spacekit.generator.augment

spacekit.generator.augment.aug_generator(X, c=9, combine=False)

Generates randomly performed image augmentations. Optionally concatenates this data with original.

Parameters:

• X (numpy array) – array of image arrays(pixel vlues)

• c (int, optional) – channel depth (product of dim and ch), by default 9

• combine (bool, optional) – concatenate original with augmented data, by default False

Returns:

randomly augmented image data (merged with original data if combine set to True).

Return type:

numpy array

spacekit.generator.augment.augment_data(xi)

Randomly apply noise to input data. For now this function is designed for SVM ensemble MLP input data. For other uses, the function tries to determine which augmentation operation to apply according to the data type - however it may not make sense for all the inputs to be augmented (e.g. if they are binary/boolean) so this should be used with caution (future updates this will be generalized properly for other use cases).

Parameters:

xi (numpy array) – 1d-array of input data

Returns:

input data with random augmentations applied

Return type:

numpy array

spacekit.generator.augment.augment_image(x, c=None, w=128, h=128, ch=3, dim=3)

Randomly applies a series of augmentation transformations. Except for random crops, random application is required to impose translational invariance.

Parameters:

• x (numpy array) – input image array

• c (int, optional) – depth (product of dim and ch), by default None

• w (int, optional) – image width, by default 128

• h (int, optional) – image height, by default 128

• ch (int, optional) – channels, by default 3

• dim (int, optional) – dimensions (or volume of image frames per input), by default 3

Returns:

random augmentation of input image

Return type:

numpy array

spacekit.generator.augment.augment_random_integer(x)

Randomly apply augmentations to integer values. Assumes final values must be absolute/non-negative.

Parameters:

x (int) – input value (integer)

Returns:

absolute (non-negative) augmented integer

Return type:

int

spacekit.generator.augment.augment_random_noise(x)

Random application of laplacian and/or logistic noise calculated relative to the input value. Augmentation transformations are applied randomly to impose translational invariance.

Parameters:

x (float) – input value

Returns:

transformed input value

Return type:

float

spacekit.generator.augment.color_jitter(x, brightness=0.4, contrast=0.4, saturation=0.4, hue=None)

Randomly applies a combination of alterations to the image brightness, contrast, saturation and hue. Note: a clipping function is performed at the end since affine transformations can disturb the natural range of RGB images.

Parameters:

• x (numpy array) – input image array

• brightness (float, optional) – decimal percentage strength applied to brightness level, by default 0.4

• contrast (float, optional) – decimal percentage strength applied to contrast level, by default 0.4

• saturation (float, optional) – decimal percentage strength applied to saturation level, by default 0.4

• hue (float, optional) – decimal percentage strength applied to hue (e.g. 0.1), by default None

Returns:

transformed image input array

Return type:

numpy array

spacekit.generator.augment.expand_dims(X, dim=3, w=128, h=128, ch=3)

Expand flattened array back into its constituent dimensions.

Parameters:

• X (numpy array) – array of image arrays

• dim (int, optional) – dimensions (volume of image frames per input), by default 3

• w (int, optional) – single image width, by default 128

• h (int, optional) – single image height, by default 128

• ch (int, optional) – channels, by default 3

Returns:

numpy array of shape – Reshaped multi-dimensional image array

Return type:

(X.shape[0], dim, w, h, ch)

spacekit.generator.augment.flip_horizontal(x)

Horizontal (x-axis) image flip

Parameters:

x (numpy array) – input image array

Returns:

horizontally flipped image

Return type:

numpy array

spacekit.generator.augment.flip_vertical(x)

Vertical (y-axis) image flip

Parameters:

x (numpy array) – input image array

Returns:

vertically flipped image

Return type:

numpy array

spacekit.generator.augment.image_index_labels(index, y, aug=False)

Creates Pandas Index and Series from numpy arrays (used for calculating FNFP data in spacekit.analyzer.compute)

Parameters:

• index (ndarray) – index of image names

• y (ndarray) – target class values

• aug (bool, optional) – concatenate (double) the index for training data, by default False

Returns:

inputs converted into index and series

Return type:

tuple of pd.Index and pd.Series

spacekit.generator.augment.laplacian_noise(x)

Calculates laplacian value relative to the input.

Parameters:

x (float) – input value

Returns:

random laplacian value relative to input.

Return type:

float

spacekit.generator.augment.logistic_noise(x)

Calculates random logistic value relative to the input.

Parameters:

x (float) – input value

Returns:

random logistic value relative to input.

Return type:

float

spacekit.generator.augment.nested_image_index(tr, ts, vl=())

creates a list of nested tuples for training, test, and validation Indices and Index-label Series

Parameters:

• tr (tuple) – (training index, training index-label series)

• ts (tuple) – (test index, test index-label series)

• vl (tuple, optional) – (val index, val index-label series), by default ()

Returns:

list of nested tuples [(indices), (tuples)]

Return type:

list

spacekit.generator.augment.random_apply(func, x, p)

Determine whether or not to apply a transformation according to a randomly calculated probability value.

Parameters:

• func (function) – augmentation function to apply if random number is less than p.

• x (float) – input value

• p (float) – probability between 0 and 1 used to determine whether or not to apply the augmentation.

Returns:

transformed input (or original input if no transformation is performed).

Return type:

float

spacekit.generator.augment.rotate_k(x)

Rotate image array k times by 90 degrees. K is a number between 0 and 3 determined randomly when the function is called.

Parameters:

x (numpy array) – input image array

Returns:

randomly rotated image

Return type:

numpy array

spacekit.generator.augment.training_data_aug(X_train, y_train)

Perform data augmentation on the training set

Parameters:

• X_train (dataframe) – training set features

• y_train (dataframe) – training set target labels

Returns:

augmented training data and target labels combined with original set (2x observations)

Return type:

dataframes

spacekit.generator.augment.training_img_aug(train, test, val=None, dim=3, w=128, h=128, ch=3)

Perform image data augmentation on the training set

Parameters:

• train (tuple) – training set tuple of (train_index, X_train, y_train)

• test (tuple) – test set tuple of (test_index, X_test, y_test)

• val (tuple, optional) – val set tuple of (val_index, X_val, y_val), by default None

• dim (int, optional) – dimensions (volume or number of image frames per observation), by default 3

• w (int, optional) – single image width, by default 128

• h (int, optional) – single image height, by default 128

• ch (int, optional) – channels (rgb: 3, grayscale: 1), by default 3

Returns:

combined image index of train, test, val; combined original and augmented training data, reshaped test and val data. If no validation data is passed in the val arg, then only train and test are returned.

Return type:

list of tuples of Pandas Index, numpy arrays