# tff.learning.templates.compose_learning_process

<br />

|---|
| [![](https://www.tensorflow.org/images/GitHub-Mark-32px.png) View source on GitHub](https://github.com/tensorflow/federated/blob/v0.87.0 Version 2.0, January 2004 Licensed under the Apache License, Version 2.0 (the) |

Composes specialized measured processes into a learning process.

    tff.learning.templates.compose_learning_process(
        initial_model_weights_fn: https://www.tensorflow.org/federated/api_docs/python/tff/Computation,
        model_weights_distributor: https://www.tensorflow.org/federated/api_docs/python/tff/learning/templates/DistributionProcess,
        client_work: https://www.tensorflow.org/federated/api_docs/python/tff/learning/templates/ClientWorkProcess,
        model_update_aggregator: https://www.tensorflow.org/federated/api_docs/python/tff/templates/AggregationProcess,
        model_finalizer: https://www.tensorflow.org/federated/api_docs/python/tff/learning/templates/FinalizerProcess
    ) -> https://www.tensorflow.org/federated/api_docs/python/tff/learning/templates/LearningProcess

### Used in the notebooks

| Used in the tutorials |
|---|
| - [Composing Learning Algorithms](https://www.tensorflow.org/federated/tutorials/composing_learning_algorithms) |

Given 4 specialized measured processes (described below) that make a learning
process, and a computation that returns initial model weights to be used for
training, this method validates that the processes fit together, and returns a
`LearningProcess`. Please see the tutorial at
<https://www.tensorflow.org/federated/tutorials/composing_learning_algorithms>
for more details on composing learning processes.

The main purpose of the 4 measured processes are:

- `model_weights_distributor`: Make global model weights at server available as the starting point for learning work to be done at clients.
- `client_work`: Produce an update to the model received at clients.
- `model_update_aggregator`: Aggregates the model updates from clients to the server.
- `model_finalizer`: Updates the global model weights using the aggregated model update at server.

The `next` computation of the created learning process is composed from the
`next` computations of the 4 measured processes, in order as visualized below.
The type signatures of the processes must be such that this chaining is
possible. Each process also reports its own metrics.

    ┌─────────────────────────┐
    │model_weights_distributor│
    └△─┬─┬────────────────────┘
     │ │┌▽──────────┐
     │ ││client_work│
     │ │└┬─────┬────┘
     │┌▽─▽────┐│
     ││metrics││
     │└△─△────┘│
     │ │┌┴─────▽────────────────┐
     │ ││model_update_aggregator│
     │ │└┬──────────────────────┘
    ┌┴─┴─▽──────────┐
    │model_finalizer│
    └┬──────────────┘
    ┌▽─────┐
    │result│
    └──────┘

The `get_hparams` computation of the created learning process produces a
nested ordered dictionary containing the result of `client_work.get_hparams`
and `finalizer.get_hparams`. The `set_hparams` computation operates similarly,
by delegating to `client_work.set_hparams` and `finalizer.set_hparams` to set
the hyperparameters in their associated states.

<br />

<br />

<br />

| ## Args ||
|---|---|
| `initial_model_weights_fn` | A [`tff.Computation`](https://www.tensorflow.org/federated/api_docs/python/tff/Computation) that returns (unplaced) initial model weights. |
| `model_weights_distributor` | A [`tff.learning.templates.DistributionProcess`](https://www.tensorflow.org/federated/api_docs/python/tff/learning/templates/DistributionProcess). |
| `client_work` | A [`tff.learning.templates.ClientWorkProcess`](https://www.tensorflow.org/federated/api_docs/python/tff/learning/templates/ClientWorkProcess). |
| `model_update_aggregator` | A [`tff.templates.AggregationProcess`](https://www.tensorflow.org/federated/api_docs/python/tff/templates/AggregationProcess). |
| `model_finalizer` | A [`tff.learning.templates.FinalizerProcess`](https://www.tensorflow.org/federated/api_docs/python/tff/learning/templates/FinalizerProcess). |

<br />

<br />

<br />

<br />

| ## Returns ||
|---|---|
| A [`tff.learning.templates.LearningProcess`](https://www.tensorflow.org/federated/api_docs/python/tff/learning/templates/LearningProcess). ||

<br />

<br />

<br />

<br />

| ## Raises ||
|---|---|
| `ClientSequenceTypeError` | If the first arg of the `next` method of the resulting `LearningProcess` is not a structure of sequences placed at [`tff.CLIENTS`](https://www.tensorflow.org/federated/api_docs/python/tff#CLIENTS). |

<br />