Compatibility with sklearn

[duboism]

Some properties of pylearn-parsimony Estimators are incompatible with sklearn Estimators. For instance in sklearn all parameters of the constructor have a default value (this allow to default construct the Estimator). There are many guidelines about the Estimator interface here (see chapters "APIs of scikit-learn objects" and "Rolling your own estimator"). Respecting this interface could help to mix parsimony and sklearn Estimators.

There are some tests in sklearn to ensure that all Estimators respect some assumptions of the API. One option to ensure the compatibility would be to use the same tests.

[tomlof]

Perhaps it's time to discuss this thoroughly? Being able to seamlessly cross over from scikit-learn to pylearn-parsimony and back would certainly be really useful!