Changed

• The Retrieval task now accepts a list of factorized metrics instead of a
  single optional metric.

Added

• tfrs.experimental.optimizers.ClippyAdagrad: a new optimizer based on
  tf.keras.optimizers.Adagrad that is able to improve training stability.
• tfrs.metrics.FactorizedTopK now accepts sample weights which are used to
  compute weighted top k metrics.

• Improved support for using TPUEmbedding under parameter server strategy.

Release fixes.

[0.7.0][2022-07-07]

A number of changes to make factorized top-K metric computation more accurate
and less prone to user error.

Changed

• tfrs.layers.embedding.TPUEmbedding now supports input features with
  dynamic shape. batch_size argument is deprecated and no longer required.

• tfrs.layers.embedding.TPUEmbedding now supports running on different
  versions of TPU.

• Pinned TensorFlow to >= 2.9.0 which works with Scann 1.2.7.

• tfrs.tasks.Ranking.call now accepts a compute_batch_metrics argument to
  allow switching off batch metric computation. Following this change,
  'compute_metrics'argument does not impact computation of batch metrics.

Breaking changes

• tfrs.metrics.FactorizedTopK requires the candidate ids for positive
  candidates to be supplied when using approximate top-K sources. Each top-K
  layer now has an exact method to broadcast its ability to return exact or
  approximate top-K results.
• Removed metrics constructor parameter for tfrs.metrics.FactorizedTopK.
  FactorizedTopK only makes sense with top-k metrics, and this change
  enforces this.
• Replaced the k constructor argument in tfrs.metrics.FactorizedTopK with
  ks: a list of k values at which to compute the top k metric.

Changed

• The tfrs.metrics.FactorizedTopK metric can now compute candidate-id based
  metrics when given the true_candidate_ids argument in its call method.

Added

• The Retrieval task now also accepts a loss_metrics argument.

Changed

• Pinned TensorFlow to >= 2.6.0, which works with Scann 1.2.3.

Breaking changes

• TopK layer indexing API changed. Indexing with datasets is now done via
  the index_from_dataset method. This change reduces the possibility of
  misaligning embeddings and candidate identifiers when indexing via
  indeterministic datasets.

Fixed

• Fixed error in default arguments to tfrs.experimental.models.Ranking
  (#311).
• Fix TPUEmbedding layer to use named parameters.

Added

• Added batch_metrics to tfrs.tasks.Retrieval for measuring how good the
  model is at picking out the true candidate for a query from other candidates
  in the batch.
• Added tfrs.experimental.layers.embedding.PartialTPUEmbedding layer, which
  uses tfrs.layers.embedding.TPUEmbedding for large embedding lookups and
  tf.keras.layers.Embedding for smaller embedding lookups.

Changed

• Supplying incompatibly-shaped candidates and identifiers inputs to
  factorized_top_k layers will now raise (to prevent issues similar to
  #286).

Added

• tfrs.experimental.models.Ranking, an experimental pre-built model for
  ranking tasks. Can be used as DLRM like model with Dot Product feature
  interaction or DCN like model with Cross layer.

Changed

• Fixed the bug in tfrs.layers.loss.SamplingProbablityCorrection that logits
  should subtract the log of item probability.
• tfrs.experimental.optimizers.CompositeOptimizer: an optimizer that
  composes multiple individual optimizers which can be applied to different
  subsets of the model's variables.
• tfrs.layers.dcn.Cross and DotInteraction layers have been moved to
  tfrs.layers.feature_interaction package.

Added

• TopK layers now come with a query_with_exclusions method, allowing
  certain candidates to be excluded from top-k retrieval.
• TPUEmbedding Keras layer for accelerating embedding lookups for large
  tables with TPU.

Changed

• factorized_top_k.Streaming layer now accepts a query model, like other
  factorized_top_k layers.

• Updated ScaNN to 1.2.0, which requires TensorFlow 2.4.x. When not using
  ScaNN, any TF >= 2.3 is still supported.

Changed

• Pinned TensorFlow to >= 2.3 when ScaNN is not being installed. When ScaNN is
  being installed, we pin on >= 2.3, < 2.4. This allows users to use TFRS on
  TF 2.4 when they are not using ScaNN.