Utils

IngeoML.utils API

class Batches

Helper class to create a set of batches.

Parameters:

• size (int) – Bath size, default=64

• strategy (str) – Procedure to create the batch, default=stratified

• remainder (str) – Method used to deal with the remainder, default=fill

• shuffle (bool) – Whether to shuffle the dataset, default=True

• random_state – Random State, default=None

>>> import numpy as np
>>> from IngeoML.utils import Batches
>>> b = Batches(size=3)
>>> X = np.empty((5, 4))
>>> b.split(X)
array([[4, 0, 2],
       [1, 3, 4]])
>>> y = np.r_[0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 2]
>>> b.split(y=y)
array([[ 0, 10,  5],
       [ 1,  6, 10],
       [ 2, 10,  7],
       [ 3, 10,  8],
       [10,  9,  4]])    

__init__(size: int = 64, strategy: str = 'stratified', remainder: str = 'fill', shuffle: bool = True, random_state: int = None) → None

blocks(index: ndarray, rows: int, columns: int)

Create the blocks :param index: :type index: np.ndarray :param rows: Number of rows :type rows: int :param columns: Number of columns :type columns: int

>>> from IngeoML.utils import Batches
>>> b = Batches(size=3)
>>> b.blocks(np.arange(3), columns=2, rows=3)
array([[0, 1],
       [2, 0],
       [1, 2]])        

static distribution(y: ndarray, size: int = 64)

Distribution

Parameters:

• y (np.ndarray) – Labels

• size – Size of the batch

split(D=None, y: ndarray = None) → ndarray

Method to create the batches

Parameters:

• D – Dataset

• y (np.ndarray) – Labels

__new__(**kwargs)

balance_class_weights(labels) → ndarray

Weights of the labels set to balance

>>> import numpy as np
>>> from IngeoML.utils import balance_class_weights
>>> balance_class_weights(np.array(['a', 'a', 'b']))
array([0.25, 0.25, 0.5 ])