Data

Simulated Data

cca_zoo.data.simulated.generate_covariance_data(n, view_features, latent_dims=1, view_sparsity=None, correlation=1, structure=None, sigma=None, decay=0.5, positive=None, random_state=None)

Function to generate CCA dataset with defined population correlations

Parameters:

• n (int) – number of samples
• view_sparsity (Optional[List[Union[int, float]]]) – level of sparsity in features in each view either as number of active variables or percentage active
• view_features (List[int]) – number of features in each view
• latent_dims (int) – number of latent dimensions
• correlation (Union[List[float], float]) – correlation either as list with element for each latent dimension or as float which is scaled by ‘decay’
• structure (Union[str, List[str], None]) – within view covariance structure (‘identity’,’gaussian’,’toeplitz’,’random’)
• sigma (Union[List[float], float, None]) – gaussian sigma
• decay (float) – ratio of second signal to first signal

Returns:

tuple of numpy arrays: view_1, view_2, true weights from view 1, true weights from view 2, overall covariance structure

Example:

>>> from cca_zoo.data import generate_covariance_data
>>> [train_view_1,train_view_2],[true_weights_1,true_weights_2]=generate_covariance_data(200,[10,10],latent_dims=1,correlation=1)

cca_zoo.data.simulated.generate_simple_data(n, view_features, view_sparsity=None, eps=0, transform=False, random_state=None)

Simple latent variable model to generate data with one latent factor

Parameters:

• n (int) – number of samples
• view_features (List[int]) – number of features view 1
• view_sparsity (Optional[List[Union[int, float]]]) – number of features view 2
• eps (float) – gaussian noise std

Returns:

view1 matrix, view2 matrix, true weights view 1, true weights view 2

Example:

>>> from cca_zoo.data import generate_simple_data
>>> [train_view_1,train_view_2],[true_weights_1,true_weights_2]=generate_covariance_data(200,[10,10])

Toy Data

Helped by https://github.com/bcdutton/AdversarialCanonicalCorrelationAnalysis (hopefully I will have my own implementation of their work soon) Check out their paper at https://arxiv.org/abs/2005.10349

class cca_zoo.data.toy.Split_MNIST_Dataset(mnist_type='MNIST', train=True, flatten=True)

Bases: torch.utils.data.dataset.Dataset

Class to generate paired noisy mnist data

Parameters:

• mnist_type (str) – “MNIST”, “FashionMNIST” or “KMNIST”
• train (bool) – whether this is train or test
• flatten (bool) – whether to flatten the data into array or use 2d images

to_numpy(indices=None)

Converts dataset to numpy array form

Parameters:indices – indices of the samples to extract into numpy arrays

class cca_zoo.data.toy.Noisy_MNIST_Dataset(mnist_type='MNIST', train=True, flatten=True)

Bases: torch.utils.data.dataset.Dataset

Class to generate paired noisy mnist data

Parameters:

• mnist_type (str) – “MNIST”, “FashionMNIST” or “KMNIST”
• train (bool) – whether this is train or test
• flatten (bool) – whether to flatten the data into array or use 2d images

class cca_zoo.data.toy.Tangled_MNIST_Dataset(mnist_type='MNIST', train=True, flatten=True)

Bases: torch.utils.data.dataset.Dataset

Class to generate paired tangled MNIST dataset

Parameters:

• mnist_type – “MNIST”, “FashionMNIST” or “KMNIST”
• train – whether this is train or test
• flatten – whether to flatten the data into array or use 2d images

Utils

class cca_zoo.data.utils.CCA_Dataset(views, labels=None)

Bases: torch.utils.data.dataset.Dataset

Class that turns numpy arrays into a torch dataset

Parameters:

• views – list/tuple of numpy arrays or array likes with the same number of rows (samples)
• labels – optional labels