I think you are right. Moreover, a transformer already exists here
<https://github.com/scikit-learn/scikit-learn/blob/85ec0fd1ae904f275f608b11044a2476ed4723e6/sklearn/feature_extraction/image.py#L380>.
I see two solutions:
* The transform() method could return the coordinates and the
inverse_transform() method could take them as argument. The
Transformer API
<http://scikit-learn.org/0.13/developers/index.html#different-objects>
doesn't specify if the transform() method can take more than one
argument (two, to be exact) and I didn't find inverse_transform() in
the API page. So I don't know if this approach would be consistent
with all the others estimators/transformers.
* The transformer could also retain the coordinates but then:
o It would be specific to images of a certain size.
o What kind of behaviour would be expected if the transform()
method is called more than once? Use the same coordinates to
extract patches in the new set of images?
o If the transform() method was used to extract patches from
various images, these images would need to be reconstructed
together.
Another question for later:
* The two functions extract_patches_2d and reconstruct_from_patches_2d
are part of the public API (they are not prefixed with _). I think
they sould be deprecated if a create a more complex transformer.
What do you think?
On 04/22/2013 12:38 AM, Gael Varoquaux wrote:
Would a transformer, with an associated inverse_transform, be useful
here? It seems to me that it would be the right pattern, however I don't
have the code in mind, so I may be wrong.
------------------------------------------------------------------------------
Precog is a next-generation analytics platform capable of advanced
analytics on semi-structured data. The platform includes APIs for building
apps and a phenomenal toolset for data science. Developers can use
our toolset for easy data analysis & visualization. Get a free account!
http://www2.precog.com/precogplatform/slashdotnewsletter
_______________________________________________
Scikit-learn-general mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
------------------------------------------------------------------------------
Precog is a next-generation analytics platform capable of advanced
analytics on semi-structured data. The platform includes APIs for building
apps and a phenomenal toolset for data science. Developers can use
our toolset for easy data analysis & visualization. Get a free account!
http://www2.precog.com/precogplatform/slashdotnewsletter
_______________________________________________
Scikit-learn-general mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
