Book a Demo!
CoCalc Logo Icon
StoreFeaturesDocsShareSupportNewsAboutPoliciesSign UpSign In
hackassin
GitHub Repository: hackassin/Coursera-Machine-Learning
Path: blob/master/Week 9/Programming Assignment - 8/ex8/fmincg.m
616 views
1
function [X, fX, i] = fmincg(f, X, options, P1, P2, P3, P4, P5)
2
% Minimize a continuous differentialble multivariate function. Starting point
3
% is given by "X" (D by 1), and the function named in the string "f", must
4
% return a function value and a vector of partial derivatives. The Polack-
5
% Ribiere flavour of conjugate gradients is used to compute search directions,
6
% and a line search using quadratic and cubic polynomial approximations and the
7
% Wolfe-Powell stopping criteria is used together with the slope ratio method
8
% for guessing initial step sizes. Additionally a bunch of checks are made to
9
% make sure that exploration is taking place and that extrapolation will not
10
% be unboundedly large. The "length" gives the length of the run: if it is
11
% positive, it gives the maximum number of line searches, if negative its
12
% absolute gives the maximum allowed number of function evaluations. You can
13
% (optionally) give "length" a second component, which will indicate the
14
% reduction in function value to be expected in the first line-search (defaults
15
% to 1.0). The function returns when either its length is up, or if no further
16
% progress can be made (ie, we are at a minimum, or so close that due to
17
% numerical problems, we cannot get any closer). If the function terminates
18
% within a few iterations, it could be an indication that the function value
19
% and derivatives are not consistent (ie, there may be a bug in the
20
% implementation of your "f" function). The function returns the found
21
% solution "X", a vector of function values "fX" indicating the progress made
22
% and "i" the number of iterations (line searches or function evaluations,
23
% depending on the sign of "length") used.
24
%
25
% Usage: [X, fX, i] = fmincg(f, X, options, P1, P2, P3, P4, P5)
26
%
27
% See also: checkgrad
28
%
29
% Copyright (C) 2001 and 2002 by Carl Edward Rasmussen. Date 2002-02-13
30
%
31
%
32
% (C) Copyright 1999, 2000 & 2001, Carl Edward Rasmussen
33
%
34
% Permission is granted for anyone to copy, use, or modify these
35
% programs and accompanying documents for purposes of research or
36
% education, provided this copyright notice is retained, and note is
37
% made of any changes that have been made.
38
%
39
% These programs and documents are distributed without any warranty,
40
% express or implied. As the programs were written for research
41
% purposes only, they have not been tested to the degree that would be
42
% advisable in any important application. All use of these programs is
43
% entirely at the user's own risk.
44
%
45
% [ml-class] Changes Made:
46
% 1) Function name and argument specifications
47
% 2) Output display
48
%
49
50
% Read options
51
if exist('options', 'var') && ~isempty(options) && isfield(options, 'MaxIter')
52
length = options.MaxIter;
53
else
54
length = 100;
55
end
56
57
58
RHO = 0.01; % a bunch of constants for line searches
59
SIG = 0.5; % RHO and SIG are the constants in the Wolfe-Powell conditions
60
INT = 0.1; % don't reevaluate within 0.1 of the limit of the current bracket
61
EXT = 3.0; % extrapolate maximum 3 times the current bracket
62
MAX = 20; % max 20 function evaluations per line search
63
RATIO = 100; % maximum allowed slope ratio
64
65
argstr = ['feval(f, X']; % compose string used to call function
66
for i = 1:(nargin - 3)
67
argstr = [argstr, ',P', int2str(i)];
68
end
69
argstr = [argstr, ')'];
70
71
if max(size(length)) == 2, red=length(2); length=length(1); else red=1; end
72
S=['Iteration '];
73
74
i = 0; % zero the run length counter
75
ls_failed = 0; % no previous line search has failed
76
fX = [];
77
[f1 df1] = eval(argstr); % get function value and gradient
78
i = i + (length<0); % count epochs?!
79
s = -df1; % search direction is steepest
80
d1 = -s'*s; % this is the slope
81
z1 = red/(1-d1); % initial step is red/(|s|+1)
82
83
while i < abs(length) % while not finished
84
i = i + (length>0); % count iterations?!
85
86
X0 = X; f0 = f1; df0 = df1; % make a copy of current values
87
X = X + z1*s; % begin line search
88
[f2 df2] = eval(argstr);
89
i = i + (length<0); % count epochs?!
90
d2 = df2'*s;
91
f3 = f1; d3 = d1; z3 = -z1; % initialize point 3 equal to point 1
92
if length>0, M = MAX; else M = min(MAX, -length-i); end
93
success = 0; limit = -1; % initialize quanteties
94
while 1
95
while ((f2 > f1+z1*RHO*d1) || (d2 > -SIG*d1)) && (M > 0)
96
limit = z1; % tighten the bracket
97
if f2 > f1
98
z2 = z3 - (0.5*d3*z3*z3)/(d3*z3+f2-f3); % quadratic fit
99
else
100
A = 6*(f2-f3)/z3+3*(d2+d3); % cubic fit
101
B = 3*(f3-f2)-z3*(d3+2*d2);
102
z2 = (sqrt(B*B-A*d2*z3*z3)-B)/A; % numerical error possible - ok!
103
end
104
if isnan(z2) || isinf(z2)
105
z2 = z3/2; % if we had a numerical problem then bisect
106
end
107
z2 = max(min(z2, INT*z3),(1-INT)*z3); % don't accept too close to limits
108
z1 = z1 + z2; % update the step
109
X = X + z2*s;
110
[f2 df2] = eval(argstr);
111
M = M - 1; i = i + (length<0); % count epochs?!
112
d2 = df2'*s;
113
z3 = z3-z2; % z3 is now relative to the location of z2
114
end
115
if f2 > f1+z1*RHO*d1 || d2 > -SIG*d1
116
break; % this is a failure
117
elseif d2 > SIG*d1
118
success = 1; break; % success
119
elseif M == 0
120
break; % failure
121
end
122
A = 6*(f2-f3)/z3+3*(d2+d3); % make cubic extrapolation
123
B = 3*(f3-f2)-z3*(d3+2*d2);
124
z2 = -d2*z3*z3/(B+sqrt(B*B-A*d2*z3*z3)); % num. error possible - ok!
125
if ~isreal(z2) || isnan(z2) || isinf(z2) || z2 < 0 % num prob or wrong sign?
126
if limit < -0.5 % if we have no upper limit
127
z2 = z1 * (EXT-1); % the extrapolate the maximum amount
128
else
129
z2 = (limit-z1)/2; % otherwise bisect
130
end
131
elseif (limit > -0.5) && (z2+z1 > limit) % extraplation beyond max?
132
z2 = (limit-z1)/2; % bisect
133
elseif (limit < -0.5) && (z2+z1 > z1*EXT) % extrapolation beyond limit
134
z2 = z1*(EXT-1.0); % set to extrapolation limit
135
elseif z2 < -z3*INT
136
z2 = -z3*INT;
137
elseif (limit > -0.5) && (z2 < (limit-z1)*(1.0-INT)) % too close to limit?
138
z2 = (limit-z1)*(1.0-INT);
139
end
140
f3 = f2; d3 = d2; z3 = -z2; % set point 3 equal to point 2
141
z1 = z1 + z2; X = X + z2*s; % update current estimates
142
[f2 df2] = eval(argstr);
143
M = M - 1; i = i + (length<0); % count epochs?!
144
d2 = df2'*s;
145
end % end of line search
146
147
if success % if line search succeeded
148
f1 = f2; fX = [fX' f1]';
149
fprintf('%s %4i | Cost: %4.6e\r', S, i, f1);
150
s = (df2'*df2-df1'*df2)/(df1'*df1)*s - df2; % Polack-Ribiere direction
151
tmp = df1; df1 = df2; df2 = tmp; % swap derivatives
152
d2 = df1'*s;
153
if d2 > 0 % new slope must be negative
154
s = -df1; % otherwise use steepest direction
155
d2 = -s'*s;
156
end
157
z1 = z1 * min(RATIO, d1/(d2-realmin)); % slope ratio but max RATIO
158
d1 = d2;
159
ls_failed = 0; % this line search did not fail
160
else
161
X = X0; f1 = f0; df1 = df0; % restore point from before failed line search
162
if ls_failed || i > abs(length) % line search failed twice in a row
163
break; % or we ran out of time, so we give up
164
end
165
tmp = df1; df1 = df2; df2 = tmp; % swap derivatives
166
s = -df1; % try steepest
167
d1 = -s'*s;
168
z1 = 1/(1-d1);
169
ls_failed = 1; % this line search failed
170
end
171
if exist('OCTAVE_VERSION')
172
fflush(stdout);
173
end
174
end
175
fprintf('\n');
176
177