Change `max_num_samples` to `max_num_clients` in `tff.learning.build_personalization_eval`, and update the docstring. No change in the API or functionality.

PiperOrigin-RevId: 343883721

PiperOrigin-RevId: 343602626

Adds function for creating useful compositions of aggregators to perform zeroing (for robustness) and clipping or differential privacy.

PiperOrigin-RevId: 343510697

PiperOrigin-RevId: 343404565

Remove clipping_compositions.py. These were intended to be a "mid-level" API linking the low level aggregators with useful high-level compositions, but it is cleaner to skip this step.

PiperOrigin-RevId: 343392858

PiperOrigin-RevId: 343151873

PiperOrigin-RevId: 342863095

PiperOrigin-RevId: 342687652

Inferring TFF types from attr.s classes is deprecated as it has shown usability issues. This change upgrades the deprecation warning to an error and removes the functionality.

PiperOrigin-RevId: 342627042

PiperOrigin-RevId: 342128507

PiperOrigin-RevId: 342064680

New classes `WeightedAggregationFactory` and `NonWeightedAggregationFactory` replace the previous `AggregationProcessFactory` in the API

This change makes explicit difference between ways to create `tff.templates.AggreagationProcess` objects with input type signatures of its `next` function `<state, value>` and `<state, value, weight>`.

Also updates `ClippingFactory` and `ZeroingFactory` to implement both newly created classes.

`MeanFactory` now only implements `WeightedAggregationFactory` class.

PiperOrigin-RevId: 341988708

Previously, `check_returns_type` only displayed the stringified format
of mismatched types, which does not include container types. This CL
changes `check_returns_type` to reuse the type diff formatting from
computation_types which will fall back to printing the `repr` of the
types in this case.

PiperOrigin-RevId: 341896484

PiperOrigin-RevId: 341870083

This removes the context manager which temporarily replaced the reference `federated_secure_sum` with `federated_sum` while executors did not support this in tests. This has been addresses, thus no need for the workaround.

PiperOrigin-RevId: 341866125

This was added recent to signal the future deprecation of converting `attr.s` classes to TFF types. However it was added to the wrong block. The intent is to deprecate conversion from a `attr.s` class *type*, but not a class *instance*.

PiperOrigin-RevId: 341702845

Secure Sum extends the request bitwidth by log2(# of clients) participating in
the sum, and masks the integers coming out of the clients to the extended
bitwidth number of bits (the actual size of values transferred across the wire).

PiperOrigin-RevId: 341676681

PiperOrigin-RevId: 341641106

Adds functions for building useful compositions of aggregators. Also removes ZeroingClippingFactory because it was determined that composing Zeroing and Clipping is superior (since it allows independent estimators for the norms).

PiperOrigin-RevId: 341517003

PiperOrigin-RevId: 341461479

PiperOrigin-RevId: 341461167

PiperOrigin-RevId: 341441687

PiperOrigin-RevId: 341092818

tensors in additional to single tensors.

PiperOrigin-RevId: 341063667

Extend the implementation to allow taking operands that have matching
structures in addition to broadcast one tensor to a structure of tensors.

PiperOrigin-RevId: 340950882

PiperOrigin-RevId: 340865946

PiperOrigin-RevId: 340863997

PiperOrigin-RevId: 340749260

outside the range of tf.int32.

Additionally, raise an error if the Python integer is outisde the range of tf.int64.

This is more inline with `tf.convert_to_tensor` behavior, without being quite
as expensive.

PiperOrigin-RevId: 340745753

PiperOrigin-RevId: 340692208

Do not use or import the ABCs from `collections`, this is deprecated since Python 3.3, and in 3.9 it will stop working. These ABCs have moved to 'collections.abc' and should be used from this package.

PiperOrigin-RevId: 340678321

PiperOrigin-RevId: 340538528

TF-nightly  (2.5.0)

PiperOrigin-RevId: 340451437

PiperOrigin-RevId: 340276637

Fixes #914. The issue was caused by multiple conflicting imports of tensorboard, which are resolved by pip uninstalling tensorboard before bringing in tensorflow_federated.

PiperOrigin-RevId: 340242994

PiperOrigin-RevId: 339961597

For now measurements are aggregated by SumFactory, but we will add support in the future for private measurements aggregation.

Also:
  1. Change state to OrderedDict (from tuple).
  2. Change default zeroing norm order from np.inf to 2 for ease of transition between ZeroingFactory and ZeroingClippingFactory (which must use 2-norm), and use tf.linalg.global_norm if norm order is 2, for efficiency.
PiperOrigin-RevId: 339908232

PiperOrigin-RevId: 339805559

PiperOrigin-RevId: 339804460

PiperOrigin-RevId: 339801656