Mercurial > pylearn

annotate doc/v2_planning/API_optimization.txt @ 1186:f111f8c2a280

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
Update after sept. 17th meeting
author Pascal Lamblin <lamblinp@iro.umontreal.ca>
date Fri, 17 Sep 2010 17:07:52 -0400
parents 4ea46ef9822a
children 073c2fab7bcd
rev   line source
1185
4ea46ef9822a small fix to make the API_optimization show on the web.
Frederic Bastien <nouiz@nouiz.org>
parents: 1182
diff changeset
1 .. _v2planning_optimization:
4ea46ef9822a small fix to make the API_optimization show on the web.
Frederic Bastien <nouiz@nouiz.org>
parents: 1182
diff changeset
2
1064
a41cc29cee26 v2planning optimization - API draft
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
3 Optimization API
a41cc29cee26 v2planning optimization - API draft
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
4 ================
a41cc29cee26 v2planning optimization - API draft
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
5
1069
16ea3e5c5a7a api_optimization: Couple questions
Olivier Delalleau <delallea@iro>
parents: 1065
diff changeset
6 Members: Bergstra, Lamblin, Delalleau, Glorot, Breuleux, Bordes
1064
a41cc29cee26 v2planning optimization - API draft
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
7 Leader: Bergstra
a41cc29cee26 v2planning optimization - API draft
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
8
a41cc29cee26 v2planning optimization - API draft
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
9
a41cc29cee26 v2planning optimization - API draft
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
10 Description
a41cc29cee26 v2planning optimization - API draft
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
11 -----------
a41cc29cee26 v2planning optimization - API draft
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
12
1065
2bbc464d6ed0 typo in doc
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1064
diff changeset
13 This API is for iterative optimization algorithms, such as:
1064
a41cc29cee26 v2planning optimization - API draft
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
14
a41cc29cee26 v2planning optimization - API draft
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
15 - stochastic gradient descent (incl. momentum, annealing)
a41cc29cee26 v2planning optimization - API draft
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
16 - delta bar delta
a41cc29cee26 v2planning optimization - API draft
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
17 - conjugate methods
a41cc29cee26 v2planning optimization - API draft
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
18 - L-BFGS
a41cc29cee26 v2planning optimization - API draft
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
19 - "Hessian Free"
a41cc29cee26 v2planning optimization - API draft
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
20 - SGD-QN
a41cc29cee26 v2planning optimization - API draft
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
21 - TONGA
a41cc29cee26 v2planning optimization - API draft
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
22
a41cc29cee26 v2planning optimization - API draft
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
23 The API includes an iterative interface based on Theano, and a one-shot
a41cc29cee26 v2planning optimization - API draft
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
24 interface similar to SciPy and MATLAB that is based on Python and Numpy, that
a41cc29cee26 v2planning optimization - API draft
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
25 only uses Theano for the implementation.
a41cc29cee26 v2planning optimization - API draft
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
26
a41cc29cee26 v2planning optimization - API draft
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
27
1100
153cf820a975 v2planning - updates to api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1069
diff changeset
28 Theano Interface
153cf820a975 v2planning - updates to api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1069
diff changeset
29 -----------------
153cf820a975 v2planning - updates to api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1069
diff changeset
30
153cf820a975 v2planning - updates to api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1069
diff changeset
31 The theano interface to optimization algorithms is to ask for a dictionary of
153cf820a975 v2planning - updates to api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1069
diff changeset
32 updates that can be used in theano.function. Implementations of iterative
153cf820a975 v2planning - updates to api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1069
diff changeset
33 optimization algorithms should be global functions with a signature like
153cf820a975 v2planning - updates to api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1069
diff changeset
34 'iterative_optimizer'.
1064
a41cc29cee26 v2planning optimization - API draft
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
35
1149
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
36 def iterative_optimizer(parameters,
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
37 cost=None,
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
38 gradients=None,
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
39 stop=None,
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
40 updates=None,
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
41 **kwargs):
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
42 """
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
43 :param parameters: list or tuple of Theano variables
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
44 that we want to optimize iteratively. If we're minimizing f(x), then
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
45 together, these variables represent 'x'. Typically these are shared
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
46 variables and their values are the initial values for the minimization
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
47 algorithm.
1064
a41cc29cee26 v2planning optimization - API draft
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
48
1149
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
49 :param cost: scalar-valued Theano variable that computes an exact or noisy estimate of
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
50 cost (what are the conditions on the noise?). Some algorithms might
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
51 need an exact cost, some algorithms might ignore the cost if the
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
52 gradients are given.
1064
a41cc29cee26 v2planning optimization - API draft
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
53
1149
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
54 :param gradients: list or tuple of Theano variables representing the gradients on
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
55 the corresponding parameters. These default to tensor.grad(cost,
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
56 parameters).
1064
a41cc29cee26 v2planning optimization - API draft
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
57
1149
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
58 :param stop: a shared variable (scalar integer) that (if provided) will be
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
59 updated to say when the iterative minimization algorithm has finished
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
60 (1) or requires more iterations (0).
1064
a41cc29cee26 v2planning optimization - API draft
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
61
1149
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
62 :param updates: a dictionary to update with the (var, new_value) items
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
63 associated with the iterative algorithm. The default is a new empty
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
64 dictionary. A KeyError is raised in case of key collisions.
1064
a41cc29cee26 v2planning optimization - API draft
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
65
1149
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
66 :param kwargs: algorithm-dependent arguments
1064
a41cc29cee26 v2planning optimization - API draft
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
67
1149
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
68 :returns: a dictionary mapping each parameter to an expression that it
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
69 should take in order to carry out the optimization procedure.
1064
a41cc29cee26 v2planning optimization - API draft
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
70
1149
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
71 If all the parameters are shared variables, then this dictionary may be
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
72 passed as the ``updates`` argument to theano.function.
1064
a41cc29cee26 v2planning optimization - API draft
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
73
1149
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
74 There may be more key,value pairs in the dictionary corresponding to
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
75 internal variables that are part of the optimization algorithm.
1064
a41cc29cee26 v2planning optimization - API draft
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
76
1149
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
77 """
1064
a41cc29cee26 v2planning optimization - API draft
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
78
a41cc29cee26 v2planning optimization - API draft
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
79
1100
153cf820a975 v2planning - updates to api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1069
diff changeset
80 Numpy Interface
153cf820a975 v2planning - updates to api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1069
diff changeset
81 ---------------
153cf820a975 v2planning - updates to api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1069
diff changeset
82
153cf820a975 v2planning - updates to api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1069
diff changeset
83 The numpy interface to optimization algorithms is supposed to mimick
153cf820a975 v2planning - updates to api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1069
diff changeset
84 scipy's. Its arguments are numpy arrays, and functions that manipulate numpy
153cf820a975 v2planning - updates to api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1069
diff changeset
85 arrays.
153cf820a975 v2planning - updates to api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1069
diff changeset
86
1149
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
87 def minimize(x0, f, df, opt_algo, **kwargs):
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
88 """
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
89 Return a point x_new with the same type as x0 that minimizes function `f`
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
90 with derivative `df`.
1064
a41cc29cee26 v2planning optimization - API draft
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
91
1149
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
92 This is supposed to provide an interface similar to scipy's minimize
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
93 routines, or MATLAB's.
1064
a41cc29cee26 v2planning optimization - API draft
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
94
1149
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
95 :type x0: numpy ndarray or list of numpy ndarrays.
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
96 :param x0: starting point for minimization
1064
a41cc29cee26 v2planning optimization - API draft
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
97
1149
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
98 :type f: python callable mapping something like x0 to a scalar
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
99 :param f: function to minimize
1064
a41cc29cee26 v2planning optimization - API draft
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
100
1149
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
101 :type df: python callable mapping something like x0 to the derivative of f at that point
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
102 :param df: derivative of `f`
1064
a41cc29cee26 v2planning optimization - API draft
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
103
1149
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
104 :param opt_algo: one of the functions that implements the
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
105 `iterative_optimizer` interface.
1064
a41cc29cee26 v2planning optimization - API draft
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
106
1149
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
107 :param kwargs: passed through to `opt_algo`
1064
a41cc29cee26 v2planning optimization - API draft
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
108
1149
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
109 """
1064
a41cc29cee26 v2planning optimization - API draft
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
110
a41cc29cee26 v2planning optimization - API draft
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
111
1149
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
112 There is also a numpy-based wrapper to the iterative algorithms.
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
113 This can be more useful than minimize() because it doesn't hog program
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
114 control. Technically minimize() is probably implemented using this
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
115 minimize_iterator interface.
1064
a41cc29cee26 v2planning optimization - API draft
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
116
1149
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
117 class minimize_iterator(object):
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
118 """
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
119 Attributes
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
120 - x - the current best estimate of the minimum
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
121 - f - the function being minimized
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
122 - df - f's derivative function
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
123 - opt_algo - the optimization algorithm at work (a serializable, callable
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
124 object with the signature of iterative_optimizer above).
1100
153cf820a975 v2planning - updates to api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1069
diff changeset
125
1149
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
126 """
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
127 def __init__(self, x0, f, df, opt_algo, **kwargs):
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
128 """Initialize state (arguments as in minimize())
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
129 """
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
130 def __iter__(self):
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
131 return self
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
132 def next(self):
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
133 """Take a step of minimization and return self raises StopIteration when
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
134 the algorithm is finished with minimization
1064
a41cc29cee26 v2planning optimization - API draft
James Bergstra <bergstrj@iro.umontreal.ca>
parents:
diff changeset
135
1149
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
136 """
1100
153cf820a975 v2planning - updates to api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1069
diff changeset
137
153cf820a975 v2planning - updates to api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1069
diff changeset
138
153cf820a975 v2planning - updates to api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1069
diff changeset
139 Examples
153cf820a975 v2planning - updates to api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1069
diff changeset
140 --------
153cf820a975 v2planning - updates to api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1069
diff changeset
141
1149
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
142 Simple stochastic gradient descent could be called like this:
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
143
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
144 sgd([p], gradients=[g], step_size=.1)
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
145
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
146 and this would return
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
147
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
148 {p:p-.1*g}
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
149
1100
153cf820a975 v2planning - updates to api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1069
diff changeset
150
153cf820a975 v2planning - updates to api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1069
diff changeset
151 Simple stochastic gradient descent with extra updates:
153cf820a975 v2planning - updates to api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1069
diff changeset
152
1149
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
153 sgd([p], gradients=[g], updates={a:b}, step_size=.1)
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
154
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
155 will return
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
156
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
157 {a:b, p:p-.1*g}
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
158
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
159
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
160 If the parameters collide with keys in a given updates dictionary an exception
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
161 will be raised:
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
162
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
163 sgd([p], gradients=[g], updates={p:b}, step_size=.1)
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
164
7c5dc11c850a cleaning up api_optimization
James Bergstra <bergstrj@iro.umontreal.ca>
parents: 1108
diff changeset
165 will raise a KeyError.