Sklearn Compatibility
=====================

psipose estimators follow sklearn conventions strictly.

Rules
------

1. ``__init__`` only stores parameters (no validation, no object construction)
2. Use ``None`` as default for mutable objects (encoders, ansatze)
3. Resolve defaults in ``fit()``, not ``__init__()``
4. Fitted attributes end with ``_`` (e.g., ``weights_``, ``classes_``)
5. ``fit()`` returns ``self``
6. Use ``sklearn.utils.validation.check_X_y`` for input validation

Example
--------

.. code-block:: python

   def __init__(self, encoder=None, ansatz=None):
       self.encoder = encoder  # Store None, resolve in fit()
       self.ansatz = ansatz

   def fit(self, X, y):
       encoder = self.encoder if self.encoder is not None else AngleFeatureMap()
       # ...
       return self