alibi_detect.utils.perturbation

Functions

apply_mask

apply_mask(X: numpy.ndarray, mask_size: tuple = (4, 4), n_masks: int = 1, coord: Optional[tuple] = None, channels: list = [0, 1, 2], mask_type: str = 'uniform', noise_distr: tuple = (0, 1), noise_rng: tuple = (0, 1), clip_rng: tuple = (0, 1)) -> Tuple[numpy.ndarray, numpy.ndarray]

Mask images. Can zero out image patches or add normal or uniformly distributed noise.

Name	Type	Default	Description
X	numpy.ndarray		Batch of instances to be masked.
mask_size	tuple	(4, 4)	Tuple with the size of the mask.
n_masks	int	1	Number of masks applied for each instance in the batch X.
coord	Optional[tuple]	None	Upper left (x,y)-coordinates for the mask.
channels	list	[0, 1, 2]	Channels of the image to apply the mask to.
mask_type	str	'uniform'	Type of mask. One of 'uniform', 'random' (both additive noise) or 'zero' (zero values for mask).
noise_distr	tuple	(0, 1)	Mean and standard deviation for noise of 'random' mask type.
noise_rng	tuple	(0, 1)	Min and max value for noise of 'uniform' type.
clip_rng	tuple	(0, 1)	Min and max values for the masked instances.

Returns

• Type: Tuple[numpy.ndarray, numpy.ndarray]

brightness

brightness(x: numpy.ndarray, strength: float, xrange: Optional[tuple] = None) -> numpy.ndarray

Change brightness of image.

Name	Type	Default	Description
x	numpy.ndarray		Instance to be perturbed.
strength	float		Strength of brightness change.
xrange	Optional[tuple]	None	Tuple with min and max data range.

Returns

• Type: numpy.ndarray

clipped_zoom

clipped_zoom(x: numpy.ndarray, zoom_factor: float) -> numpy.ndarray

Helper function for zoom blur.

Name	Type	Default	Description
x	numpy.ndarray		Instance to be perturbed.
zoom_factor	float		Zoom strength.

Returns

• Type: numpy.ndarray

contrast

contrast(x: numpy.ndarray, strength: float, xrange: Optional[tuple] = None) -> numpy.ndarray

Change contrast of image.

Name	Type	Default	Description
x	numpy.ndarray		Instance to be perturbed.
strength	float		Strength of contrast change. Lower is actually more contrast.
xrange	Optional[tuple]	None	Tuple with min and max data range.

Returns

• Type: numpy.ndarray

defocus_blur

defocus_blur(x: numpy.ndarray, radius: int, alias_blur: float, xrange: Optional[tuple] = None) -> numpy.ndarray

Apply defocus blur.

Name	Type	Default	Description
x	numpy.ndarray		Instance to be perturbed.
radius	int		Radius for the Gaussian kernel.
alias_blur	float		Standard deviation for the Gaussian kernel in both X and Y directions.
xrange	Optional[tuple]	None	Tuple with min and max data range.

Returns

• Type: numpy.ndarray

disk

disk(radius: float, alias_blur: float = 0.1, dtype = <class 'numpy.float32'>) -> numpy.ndarray

Helper function for defocus blur.

Name	Type	Default	Description
radius	float		Radius for the Gaussian kernel.
alias_blur	float	0.1	Standard deviation for the Gaussian kernel in both X and Y directions.
dtype		<class 'numpy.float32'>	Data type.

Returns

• Type: numpy.ndarray

elastic_transform

elastic_transform(x: numpy.ndarray, mult_dxdy: float, sigma: float, rnd_rng: float, xrange: Optional[tuple] = None) -> numpy.ndarray

Apply elastic transformation to instance.

Name	Type	Default	Description
x	numpy.ndarray		Instance to be perturbed.
mult_dxdy	float		Multiplier for the Gaussian noise in x and y directions.
sigma	float		Standard deviation determining the strength of the Gaussian perturbation.
rnd_rng	float		Range for random uniform noise.
xrange	Optional[tuple]	None	Tuple with min and max data range.

Returns

• Type: numpy.ndarray

fog

fog(x: numpy.ndarray, fractal_mult: float, wibbledecay: float, xrange: Optional[tuple] = None) -> numpy.ndarray

Apply fog to instance.

Name	Type	Default	Description
x	numpy.ndarray		Instance to be perturbed.
fractal_mult	float		Strength applied to plasma_fractal output.
wibbledecay	float		Decay factor for size of noise that is applied.
xrange	Optional[tuple]	None	Tuple with min and max data range.

Returns

• Type: numpy.ndarray

gaussian_blur

gaussian_blur(x: numpy.ndarray, sigma: float, channel_axis: int = -1, xrange: Optional[tuple] = None) -> numpy.ndarray

Apply Gaussian blur.

Name	Type	Default	Description
x	numpy.ndarray		Instance to be perturbed.
sigma	float		Standard deviation determining the strength of the blur.
channel_axis	int	-1	Denotes the axis of the colour channel. If None the image is assumed to be grayscale.
xrange	Optional[tuple]	None	Tuple with min and max data range.

Returns

• Type: numpy.ndarray

gaussian_noise

gaussian_noise(x: numpy.ndarray, stdev: float, xrange: Optional[tuple] = None) -> numpy.ndarray

Inject Gaussian noise.

Name	Type	Default	Description
x	numpy.ndarray		Instance to be perturbed.
stdev	float		Standard deviation of noise.
xrange	Optional[tuple]	None	Tuple with min and max data range.

Returns

• Type: numpy.ndarray

glass_blur

glass_blur(x: numpy.ndarray, sigma: float, max_delta: int, iterations: int, xrange: Optional[tuple] = None) -> numpy.ndarray

Apply glass blur.

Name	Type	Default	Description
x	numpy.ndarray		Instance to be perturbed.
sigma	float		Standard deviation determining the strength of the Gaussian perturbation.
max_delta	int		Maximum pixel range for the blurring.
iterations	int		Number of blurring iterations.
xrange	Optional[tuple]	None	Tuple with min and max data range.

Returns

• Type: numpy.ndarray

impulse_noise

impulse_noise(x: numpy.ndarray, amount: float, xrange: Optional[tuple] = None) -> numpy.ndarray

Inject salt & pepper noise.

Name	Type	Default	Description
x	numpy.ndarray		Instance to be perturbed.
amount	float		Proportion of pixels to replace with noise.
xrange	Optional[tuple]	None	Tuple with min and max data range.

Returns

• Type: numpy.ndarray

inject_outlier_categorical

inject_outlier_categorical(X: numpy.ndarray, cols: List[int], perc_outlier: int, y: Optional[numpy.ndarray] = None, cat_perturb: Optional[dict] = None, X_fit: Optional[numpy.ndarray] = None, disc_perc: list = [25, 50, 75], smooth: float = 1.0) -> alibi_detect.utils.data.Bunch

Inject outliers in categorical variables of tabular data.

Name	Type	Default	Description
X	numpy.ndarray		Tabular data with categorical variables to perturb (inject outliers).
cols	List[int]		Columns of X that are categorical and can be perturbed.
perc_outlier	int		Percentage of observations which are perturbed to outliers. For multiple numerical features, the percentage is evenly split across the features.
y	Optional[numpy.ndarray]	None	Outlier labels.
cat_perturb	Optional[dict]	None	Dictionary mapping each category in the categorical variables to their furthest neighbour.
X_fit	Optional[numpy.ndarray]	None	Optional data used to infer pairwise distances from.
disc_perc	list	[25, 50, 75]	List with percentiles used in binning of numerical features used for the 'abdm' pairwise distance measure.
smooth	float	1.0	Smoothing exponent between 0 and 1 for the distances. Lower values will smooth the difference in distance metric between different features.

Returns

• Type: alibi_detect.utils.data.Bunch

inject_outlier_tabular

inject_outlier_tabular(X: numpy.ndarray, cols: List[int], perc_outlier: int, y: Optional[numpy.ndarray] = None, n_std: float = 2.0, min_std: float = 1.0) -> alibi_detect.utils.data.Bunch

Inject outliers in numerical tabular data.

Name	Type	Default	Description
X	numpy.ndarray		Tabular data to perturb (inject outliers).
cols	List[int]		Columns of X that are numerical and can be perturbed.
perc_outlier	int		Percentage of observations which are perturbed to outliers. For multiple numerical features, the percentage is evenly split across the features.
y	Optional[numpy.ndarray]	None	Outlier labels.
n_std	float	2.0	Number of feature-wise standard deviations used to perturb the original data.
min_std	float	1.0	Minimum number of standard deviations away from the current observation. This is included because of the stochastic nature of the perturbation which could lead to minimal perturbations without a floor.

Returns

• Type: alibi_detect.utils.data.Bunch

inject_outlier_ts

inject_outlier_ts(X: numpy.ndarray, perc_outlier: int, perc_window: int = 10, n_std: float = 2.0, min_std: float = 1.0) -> alibi_detect.utils.data.Bunch

Inject outliers in both univariate and multivariate time series data.

Name	Type	Default	Description
X	numpy.ndarray		Time series data to perturb (inject outliers).
perc_outlier	int		Percentage of observations which are perturbed to outliers. For multivariate data, the percentage is evenly split across the individual time series.
perc_window	int	10	Percentage of the observations used to compute the standard deviation used in the perturbation.
n_std	float	2.0	Number of standard deviations in the window used to perturb the original data.
min_std	float	1.0	Minimum number of standard deviations away from the current observation. This is included because of the stochastic nature of the perturbation which could lead to minimal perturbations without a floor.

Returns

• Type: alibi_detect.utils.data.Bunch

jpeg_compression

jpeg_compression(x: numpy.ndarray, strength: float, xrange: Optional[tuple] = None) -> numpy.ndarray

Simulate changes due to JPEG compression for an image.

Name	Type	Default	Description
x	numpy.ndarray		Instance to be perturbed.
strength	float		Strength of compression (>1). Lower is actually more compressed.
xrange	Optional[tuple]	None	Tuple with min and max data range.

Returns

• Type: numpy.ndarray

pixelate

pixelate(x: numpy.ndarray, strength: float, xrange: Optional[tuple] = None) -> numpy.ndarray

Change coarseness of pixels for an image.

Name	Type	Default	Description
x	numpy.ndarray		Instance to be perturbed.
strength	float		Strength of pixelation (<1). Lower is actually more pixelated.
xrange	Optional[tuple]	None	Tuple with min and max data range.

Returns

• Type: numpy.ndarray

plasma_fractal

plasma_fractal(mapsize: int = 256, wibbledecay: float = 3.0) -> numpy.ndarray

Helper function to apply fog to instance.

Generates a heightmap using diamond-square algorithm. Returns a square 2d array, side length 'mapsize', of floats in range 0-255. 'mapsize' must be a power of two.

Name	Type	Default	Description
mapsize	int	256
wibbledecay	float	3.0

Returns

• Type: numpy.ndarray

saturate

saturate(x: numpy.ndarray, strength: tuple, xrange: Optional[tuple] = None) -> numpy.ndarray

Change colour saturation of image.

Name	Type	Default	Description
x	numpy.ndarray		Instance to be perturbed.
strength	tuple		Strength of saturation change. Tuple consists of (multiplier, shift) of the perturbation.
xrange	Optional[tuple]	None	Tuple with min and max data range.

Returns

• Type: numpy.ndarray

scale_minmax

scale_minmax(x: numpy.ndarray, xrange: Optional[tuple] = None) -> Tuple[numpy.ndarray, bool]

Minmax scaling to [0,1].

Name	Type	Default	Description
x	numpy.ndarray		Numpy array to be scaled.
xrange	Optional[tuple]	None	Tuple with min and max data range.

Returns

• Type: Tuple[numpy.ndarray, bool]

shot_noise

shot_noise(x: numpy.ndarray, lam: float, xrange: Optional[tuple] = None) -> numpy.ndarray

Inject Poisson noise.

Name	Type	Default	Description
x	numpy.ndarray		Instance to be perturbed.
lam	float		Scalar for the lambda parameter determining the expectation of the interval.
xrange	Optional[tuple]	None	Tuple with min and max data range.

Returns

• Type: numpy.ndarray

speckle_noise

speckle_noise(x: numpy.ndarray, stdev: float, xrange: Optional[tuple] = None) -> numpy.ndarray

Inject speckle noise.

Name	Type	Default	Description
x	numpy.ndarray		Instance to be perturbed.
stdev	float		Standard deviation of noise.
xrange	Optional[tuple]	None	Tuple with min and max data range.

Returns

• Type: numpy.ndarray

zoom_blur

zoom_blur(x: numpy.ndarray, max_zoom: float, step_zoom: float, xrange: Optional[tuple] = None) -> numpy.ndarray

Apply zoom blur.

Name	Type	Default	Description
x	numpy.ndarray		Instance to be perturbed.
max_zoom	float		Max zoom strength.
step_zoom	float		Step size to go from 1 to max_zoom strength.
xrange	Optional[tuple]	None	Tuple with min and max data range.

Returns

• Type: numpy.ndarray