tff.learning.build_local_evaluation

Stay organized with collections Save and categorize content based on your preferences.

Builds the local TFF computation for evaluation of the given model.

tff.learning.build_local_evaluation(
    model_fn: Callable[[], tff.learning.Model],
    model_weights_type: tff.types.StructType,
    batch_type: tff.types.Type,
    use_experimental_simulation_loop: bool = False
) -> tff.Computation

This produces an unplaced function that evaluates a tff.learning.models.VariableModel on a tf.data.Dataset. This function can be mapped to placed data, i.e. is mapped to client placed data in build_federated_evaluation.

The TFF type notation for the returned computation is:

(<M, D*> → <local_outputs=N, num_examples=tf.int64>)

Where M is the model weights type structure, D is the type structure of a single data point, and N is the type structure of the local metrics.

model_fn A no-arg function that returns a tff.learning.models.VariableModel.
model_weights_type The tff.Type of the model parameters that will be used to initialize the model during evaluation.
batch_type The type of one entry in the dataset.
use_experimental_simulation_loop Controls the reduce loop function for input dataset. An experimental reduce loop is used for simulation.

A federated computation (an instance of tff.Computation) that accepts model parameters and sequential data, and returns the evaluation metrics.