API pairwise_distances will require explicit V/VI param if Y is given

[jnothman]

Deprecation until version 0.25.

The current approach in _precompute_metric_params
(

scikit-learn/sklearn/metrics/pairwise.py

Lines 1429 to 1444 in f82a2cb

	def _precompute_metric_params(X, Y, metric=None, **kwds):
	"""Precompute data-derived metric parameters if not provided
	"""
	if metric == "seuclidean" and 'V' not in kwds:
	if X is Y:
	V = np.var(X, axis=0, ddof=1)
	else:
	V = np.var(np.vstack([X, Y]), axis=0, ddof=1)
	return {'V': V}
	if metric == "mahalanobis" and 'VI' not in kwds:
	if X is Y:
	VI = np.linalg.inv(np.cov(X.T)).T
	else:
	VI = np.linalg.inv(np.cov(np.vstack([X, Y]).T)).T
	return {'VI': VI}
	return {}

)
means that we may be applying a different metric at training and test
time. Ideally we'd have a framework for fitting a metric on some
specific training data, but in the meantime, this deprecation stops
users making mistakes.

[adrinjalali]

Are there other metrics where we have a similar pattern?

[jnothman]

Only Gower in progress.

[jnothman]

Good to go, @thomasjpfan?

[adrinjalali]

tagging for inclusion #17010