Registration is open for TensorFlow Dev Summit 2020 Learn more


View source on GitHub

Class EnhancedModel

A wrapper around a Model that adds sanity checking and metadata helpers.

Inherits From: Model


View source


Initialize self. See help(type(self)) for accurate signature.



Performs federated aggregation of the Model's local_outputs.

This is typically used to aggregate metrics across many clients, e.g. the body of the computation might be:

return {
    'num_examples': tff.federated_sum(local_outputs.num_examples),
    'loss': tff.federated_mean(local_outputs.loss)

N.B. It is assumed all TensorFlow computation happens in the report_local_outputs method, and this method only uses TFF constructs to specify aggregations across clients.


Either a tff.Computation, or None if no federated aggregation is needed.

The tff.Computation should take as its single input a tff.CLIENTS-placed tff.Value corresponding to the return value of Model.report_local_outputs, and return a dictionary or other structure of tff.SERVER-placed values; consumers of this method should generally provide these server-placed values as outputs of the overall computation consuming the model.


The type specification of the batch_input parameter for forward_pass.

A nested structure of tf.TensorSpec objects, that matches the structure of arguments that will be passed as the batch_input argument of forward_pass. The tensors must include a batch dimension as the first dimension, but the batch dimension may be undefined.

Similar in spirit to tf.keras.models.Model.input_spec.


An iterable of tf.Variable objects, see class comment for details.


An iterable of tf.Variable objects, see class comment for details.


An iterable of tf.Variable objects, see class comment for details.


Returns a tff.learning.ModelWeights.



View source


Runs the forward pass and returns results.

This method should not modify any variables that are part of the model, that is, variables that influence the predictions; for that, see TrainableModel.train_on_batch.

However, this method may update aggregated metrics computed across calls to forward_pass; the final values of such metrics can be accessed via aggregated_outputs.

Uses in TFF:

  • To implement model evaluation.
  • To implement federated gradient descent and other non-Federated-Averaging algorithms, where we want the model to run the forward pass and update metrics, but there is no optimizer (we might only compute gradients on the returned loss).
  • To implement Federated Averaging, when augmented as a TrainableModel.



A BatchOutput object. The object must include the loss tensor if the model will be trained via a gradient-based algorithm.


View source


Returns tensors representing values aggregated over forward_pass calls.

In federated learning, the values returned by this method will typically be further aggregated across clients and made available on the server.

This method returns results from aggregating across all previous calls to forward_pass, most typically metrics like accuracy and loss. If needed, we may add a clear_aggregated_outputs method, which would likely just run the initializers on the local_variables.

In general, the tensors returned can be an arbitrary function of all the tf.Variables of this model, not just the local_variables; for example, this could return tensors measuring the total L2 norm of the model (which might have been updated by training).

This method may return arbitrarily shaped tensors, not just scalar metrics. For example, it could return the average feature vector or a count of how many times each feature exceed a certain magnitude.


A structure of tensors (as supported by tf.nest) to be aggregated across clients.