Mercurial > pylearn

comparison doc/v2_planning/optimization.txt @ 1149:7c5dc11c850a

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
cleaning up api_optimization
author James Bergstra <bergstrj@iro.umontreal.ca>
date Thu, 16 Sep 2010 16:26:21 -0400
parents 2cf3ad953bf9
children f2105a06201c
comparison
equal deleted inserted replaced
1148:2da593b0f29d 1149:7c5dc11c850a
49 Proposal for API 49 Proposal for API
50 ================ 50 ================
51 51
52 See api_optimization.txt. 52 See api_optimization.txt.
53 53
54 OD: Do we really need a different file? If yes, maybe create a subdirectory to 54 OD asks: Do we really need a different file? If yes, maybe create a subdirectory
55 be able to easily find all files related to optimization? 55 to be able to easily find all files related to optimization?
56
57 JB replies: Yoshua's orders.
58
59
60
61 OD asks: Could it be more convenient for x0 to be a list?
62
63 JB replies: Yes, but that's not the interface used by other minimize()
64 routines (e.g. in scipy). Maybe another list-based interface is required?
65
66 OD replies: I think most people would prefer to use a list-based interface, so
67 they don't have to manually pack / unpack multiple arrrays of parameters. So I
68 would vote in favor or having both (where the main reason to also provide a
69 non-list interface would be to allow one to easily switch e.g. to scipy's
70 minimize).
71 I would guess the reason scipy's interface is like this is because it makes
72 it easier for the optimization algorithm. However, this does not really
73 matter if we are just wrapping a theano-based algorithm (that already has
74 to handle multiple parameters), and avoiding useless data copies on each call
75 to f / df can only help speed-wise.
76 JB replies: Done, I added possibility that x0 is list of ndarrays to the api
77 doc.
78
79
80
81 OD asks: Why make a difference between iterative and one-shot versions? A one-shot
82 algorithm can be seen as an iterative one that stops after its first
83 iteration. The difference I see between the two interfaces proposed here
84 is mostly that one relies on Theano while the other one does not, but
85 hopefully a non-Theano one can be created by simply wrapping around the
86 Theano one.
87
88 JB replies: Right, it would make more sense to distinguish them by the fact that
89 one works on Theano objects, and the other on general Python callable functions.
90 There is room for an iterative numpy interface, but I didn't make it yet. Would
91 that answer your question?
92
93 OD replies and asks: Partly. Do we really need a non-iterative interface?
94