API Reference

dnn_locate.loc_model

loc_model

class dnn_locate.loc_model.loc_model(input_shape, discriminator, localizer_backend=None, tau_range=array([1, 2, 3, 4, 5, 6, 7, 8, 9]), target_r_square='auto', r_metric='acc', activation='tanh+relu', save_path=False)

Bases: object

class for discriminative feature detection for deep learning models.

Parameters:

input_shape{tuple-like} (shape of the feature/image)

For example, in MNIST dataset input_shape = (28, 28, 1).

localizer_backend: {keras-defined neural network}

A backend neural network before reshape and Truncated ReLU activation.

discriminator: {keras-defined neural network}

A pretrained neural network needs to be explained.

tau_range{list-like}

List of tau to define the localizers.

target_r_square: {float, [0,1]}

A target R_square to be explained.

R_square: {list-like}

Records for R_square values based on a dataset.

DA_plot(X, y, demo_ind=None, threshold=None, plt1_params={'cmap': 'binary'}, plt2_params={'cmap': 'OrRd'})

Plots data-adaptive detected region for the fitted localizer.

Parameters:

X{array-like} of shape (n_samples, dim_features)

Instances matrix/tensor, where n_samples in the number of samples and dim_features is the dimension of the features.

y{array-like} of shape (n_samples,)

Output vector/matrix relative to X.

demo_ind{array-like} of shape (num_instance, num_labels) default=None

threshold{array-like} or float, default=None

threshold to truncate the small detected pixels

R_sqaure_path()

Plot solution path for the proposed method wrt tau_range

R_square(X, y)

Report R_square for the fitted localizer based on a given dataset

Parameters:

X{array-like} of shape (n_samples, dim_features)

Instances matrix/tensor, where n_samples in the number of samples and dim_features is the dimension of the features.

y{array-like} of shape (n_samples,)

Output vector/matrix relative to X.

build_combined(optimizer=<keras.optimizers.optimizer_v2.adam.Adam object>)

Building a localizer and a combined model for the proposed framework

Parameters:

localizer: {keras-defined neural network}

A neural network for localizer.

optimizer: {keras-defined optimizer: ``tensorflow.keras.optimizers``}, default = ‘SGD(lr=.0005)’

A optimizer used to train the localizer.

build_localizer(tau=10.0, max_value=1.0)

Building a localizer for the proposed framework

Parameters:

tau: {float}

The magnitude of the localizer.

max_value: {float, (0, 1]}

The maximum proportion of the localizer.

fit(X_train, y_train, fit_params, datagen=None, demo_ind=None, X_test=None, y_test=None, optimizer=<keras.optimizers.optimizer_v2.gradient_descent.SGD object>)

Fitting the localizer based on a given dataset.

Parameters:

X_train{array-like} of shape (n_samples, dim_features)

Instances matrix/tensor, where n_samples in the number of samples and dim_features is the dimension of the features.

y_train{array-like} of shape (n_samples,)

Output vector/matrix relative to X.

X_test{array-like} of shape (n_samples, dim_features), default = None

Instances features to compute the r_square. If None, X_test = X_train

y_test{array-like} of shape (n_samples,), default = None

Output to compute the r_square. If None, y_test = y_train

localizer{keras-defined neural network}

A neural network for localizer before Truncated RELU activation.

fit_params: {dict of fitting parameters}

See keras fit: (https://keras.rstudio.com/reference/fit.html), including batch_size, epoch, callbacks,

validation_split, validation_data, and so on.

demo_ind{array-like} default=None

The index set for demonstrated instances.

optimizer: {keras-defined optimizer: ``tensorflow.keras.optimizers``}, default = ‘SGD(lr=.0005)’

A optimizer used to train the localizer.

locate(X)

Return the localized discriminative features by the fitted localizer.

Parameters:

X{array-like} of shape (n_samples, dim_features)

Instances matrix/tensor, where n_samples in the number of samples and dim_features is the dimension of the features.

path_plot(threshold=None, plt1_params={'cmap': 'binary'}, plt2_params={'cmap': 'OrRd'})

Plots generalized partial R values and its corresponding images wrt tau_range.

Parameters:

threshold{array-like} or float

threshold to truncate the small detected pixels

plt1_params{dict-like}

dict for imshow for X_demo

plt2_params{dict-like}

dict for imshow for X_diff_demo