deep_bottleneck package

Subpackages

Submodules

deep_bottleneck.artifact_viewer module

class deep_bottleneck.artifact_viewer.Artifact(name, file)[source]

Bases: object

Displays or saves an artifact.

content
extension = ''
save()[source]
class deep_bottleneck.artifact_viewer.ArtifactLoader(mongo_uri='mongodb://<MONGO_INITDB_ROOT_USERNAME>:<MONGO_INITDB_ROOT_PASSWORD>@<server_ip_address>:27017/?authMechanism=SCRAM-SHA-1', db_name='<MONGO_DATABASE>')[source]

Bases: object

Loads artifacts related to experiments.

load[source]
class deep_bottleneck.artifact_viewer.CSVArtifact(name, file)[source]

Bases: deep_bottleneck.artifact_viewer.Artifact

Displays and saves a CSV artifact

extension = 'csv'
show()[source]
class deep_bottleneck.artifact_viewer.MP4Artifact(name, file)[source]

Bases: deep_bottleneck.artifact_viewer.Artifact

Displays or saves a MP4 artifact

extension = 'mp4'
show()[source]
class deep_bottleneck.artifact_viewer.PNGArtifact(name, file)[source]

Bases: deep_bottleneck.artifact_viewer.Artifact

Displays or saves a PNG artifact.

extension = 'png'
show(figsize=(10, 10))[source]

deep_bottleneck.run_experiments module

This script can be used to submit several experiments to the grid. All the experiments need to be specified as separate JSON.

deep_bottleneck.run_experiments.create_output_directory()[source]
deep_bottleneck.run_experiments.main()[source]
deep_bottleneck.run_experiments.parse_command_line_args()[source]
deep_bottleneck.run_experiments.start_experiment(root, file, local_execution)[source]
deep_bottleneck.run_experiments.start_experiments(config_dir_or_file, local_execution)[source]

Recursively walk through the config dir and submit all experiment configurations in there to the grid.

deep_bottleneck.utils module

deep_bottleneck.utils.construct_full_dataset(training, test)[source]

Concatenates training and test data splits to obtain the full dataset.

The input arguments use the following naming convention:
  • X is the training data
  • y is training class, with numbers from 0 to 1
  • Y is training class, but coded as a 2-dim vector with one entry set to 1 at the column index corresponding to the class
Parameters:
  • training – Namedtuple with fields X, y and Y:
  • test – Namedtuple with fields X, y and Y:
Returns:

A new Namedtuple with fields X, y and Y containing the concatenation of training and test data
deep_bottleneck.utils.data_shuffle(data_sets_org, percent_of_train, min_test_data=80, shuffle_data=False)[source]

Divided the data to train and test and shuffle it

deep_bottleneck.utils.get_min_max(activations_summary, layer_number, neuron_number=None)[source]

Get minimum and maximum of activations of a specific layer or a specific neuron over all epochs :param activations_summary: numpy ndarray :param layer_number: Index of the layer :param neuron_number: Index of the neuron. If None, activations of the whole layer serve as a basis

Returns:Minimum and maximum value of activations over all epochs
deep_bottleneck.utils.get_mongo_config()[source]
deep_bottleneck.utils.is_dense_like(layer)[source]

Check whether a layer has attribute ‘kernel’, which is true for dense-like layers :param layer: Keras layer to check for attribute ‘kernel’

Returns:True if layer has attribute ‘kernel’, False otherwise
deep_bottleneck.utils.shuffle_in_unison_inplace(a, b)[source]

Shuffles both array a and b randomly in unison :param a: An Array, for example containing data samples :param b: An Array, fpor example containing labels

Returns:Both arrays shuffled in the same way

Module contents