mlnext.data.detemporalize#

mlnext.data.detemporalize(data: ndarray, *, stride: int = 0, last_point_only: bool = False, verbose: bool = False) → ndarray[source]#

Transforms a 3 dimensional array (rows, timesteps, features) into a 2 dimensional array (new_rows, features). If stride >= timesteps or 0, then the operation is equivalent to data.reshape(-1, features) and new_rows equals rows * timesteps. If 0 < stride < timesteps, the stride induced elements will be removed and new_rows equals (rows - timesteps) * timesteps. If last_point_only=True then only the last point in each window is kept and new_rows equals (rows, features).

Parameters:

data (np.ndarray) – Array to transform.
stride (np.ndarray) – Stride that was used to transform the array from 2d into 3d.
last_point_only (np.ndarray) – Whether to only take the last point of each window.
verbose (bool) – Whether to print old and new shape.

Returns:

Returns an array of shape (rows * timesteps) x features.

Return type:

np.ndarray

Example

>>> import numpy as np
>>> import mlnext

>>> # setup data
>>> i, j = np.ogrid[:6, :3]
>>> data = 10 * i + j
>>> print(data)
[[ 0  1  2]
 [10 11 12]
 [20 21 22]
 [30 31 32]
 [40 41 42]
 [50 51 52]]

>>> # Transform 3d data into 2d
>>> data_3d = mlnext.temporalize(data, timesteps=2)
>>> print(data_3d)
[[[ 0  1  2]
  [10 11 12]]
 [[20 21 22]
   [30 31 32]]
 [[40 41 42]
  [50 51 52]]]
>>> mlnext.detemporalize(data_3d, verbose=True)
Old shape: (3, 2, 3). New shape: (6, 3).
[[ 0  1  2]
 [10 11 12]
 [20 21 22]
 [30 31 32]
 [40 41 42]
 [50 51 52]]

>>> # Transform 3d data into 2d with stride=1
>>> data_3d = mlnext.temporalize(data,
... timesteps=3, stride=1, verbose=True)
Old shape: (6, 3). New shape: (4, 3, 3).
>>> print(data_3d)
[[[ 0  1  2]
  [10 11 12]
  [20 21 22]]
 [[10 11 12]
  [20 21 22]
  [30 31 32]]
 [[20 21 22]
  [30 31 32]
  [40 41 42]]
 [[30 31 32]
  [40 41 42]
  [50 51 52]]]
>>> mlnext.detemporalize(data_3d, stride=1, verbose=True)
Old shape: (4, 3, 3). New shape: (6, 3).
[[ 0  1  2]
 [10 11 12]
 [20 21 22]
 [30 31 32]
 [40 41 42]
 [50 51 52]]
>>> # Take only the last point from each window
>>> mlnext.detemporalize(data_3d, last_point_only=True, verbose=True)
Old shape: (4, 3, 3). New shape: (4, 3).
[[20 21 22]
 [30 31 32]
 [40 41 42]
 [50 51 52]]