multiml.task.keras package

Subpackages

Submodules

Module contents

class multiml.task.keras.KerasBaseTask(run_eagerly=None, callbacks=['EarlyStopping', 'ModelCheckpoint'], save_tensorboard=False, **kwargs)

Bases: MLBaseTask

Base task for Keras model.

Examples

>>> # your keras model
>>> class MyKerasModel(Model):
>>>     def __init__(self, units=1):
>>>         super(MyKerasModel, self).__init__()
>>>
>>>         self.dense = Dense(units)
>>>         self.relu = ReLU()
>>>
>>>     def call(self, x):
>>>         return self.relu(self.dense(x))
>>>
>>> # create task instance
>>> task = KerasBaseTask(storegate=storegate,
>>>                      model=MyKerasModel,
>>>                      input_var_names=('x0', 'x1'),
>>>                      output_var_names='outputs-keras',
>>>                      true_var_names='labels',
>>>                      optimizer='adam',
>>>                      optimizer_args=dict(lr=0.1),
>>>                      loss='binary_crossentropy')
>>> task.set_hps({'num_epochs': 5})
>>> task.execute()
>>> task.finalize()
__init__(run_eagerly=None, callbacks=['EarlyStopping', 'ModelCheckpoint'], save_tensorboard=False, **kwargs)
Parameters:

  • run_eagerly (bool) – Run on eager execution mode (not graph mode).

  • callbacks (list(str or keras.Callback)) – callback for keras model training. Predefined callbacks (EarlyStopping, ModelCheckpoint, and TensorBoard) can be selected by str. Other user-defined callbacks should be given as keras.Callback object.

  • save_tensorboard (bool) – use tensorboard callback in training.

  • **kwargs – Arbitrary keyword arguments.

compile_model()

Compile keras model.

compile_loss()

Compile keras model.

load_model()

Load pre-trained keras model weights.

dump_model(extra_args=None)

Dump current keras model.

fit(train_data=None, valid_data=None)

Training model.

Returns:

training results.

Return type:

dict

predict(data=None, phase=None)

Evaluate model prediction.

Parameters:

phase (str) – data type (train, valid, test or None)

Returns:

prediction by the model ndarray: target

Return type:

ndarray

get_inputs()

Returns keras Input from input_var_names.

class multiml.task.keras.MLPTask(input_shapes=None, layers=None, activation=None, activation_last=None, kernel_regularizer=None, bias_regularizer=None, batch_norm=False, **kwargs)

Bases: KerasBaseTask

Keras MLP task.

__init__(input_shapes=None, layers=None, activation=None, activation_last=None, kernel_regularizer=None, bias_regularizer=None, batch_norm=False, **kwargs)
Parameters:

  • input_shapes (tuple) – shape for Keras.Inputs

  • layers (list) – list of hidden layers

  • activation (str) – activation function for MLP

  • activation_last (str) – activation function in last layer

  • kernel_regularizer (str) – kernel regularizer

  • bias_regularizer (str) – bias regularizer

  • batch_norm (bool) – use batch normalization

  • **kwargs – Arbitrary keyword arguments

set_hps(hps)

Set thresholds. Cut-names are given by get_hyperparameters methods.

Parameters:

hps (dict) – (hyperparameter name => hyperparameter value)

build_model()

Build a Keras MLP model.

get_inputs()

Returns keras Input from input_var_names.

class multiml.task.keras.Conv2DTask(conv2d_layers=None, **kwargs)

Bases: MLPTask

Keras MLP task.

__init__(conv2d_layers=None, **kwargs)
Parameters:

  • list (conv2d_layers) – list of conv2d layer config(op_name, op_args).

  • **kwargs – Arbitrary keyword arguments

build_model()

Build a Keras MLP model.

class multiml.task.keras.EnsembleTask(subtasks, dropout_rate=None, individual_loss=False, individual_loss_weights=0.0, **kwargs)

Bases: KerasBaseTask

__init__(subtasks, dropout_rate=None, individual_loss=False, individual_loss_weights=0.0, **kwargs)
Parameters:

  • subtasks (list) – list of task instances.

  • dropout_rate (float) – dropout_rate for ensemble weights. If None, no dropout.

  • individual_loss (bool) – use multiple outputs

  • individual_loss_weights (float) – coefficient for multiple outputs

  • **kwargs – Arbitrary keyword arguments

compile()

Compile model, optimizer and loss.

Compiled objects will be avaialble via self.ml.model, self.ml.optimizer and self.ml.loss.

Examples

>>> # compile all together,
>>> self.compile()
>>> # which is equivalent to:
>>> self.build_model() # set self._model
>>> self.compile_model() # set self.ml.model
>>> self.compile_optimizer() # set self.ml.optimizer
>>> self.compile_loss() # set self.ml.loss
compile_loss()

Compile keras model.

build_model()

Build model.

get_input_true_data(phase)

Get input and true data.

Parameters:

phase (str) – data type (train, valid, test or None).

Returns:

(input, true) data for model.

Return type:

tuple

get_inputs()

Returns keras Input from input_var_names.

get_submodel_names()

Returns subtask_id used in ensembling.

Returns:

list of subtask_id

Return type:

list (str)

get_submodel(i_models)

Get a submodel by model index.

Parameters:

i_models (int) – submodel index

Returns:

submodel for the input index

Return type:

subtasktuple

static get_ensemble_weights(model)

Collect ensemble_weights in the keras model.

Parameters:

model (keras.Model) –

Returns:

list of ensemble weights

Return type:

list (tf.Variable)

class multiml.task.keras.ModelConnectionTask(subtasks, loss_weights=None, variable_mapping=None, **kwargs)

Bases: ModelConnectionTask, KerasBaseTask

Keras implementation of ModelConnectionTask.

build_model()

Build model.

class multiml.task.keras.DARTSTask(optimizer_alpha, optimizer_weight, learning_rate_alpha, learning_rate_weight, zeta=0.01, **kwargs)

Bases: ModelConnectionTask

__init__(optimizer_alpha, optimizer_weight, learning_rate_alpha, learning_rate_weight, zeta=0.01, **kwargs)
Parameters:

  • optimizer_darts_alpha (str) – optimizer for alpha in DARTS optimization

  • optimizer_darts_weight (str) – optimizer for weight in DARTS optimization

  • learning_rate_darts_alpha (float) – learning rate (epsilon) for alpha in DARTS optimization

  • learning_rate_darts_weight (float) – learning rate (epsilon) for weight in DARTS optimization

  • zeta (float) – zeta parameter in DARTS optimization

  • **kwargs – Arbitrary keyword arguments

fit(train_data=None, valid_data=None)

Training model.

Returns:

training results.

Return type:

dict

load_metadata()

Load metadata.

get_best_submodels()

Returns indices of the best submodels determined by the alpha.

Returns:

list of index of the selected submodels

Return type:

list (int)

build_model()

Build model.

dump_model(extra_args=None)

Dump current DARTS model.