line_search_strong_wolfe.py 10.2 KB
Newer Older
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
# NIFTy
# Copyright (C) 2017  Theo Steininger
#
# Author: Theo Steininger
#
# This program is free software: you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation, either version 3 of the License, or
# (at your option) any later version.
#
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
# GNU General Public License for more details.
#
# You should have received a copy of the GNU General Public License
# along with this program.  If not, see <http://www.gnu.org/licenses/>.

19
20
21
22
23
24
25
26
27
28
29
30
import numpy as np

from .line_search import LineSearch


class LineSearchStrongWolfe(LineSearch):
    """
    Class for finding a step size that satisfies the strong Wolfe conditions.
    """

    def __init__(self, c1=1e-4, c2=0.9,
                 max_step_size=50, max_iterations=10,
31
                 max_zoom_iterations=10):
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65

        """
        Parameters
        ----------

        f : callable f(x, *args)
            Objective function.

        fprime : callable f'(x, *args)
            Objective functions gradient.

        f_args : tuple (optional)
            Additional arguments passed to objective function and its
            derivation.

        c1 : float (optional)
            Parameter for Armijo condition rule.

        c2 : float (optional)
            Parameter for curvature condition rule.

        max_step_size : float (optional)
            Maximum step size
        """

        super(LineSearchStrongWolfe, self).__init__()

        self.c1 = np.float(c1)
        self.c2 = np.float(c2)
        self.max_step_size = max_step_size
        self.max_iterations = int(max_iterations)
        self.max_zoom_iterations = int(max_zoom_iterations)
        self._last_alpha_star = 1.

66
67
    def perform_line_search(self, energy, pk, f_k_minus_1=None):
        self._set_line_energy(energy, pk, f_k_minus_1=f_k_minus_1)
68
69
70
71
72
73
74
        c1 = self.c1
        c2 = self.c2
        max_step_size = self.max_step_size
        max_iterations = self.max_iterations

        # initialize the zero phis
        old_phi_0 = self.f_k_minus_1
75
76
77
        energy_0 = self.line_energy.at(0)
        phi_0 = energy_0.value
        phiprime_0 = energy_0.gradient
78
79
80
81
82
83
84

        if phiprime_0 == 0:
            self.logger.warn("Flat gradient in search direction.")
            return 0., 0.

        # set alphas
        alpha0 = 0.
85
86
87
        if self.prefered_initial_step_size is not None:
            alpha1 = self.prefered_initial_step_size
        elif old_phi_0 is not None and phiprime_0 != 0:
88
89
90
91
92
93
94
95
96
97
98
99
            alpha1 = min(1.0, 1.01*2*(phi_0 - old_phi_0)/phiprime_0)
            if alpha1 < 0:
                alpha1 = 1.0
        else:
            alpha1 = 1.0

        # give the alpha0 phis the right init value
        phi_alpha0 = phi_0
        phiprime_alpha0 = phiprime_0

        # start the minimization loop
        for i in xrange(max_iterations):
100
101
            energy_alpha1 = self.line_energy.at(alpha1)
            phi_alpha1 = energy_alpha1.value
102
103
104
105
            if alpha1 == 0:
                self.logger.warn("Increment size became 0.")
                alpha_star = 0.
                phi_star = phi_0
106
                energy_star = energy_0
107
108
109
110
                break

            if (phi_alpha1 > phi_0 + c1*alpha1*phiprime_0) or \
               ((phi_alpha1 >= phi_alpha0) and (i > 1)):
111
112
                (alpha_star, phi_star, energy_star) = self._zoom(
                                                    alpha0, alpha1,
113
114
115
116
117
118
119
                                                    phi_0, phiprime_0,
                                                    phi_alpha0,
                                                    phiprime_alpha0,
                                                    phi_alpha1,
                                                    c1, c2)
                break

120
            phiprime_alpha1 = energy_alpha1.gradient
121
122
123
            if abs(phiprime_alpha1) <= -c2*phiprime_0:
                alpha_star = alpha1
                phi_star = phi_alpha1
124
                energy_star = energy_alpha1
125
126
127
                break

            if phiprime_alpha1 >= 0:
128
129
                (alpha_star, phi_star, energy_star) = self._zoom(
                                                    alpha1, alpha0,
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
                                                    phi_0, phiprime_0,
                                                    phi_alpha1,
                                                    phiprime_alpha1,
                                                    phi_alpha0,
                                                    c1, c2)
                break

            # update alphas
            alpha0, alpha1 = alpha1, min(2*alpha1, max_step_size)
            phi_alpha0 = phi_alpha1
            phiprime_alpha0 = phiprime_alpha1
            # phi_alpha1 = self._phi(alpha1)

        else:
            # max_iterations was reached
            alpha_star = alpha1
            phi_star = phi_alpha1
147
            energy_star = energy_alpha1
148
149
150
            self.logger.error("The line search algorithm did not converge.")

        self._last_alpha_star = alpha_star
151
152
153
154
155

        # extract the full energy from the line_energy
        energy_star = energy_star.energy

        return alpha_star, phi_star, energy_star
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193

    def _zoom(self, alpha_lo, alpha_hi, phi_0, phiprime_0,
              phi_lo, phiprime_lo, phi_hi, c1, c2):

        max_iterations = self.max_zoom_iterations
        # define the cubic and quadratic interpolant checks
        cubic_delta = 0.2  # cubic
        quad_delta = 0.1  # quadratic

        # initialize the most recent versions (j-1) of phi and alpha
        alpha_recent = 0
        phi_recent = phi_0

        for i in xrange(max_iterations):
            delta_alpha = alpha_hi - alpha_lo
            if delta_alpha < 0:
                a, b = alpha_hi, alpha_lo
            else:
                a, b = alpha_lo, alpha_hi

            # Try cubic interpolation
            if i > 0:
                cubic_check = cubic_delta * delta_alpha
                alpha_j = self._cubicmin(alpha_lo, phi_lo, phiprime_lo,
                                         alpha_hi, phi_hi,
                                         alpha_recent, phi_recent)
            # If cubic was not successful or not available, try quadratic
            if (i == 0) or (alpha_j is None) or (alpha_j > b - cubic_check) or\
               (alpha_j < a + cubic_check):
                quad_check = quad_delta * delta_alpha
                alpha_j = self._quadmin(alpha_lo, phi_lo, phiprime_lo,
                                        alpha_hi, phi_hi)
                # If quadratic was not successfull, try bisection
                if (alpha_j is None) or (alpha_j > b - quad_check) or \
                   (alpha_j < a + quad_check):
                    alpha_j = alpha_lo + 0.5*delta_alpha

            # Check if the current value of alpha_j is already sufficient
194
195
            energy_alphaj = self.line_energy.at(alpha_j)
            phi_alphaj = energy_alphaj.value
196

197
198
199
200
201
202
203
            # If the first Wolfe condition is not met replace alpha_hi
            # by alpha_j
            if (phi_alphaj > phi_0 + c1*alpha_j*phiprime_0) or\
               (phi_alphaj >= phi_lo):
                alpha_recent, phi_recent = alpha_hi, phi_hi
                alpha_hi, phi_hi = alpha_j, phi_alphaj
            else:
204
                phiprime_alphaj = energy_alphaj.gradient
205
206
207
208
                # If the second Wolfe condition is met, return the result
                if abs(phiprime_alphaj) <= -c2*phiprime_0:
                    alpha_star = alpha_j
                    phi_star = phi_alphaj
209
                    energy_star = energy_alphaj
210
211
212
213
214
215
216
217
218
219
220
221
                    break
                # If not, check the sign of the slope
                if phiprime_alphaj*delta_alpha >= 0:
                    alpha_recent, phi_recent = alpha_hi, phi_hi
                    alpha_hi, phi_hi = alpha_lo, phi_lo
                else:
                    alpha_recent, phi_recent = alpha_lo, phi_lo
                # Replace alpha_lo by alpha_j
                (alpha_lo, phi_lo, phiprime_lo) = (alpha_j, phi_alphaj,
                                                   phiprime_alphaj)

        else:
222
223
            alpha_star, phi_star, energy_star = \
                alpha_j, phi_alphaj, energy_alphaj
224
225
226
            self.logger.error("The line search algorithm (zoom) did not "
                              "converge.")

227
        return alpha_star, phi_star, energy_star
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277

    def _cubicmin(self, a, fa, fpa, b, fb, c, fc):
        """
        Finds the minimizer for a cubic polynomial that goes through the
        points (a,fa), (b,fb), and (c,fc) with derivative at a of fpa.
        If no minimizer can be found return None
        """
        # f(x) = A *(x-a)^3 + B*(x-a)^2 + C*(x-a) + D

        with np.errstate(divide='raise', over='raise', invalid='raise'):
            try:
                C = fpa
                db = b - a
                dc = c - a
                denom = (db * dc) ** 2 * (db - dc)
                d1 = np.empty((2, 2))
                d1[0, 0] = dc ** 2
                d1[0, 1] = -db ** 2
                d1[1, 0] = -dc ** 3
                d1[1, 1] = db ** 3
                [A, B] = np.dot(d1, np.asarray([fb - fa - C * db,
                                                fc - fa - C * dc]).flatten())
                A /= denom
                B /= denom
                radical = B * B - 3 * A * C
                xmin = a + (-B + np.sqrt(radical)) / (3 * A)
            except ArithmeticError:
                return None
        if not np.isfinite(xmin):
            return None
        return xmin

    def _quadmin(self, a, fa, fpa, b, fb):
        """
        Finds the minimizer for a quadratic polynomial that goes through
        the points (a,fa), (b,fb) with derivative at a of fpa,
        """
        # f(x) = B*(x-a)^2 + C*(x-a) + D
        with np.errstate(divide='raise', over='raise', invalid='raise'):
            try:
                D = fa
                C = fpa
                db = b - a * 1.0
                B = (fb - D - C * db) / (db * db)
                xmin = a - C / (2.0 * B)
            except ArithmeticError:
                return None
        if not np.isfinite(xmin):
            return None
        return xmin