\filler\filler.py 
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
class ForwardFiller(Filler):
    """
    Fills missing values based on last existing value. 

    Corresponds to enum [`FillerType.FORWARD_FILLER`][cesnet_tszoo.utils.enums.FillerType] or literal `forward_filler`.
    """

    def __init__(self):
        self.last_values = {}
        self.initialized = False

    def __init_attributes(self, batch_values: np.ndarray):
        for name in batch_values.dtype.names:
            self.last_values[name] = None

        self.initialized = True

    def __fill_section(self, values: np.ndarray, missing_mask: np.ndarray, last_values: np.ndarray, name: str) -> np.ndarray:
        if last_values is not None and np.any(missing_mask[0]):
            values[0, missing_mask[0]] = last_values[missing_mask[0]]

        orig_shape = values.shape
        t = orig_shape[0]
        flat_size = int(np.prod(orig_shape[1:]))

        values_2d = values.reshape(t, flat_size)
        mask_2d = missing_mask.reshape(t, flat_size)

        mask = mask_2d.T
        values_t = values_2d.T

        idx = np.where(~mask, np.arange(mask.shape[1]), 0)
        np.maximum.accumulate(idx, axis=1, out=idx)

        values_t[mask] = values_t[np.nonzero(mask)[0], idx[mask]]
        values_t = values_t.T

        values = values_2d.reshape(orig_shape)

        self.last_values[name] = np.copy(values[-1])

        return values

    def fill(self, batch_values: np.ndarray, missing_masks: dict[str, np.ndarray], **kwargs) -> np.ndarray:

        if not self.initialized:
            self.__init_attributes(batch_values)

        for name in batch_values.dtype.names:

            values = batch_values[name].view()
            missing_mask = missing_masks[name]
            last_values = self.last_values[name]

            self.__fill_section(values, missing_mask, last_values, name)

        return batch_values

LinearInterpolationFiller

Bases: Filler

Fills values with linear interpolation.

Corresponds to enum FillerType.LINEAR_INTERPOLATION_FILLER or literal linear_interpolation_filler.

Source code in cesnet_tszoo\utils\filler\filler.py 
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
class LinearInterpolationFiller(Filler):
    """
    Fills values with linear interpolation. 

    Corresponds to enum [`FillerType.LINEAR_INTERPOLATION_FILLER`][cesnet_tszoo.utils.enums.FillerType] or literal `linear_interpolation_filler`.
    """

    def __init__(self):
        self.last_values = {}
        self.last_values_x_pos = {}
        self.initialized = False

    def __init_attributes(self, batch_values: np.ndarray):
        for name in batch_values.dtype.names:
            self.last_values[name] = None
            self.last_values_x_pos[name] = None

        self.initialized = True

    def __flatten(self, values: np.ndarray, mask: np.ndarray, last_values: np.ndarray | None) -> tuple[np.ndarray]:
        orig_shape = values.shape
        t = orig_shape[0]
        flat_size = int(np.prod(orig_shape[1:]))

        values_2d = values.reshape(t, flat_size)
        mask_2d = mask.reshape(t, flat_size)

        if last_values is not None:
            last_values = last_values.reshape(flat_size)

        return values_2d, mask_2d, orig_shape, last_values

    def __restore_shape(self, values_2d: np.ndarray, orig_shape: tuple) -> np.ndarray:
        return values_2d.reshape(orig_shape)

    def __fill_section(self, values: np.ndarray, missing_mask: np.ndarray, last_values: np.ndarray | None, last_values_x_pos: np.ndarray, default_values: np.ndarray, name: str) -> np.ndarray:
        values, missing_mask, orig_shape, last_values = self.__flatten(values, missing_mask, last_values)
        t, flat_size = values.shape
        x_positions = np.arange(t)

        existing_mask = ~missing_mask
        any_existing = np.any(existing_mask, axis=0)
        any_missing = np.any(missing_mask, axis=0)

        if not np.any(any_missing):
            self.last_values[name] = np.copy(values[-1, :])
            self.last_values_x_pos[name] = -1
            return

        if not np.any(any_existing) and last_values is None:
            return

        for i in range(flat_size):
            existing_x = x_positions[existing_mask[:, i]]
            existing_y = values[existing_mask[:, i], i]

            if last_values is not None:
                existing_x = np.insert(existing_x, 0, last_values_x_pos)
                existing_y = np.insert(existing_y, 0, last_values[i])

            missing_x = x_positions[missing_mask[:, i]]

            if len(missing_x) > 0 and len(existing_x) > 0:
                values[missing_mask[:, i], i] = np.interp(
                    missing_x,
                    existing_x,
                    existing_y,
                    left=default_values[i],
                    right=default_values[i],
                )

        values = self.__restore_shape(values, orig_shape)

        self.last_values[name] = np.copy(values[-1, :])
        self.last_values_x_pos[name] = -1

        return values

    def fill(self, batch_values: np.ndarray, missing_masks: dict[str, np.ndarray], **kwargs) -> np.ndarray:

        if not self.initialized:
            self.__init_attributes(batch_values)

        for name in batch_values.dtype.names:

            default_values = kwargs["default_values"][name]
            values = batch_values[name].view()
            missing_mask = missing_masks[name]
            last_values = self.last_values[name]
            last_values_x_pos = self.last_values_x_pos[name]

            batch_values[name] = self.__fill_section(values, missing_mask, last_values, last_values_x_pos, default_values, name)

        return batch_values

NoFiller

Bases: Filler

Does nothing.

Corresponds to enum FillerType.NO_FILLER or literal no_filler.

Source code in cesnet_tszoo\utils\filler\filler.py 
292
293
294
295
296
297
298
299
300
class NoFiller(Filler):
    """
    Does nothing. 

    Corresponds to enum [`FillerType.NO_FILLER`][cesnet_tszoo.utils.enums.FillerType] or literal `no_filler`.
    """

    def fill(self, batch_values: np.ndarray, missing_masks: dict[str, np.ndarray], **kwargs) -> np.ndarray:
        return batch_values