scalarstop.model_template

A class that builds models with hyperparameters.

A ModelTemplate is a description of how to create a compiled machine learning model and the hyperparameters that the model depends on.

Subclass ModelTemplate to describe the architecture of your model and the hyperparameters that are used to construct the model (and the model’s optimizer).

Then, pass an instance of your ModelTemplate subclass to as subclass of Model to train an instance of a machine learning model created from your ModelTemplate.

>>> import tensorflow as tf
>>> import scalarstop as sp
>>>
>>> class small_dense_10_way_classifier_v1(sp.ModelTemplate):
...    @sp.dataclass
...
...    class Hyperparams(sp.HyperparamsType):
...        hidden_units: int
...        optimizer: str = "adam"
...
...    def new_model(self):
...        model = tf.keras.Sequential(
...            layers=[
...                tf.keras.layers.Flatten(input_shape=(28, 28)),
...                tf.keras.layers.Dense(
...                    units=self.hyperparams.hidden_units,
...                    activation="relu",
...                ),
...                tf.keras.layers.Dense(units=10)
...           ],
...            name=self.name,
...        )
...        model.compile(
...            optimizer=self.hyperparams.optimizer,
...            loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),
...            metrics=["accuracy"],
...        )
...        return model

>>> model_template = small_dense_10_way_classifier_v1(hyperparams=dict(hidden_units=20))

>>> model_template.name
'small_dense_10_way_classifier_v1-zc9r3do1baeeffafanjnjmou'

Module Contents

Classes

ModelTemplate

Describes machine learning model architectures and hyperparameters. Used to generate new machine learning model objects that are passed into Model objects.
class ModelTemplate(*, hyperparams: Optional[Union[Mapping[str, Any], scalarstop.hyperparams.HyperparamsType]] = None, **kwargs)

Bases: scalarstop._single_namespace.SingleNamespace

Describes machine learning model architectures and hyperparameters. Used to generate new machine learning model objects that are passed into Model objects.

Parameters

hyperparams – The hyperparameters to initialize this class with.

Hyperparams :Type[scalarstop.hyperparams.HyperparamsType]
new_model(self) → Any

Create a new compiled model with the current hyperparameters.

When you override this method, make sure to create a new model object every single time this function is called.

classmethod calculate_name(cls, *, hyperparams: Optional[Union[Mapping[str, Any], scalarstop.hyperparams.HyperparamsType]] = None) → str

Calculate the hashed name of this object, given the hyperparameters.

This classmethod can be used to calculate what an object would be without actually having to call __init__().

property group_name(self) → str

The “group” name is this object’s Python class name.

property name(self) → str

The group (class) name and a calculated hash of the hyperparameters.

property hyperparams(self) → scalarstop.hyperparams.HyperparamsType

Returns a HyperparamsType instance containing hyperparameters.

property hyperparams_flat(self) → Dict[str, Any]

Returns a Python dictionary of “flattened” hyperparameters.

AppendDataBlob objects modify a “parent” DataBlob, nesting the parent’s Hyperparams within the AppendDataBlob ‘s own Hyperparams.

This makes it hard to look up a given hyperparams key. A value at parent_datablob.hyperparams.a is stored at child_datablob.hyperparams.parent.hyperparams.a.

This hyperparams_flat property provides all nested hyperparams keys as a flat Python dictionary. If a child AppendDataBlob has a hyperparameter key that that conflicts with the parent, the child’s value will overwrite the parent’s value.