# Add a search function

A search function generates the pool of candidates that an {py:class}`Acquisition function <alf_core.optimizer.acquisition_function.AcquisitionFunction>` then scores.
Subclass {py:class}`BaseSearch <alf_core.optimizer.search.BaseSearch>` to control where candidates come from (e.g. enumerating a dataset with
`DatasetSearch`, or sampling from a generative model with `GeneratorSearch`).

**Base class:** {py:class}`BaseSearch <alf_core.optimizer.search.BaseSearch>`. **Key method:** `__call__`.

## Steps

1. Subclass {py:class}`BaseSearch <alf_core.optimizer.search.BaseSearch>` under `tools/alf_tools/optimizer/search/`.
2. Implement `__call__` to return the candidate pool for the current round.
3. Pair it with an acquisition function in an {py:class}`Optimizer <alf_core.optimizer.optimizer.Optimizer>` and run a {py:class}`Task <alf_core.tasks.base_task.BaseTask>`.
4. Add tests and run the build / pre-commit checks.

## Reference notebook

- [Search functions](https://github.com/instadeepai/alf/blob/main/tutorials/extending_base_classes/search_functions.ipynb): implement a custom {py:class}`BaseSearch <alf_core.optimizer.search.BaseSearch>`.