PiperOrigin-RevId: 321785542

PiperOrigin-RevId: 313410789

Replace `import tensorflow.compat.v2 as tf` with `import tensorflow as tf` in the `learning` module.

PiperOrigin-RevId: 310572253

PiperOrigin-RevId: 310377468

* Replaced `# Unsued` with `# Unused.`
* Replaced `# unused` with `# Unused.`

PiperOrigin-RevId: 301667453

This has been replaced with passing a `tff.learning.Model` and two optimizers (client and server)
the federated optimization processes.

PiperOrigin-RevId: 301266327

Step 1 of 3.

PiperOrigin-RevId: 297684545

PiperOrigin-RevId: 287860207

PiperOrigin-RevId: 287599572

PiperOrigin-RevId: 285801872

PiperOrigin-RevId: 285116495

de-duplication strategy in non-eager mode.

TF, under certain circumstances, prepends `_#` numbers to variables with the
same name. This appears to happen in non-eager mode and in some occasions with
Keras. TFF previously used these names for the type signature, which would
cause problems when the same model was created more than once in the same graph
context.

Instead, rely on the ordering of the sequence of variables contained in the
model and use the unnamed structure for types.

PiperOrigin-RevId: 285032395

PiperOrigin-RevId: 280784907

PiperOrigin-RevId: 280767713

Previously tried calling shape on a nested structure of tensors, which will not
work. Instead flatten the structre and call shape on the first tensor (all
input tensors are expected to have the same first dimension size, the batch
size).

Updated example models and tests, ensured the test failed under the current
code and passes after this change.

Fixes #785

PiperOrigin-RevId: 280221692

The unittest adding the example to FedAvg tests currently does not run because
TFF doesn't have its hands on the lookup table initializer (see
#783). This test is disabled by
default (since it won't pass) and any fix should enable and ensure it passes.

Switch to the local executor for tests to make them faster, and decrease the
size to 'medium' to reflect this improvement.

PiperOrigin-RevId: 280083440

This is a prerequisite for making TensorShape and Dimension more strict
about the types of their arguments.

PiperOrigin-RevId: 274176324

PiperOrigin-RevId: 264416280

PiperOrigin-RevId: 263172828

`tf.compat.v1.train.GradientDescentOptimizer`.

PiperOrigin-RevId: 253804687

Convert usages of `tf.assign(variable, tesnor)` to `variable.assign(tensor)` to be TF2.0 compatible.

PiperOrigin-RevId: 253102645

Without this, input lists would be converted to single tensors which prevented
the use of Keras models that support multiple inputs.

PiperOrigin-RevId: 248348837

Now that TFF has moved to a more recent version of TensorFlow, this no longer needs to be disabled.

- Remove autograph=False from tf.function decorators
- Consistently use tf.function everywhere
- Replace tf.cond() and tf.control_dependencies() with Python control flow

PiperOrigin-RevId: 245817277

PiperOrigin-RevId: 244269098

This is a workaround for issue #258, which uncovered that tf.keras.Model.weights and tf.keras.Model.get_weights() are not ordered the same.

- Add a toy example model that uses batch norm (includes non-trainable variables), which will fail without this change.
- Move client optimizer variables to local_variables, this includes variables such as iteration number.

PiperOrigin-RevId: 241075678

This will better align naming with TF and numpy, and reduce confusion with the Federated Averaging optimiztion algorithms.

PiperOrigin-RevId: 239638931

PiperOrigin-RevId: 233541246

PiperOrigin-RevId: 233472320

PiperOrigin-RevId: 233095312

PiperOrigin-RevId: 233063673

Implement metrics federated aggreation and return as a side output in federated learning optimizations.

PiperOrigin-RevId: 232972496

Implement report_local_outputs() and federated_output_computation() for tff.Model wrapped Keras models.

PiperOrigin-RevId: 232755857

- Add pydoc to BatchOutput
- Change tff.learning.Model to expose a new `input_spec`
- Extend model_utils keras methods to require a dummy_batch argument (necessary for subclassed tf.keras.models.Model)
- Update `build_model_delta_optimizer_process` to use tff.learning.Model.input_spec, isntead of hard coded type

PiperOrigin-RevId: 231294914

- Add the "federated orchestration" logic to tie together local client training with global averaging and global model updates.
- Rename build_model_delta_optimizer_tff -> build_model_delta_optimizer_process
- Add a new test utility class that matches any type.

PiperOrigin-RevId: 230965510

PiperOrigin-RevId: 230591634

Sketch an approach to handling model-controlled metric aggregation via the Model.federated_output() method.

Also renames Model.aggregated_outputs to Model.report_local_outputs.

PiperOrigin-RevId: 228270629

* Added missing BUILD dependencies.
* Added six python dependencies.
* Fixed whitespace.

PiperOrigin-RevId: 227725720

Adds initial TFF orchestration for optimizers based on computing model deltas (FedAvg and FedSgd). Also moves common server optimizer functionality to framework/optimizer_utils.py, and adds tests. Moves @graph_mode_test to test_utils.py

PiperOrigin-RevId: 227157965

Implements the client logic for federated_sgd, and adds a common interface for client optimizers that average model deltas (FedSgd and FedAvg).  Common functionality for building federated optimizers moves to framework/optimizer_utils.py.

PiperOrigin-RevId: 227026404