Estimator#

class mergernet.estimators.base.Estimator[source]#

Bases: ABC

Attributes

`registry`
`tf_model`

abstract build(freeze_conv: bool = False)[source]#

compile_model(tf_model: Model, optimizer: Optimizer, metrics: list = [], label_smoothing: float = 0.0)[source]#

download(config: EstimatorConfig, replace: bool = False)[source]#

get_conv_arch(pretrained_arch: str) → Tuple[Callable, Callable][source]#

get_dataaug_block(flip_horizontal: bool = True, flip_vertical: bool = True, rotation: Tuple[float, float] | bool = (-0.08, 0.08), zoom: Tuple[float, float] | bool = (-0.15, 0.0))[source]#

get_metric(metric: str)[source]#

get_optimizer(optimizer: str, lr: float | LearningRateSchedule) → Optimizer[source]#

get_scheduler(scheduler: str, lr: float) → LearningRateSchedule[source]#

For cosine_restarts scheduler, the learning rate multiplier first decays from 1 to alpha for first_decay_steps steps. Then, a warm restart is performed. Each new warm restart runs for t_mul times more steps and with m_mul times initial learning rate as the new learning rate.

Parameters:

scheduler (str) – Scheduler name
lr (float) – Initial learning rate

Returns:

A LearningRateSchedule instance

Return type:

tf.keras.optimizers.schedules.LearningRateSchedule

plot(filename: str = 'model.png')[source]#

abstract predict()[source]#

set_trainable(tf_model: Model, layer: str, trainable: bool)[source]#

abstract train(run_name: str = '') → Tuple[Model, History][source]#

_abc_impl = <_abc_data object>#

registry = <mergernet.estimators.base.EstimatorRegistry object>#

property tf_model#