base

base#

Classes#

BaseModel

BaseModel for Machine Learning Models

ModelInstanceDescription

Provide a description of a model instance to allow serialization.

Module Contents#

class base.BaseModel(features: list[str], targets: list[str] | None = None, name: str | None = None, distribution_strategy=None, output_dir: pathlib.Path = Path(gettempdir()))#

Bases: abc.ABC

BaseModel for Machine Learning Models

Parameters:

  • name – A name for this model - by default this will be the class name

  • features – List of the input feature names

  • targets – List of the target feature names

  • distribution_strategy – A tensorflow distribution strategy when trying to distribute the training across multiple devices

  • output_dir – model / experiment specific output directory

input_specs: ClassVar[OrderedDict[str, damast.core.DataSpecification]]#
features: list[str]#
output_specs: ClassVar[OrderedDict[str, damast.core.DataSpecification]]#
targets: list[str]#
name: str#
model: keras.Model#
distribution_strategy: object | None#
model_dir: pathlib.Path#
history: keras.callbacks.History#
abstract _init_model()#

Initialize the internal keras model, i.e. self.model.

load_weights(checkpoint_filepath: str | pathlib.Path)#

Load the model weight from an existing checkpoint named by the checkpoint_filepath

Parameters:

checkpoint_filepath – Name of the checkpoint filepath (prefix)

build(optimizer: str | keras.optimizers.Optimizer = keras.optimizers.Adam(0.001), loss_function: str | keras.losses.Loss = 'mse', metrics: list[str] | list[keras.metrics.Metric] = ['mse'], **kwargs)#

Build / Compile the actual model

Parameters:

  • optimizer – Name of the optimizer or an instance of keras.optimizers.Optimizer

  • loss_function – Name of the loss function or an instance of keras.losses.Loss

  • metrics – List of names of the metrics that shall be used, or list of instance of keras.metrics.Metrics

  • kwargs – additional arguments that can be forwarded to keras.engine.training.Model.compile

plot(suffix='.png') → pathlib.Path#

Plot the current model instance.

Parameters:

suffix – Suffix that shall be used for the plot file

Returns:

Path to the plotted file

property checkpoints_dir: pathlib.Path#

Directory for the checkpoints.

Returns:

Path to the checkpoints directory

property evaluation_file: pathlib.Path#

Path to the evaluation file storing the results

Returns:

evaluation file

train(training_data: tf.data.Dataset, validation_data: tf.data.Dataset, monitor: str = 'val_loss', mode: str = 'min', epochs: int = 1, initial_epoch: int = 0, save_history: bool = True, **kwargs) → None#
get_evaluations() → damast.core.types.DataFrame#

Read the evaluation results from the evaluation file.

Raises:

FileNotFoundError – If evaluation result file does not exist

save() → pathlib.Path#

Save the model in the model directory.

Returns:

Path to the saved model.

evaluate(label: str, evaluation_data: tf.data.Dataset, **kwargs) → dict[str, any]#

Evaluate this model.

Parameters:

  • label – Custom label for this evaluation, e.g., can be the name of the dataset being used

  • evaluation_data – The data that shall be used for evaluation

Returns:

DataFrame listing all performed evaluation results for this model

class base.ModelInstanceDescription#

Bases: NamedTuple

Provide a description of a model instance to allow serialization.

model: BaseModel#
parameters: dict[str, str]#
classmethod from_dict(data: dict[str, any]) → ModelInstanceDescription#
__iter__()#
Contents