line_search_strong_wolfe.py 14.7 KB
Newer Older
1
2
3
4
5
6
7
8
9
10
11
12
# 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/>.
Theo Steininger's avatar
Theo Steininger committed
13
14
15
16
17
#
# Copyright(C) 2013-2017 Max-Planck-Society
#
# NIFTy is being developed at the Max-Planck-Institut fuer Astrophysik
# and financially supported by the Studienstiftung des deutschen Volkes.
18

19
20
21
22
23
24
import numpy as np

from .line_search import LineSearch


class LineSearchStrongWolfe(LineSearch):
25
    """Class for finding a step size that satisfies the strong Wolfe conditions.
26

Martin Reinecke's avatar
Martin Reinecke committed
27
28
29
    Algorithm contains two stages. It begins whit a trial step length and
    keeps increasing it until it finds an acceptable step length or an
    interval. If it does not satisfy the Wolfe conditions, it performs the Zoom
30
31
32
    algorithm (second stage). By interpolating it decreases the size of the
    interval until an acceptable step length is found.

33
34
    Parameters
    ----------
35
    c1 : float
36
        Parameter for Armijo condition rule. (Default: 1e-4)
37
    c2 : float
38
        Parameter for curvature condition rule. (Default: 0.9)
39
    max_step_size : float
40
        Maximum step allowed in to be made in the descent direction.
41
42
43
44
45
        (Default: 50)
    max_iterations : integer
        Maximum number of iterations performed by the line search algorithm.
        (Default: 10)
    max_zoom_iterations : integer
46
        Maximum number of iterations performed by the zoom algorithm.
47
        (Default: 10)
48

49
50
51
52
53
54
    Attributes
    ----------
    c1 : float
        Parameter for Armijo condition rule.
    c2 : float
        Parameter for curvature condition rule.
55
    max_step_size : float
56
        Maximum step allowed in to be made in the descent direction.
57
58
59
60
    max_iterations : integer
        Maximum number of iterations performed by the line search algorithm.
    max_zoom_iterations : integer
        Maximum number of iterations performed by the zoom algorithm.
61

62
63
64
    """

    def __init__(self, c1=1e-4, c2=0.9,
Martin Reinecke's avatar
Martin Reinecke committed
65
                 max_step_size=1000000000, max_iterations=100,
Martin Reinecke's avatar
Martin Reinecke committed
66
                 max_zoom_iterations=30):
67
68
69
70
71
72
73
74
75

        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)

76
    def perform_line_search(self, energy, pk, f_k_minus_1=None):
77
        """Performs the first stage of the algorithm.
78
79
80

        It starts with a trial step size and it keeps increasing it until it
        satisfy the strong Wolf conditions. It also performs the descent and
81
        returns the optimal step length and the new enrgy.
82

83
84
85
86
87
88
89
90
        Parameters
        ----------
        energy : Energy object
            Energy object from which we will calculate the energy and the
            gradient at a specific point.
        pk : Field
            Unit vector pointing into the search direction.
        f_k_minus_1 : float
91
            Value of the fuction (which is being minimized) at the k-1
92
            iteration of the line search procedure. (Default: None)
93

94
95
96
97
98
99
100
101
        Returns
        -------
        alpha_star : float
            The optimal step length in the descent direction.
        phi_star : float
            Value of the energy after the performed descent.
        energy_star : Energy object
            The new Energy object on the new position.
102
103
104

        """

105
        self._set_line_energy(energy, pk, f_k_minus_1=f_k_minus_1)
106
107
108
109
110
        max_step_size = self.max_step_size
        max_iterations = self.max_iterations

        # initialize the zero phis
        old_phi_0 = self.f_k_minus_1
Martin Reinecke's avatar
Martin Reinecke committed
111
112
        le_0 = self.line_energy.at(0)
        phi_0 = le_0.value
113
        phiprime_0 = le_0.directional_derivative
Martin Reinecke's avatar
Martin Reinecke committed
114
        assert phiprime_0<0, "input direction must be a descent direction"
115
116
117

        # set alphas
        alpha0 = 0.
118
119
        if self.preferred_initial_step_size is not None:
            alpha1 = self.preferred_initial_step_size
120
        elif old_phi_0 is not None and phiprime_0 != 0:
121
122
123
124
125
126
127
128
129
130
131
132
            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):
Martin Reinecke's avatar
Martin Reinecke committed
133
134
            le_alpha1 = self.line_energy.at(alpha1)
            phi_alpha1 = le_alpha1.value
135
136
137
138
            if alpha1 == 0:
                self.logger.warn("Increment size became 0.")
                alpha_star = 0.
                phi_star = phi_0
Martin Reinecke's avatar
Martin Reinecke committed
139
                le_star = le_0
140
141
                break

Martin Reinecke's avatar
Martin Reinecke committed
142
            if (phi_alpha1 > phi_0 + self.c1*alpha1*phiprime_0) or \
Martin Reinecke's avatar
Martin Reinecke committed
143
               ((phi_alpha1 >= phi_alpha0) and (i > 0)):
Martin Reinecke's avatar
Martin Reinecke committed
144
                (alpha_star, phi_star, le_star) = self._zoom(
145
                                                    alpha0, alpha1,
146
147
148
                                                    phi_0, phiprime_0,
                                                    phi_alpha0,
                                                    phiprime_alpha0,
Martin Reinecke's avatar
Martin Reinecke committed
149
                                                    phi_alpha1)
150
151
                break

152
            phiprime_alpha1 = le_alpha1.directional_derivative
Martin Reinecke's avatar
Martin Reinecke committed
153
            if abs(phiprime_alpha1) <= -self.c2*phiprime_0:
154
155
                alpha_star = alpha1
                phi_star = phi_alpha1
Martin Reinecke's avatar
Martin Reinecke committed
156
                le_star = le_alpha1
157
158
159
                break

            if phiprime_alpha1 >= 0:
Martin Reinecke's avatar
Martin Reinecke committed
160
                (alpha_star, phi_star, le_star) = self._zoom(
161
                                                    alpha1, alpha0,
162
163
164
                                                    phi_0, phiprime_0,
                                                    phi_alpha1,
                                                    phiprime_alpha1,
Martin Reinecke's avatar
Martin Reinecke committed
165
                                                    phi_alpha0)
166
167
168
169
                break

            # update alphas
            alpha0, alpha1 = alpha1, min(2*alpha1, max_step_size)
Martin Reinecke's avatar
Martin Reinecke committed
170
            if alpha1 == max_step_size:
Martin Reinecke's avatar
Martin Reinecke committed
171
                print "reached max step size, bailing out"
Martin Reinecke's avatar
Martin Reinecke committed
172
173
174
175
                alpha_star = alpha1
                phi_star = phi_alpha1
                le_star = le_alpha1
                break
176
177
178
179
180
181
182
            phi_alpha0 = phi_alpha1
            phiprime_alpha0 = phiprime_alpha1

        else:
            # max_iterations was reached
            alpha_star = alpha1
            phi_star = phi_alpha1
Martin Reinecke's avatar
Martin Reinecke committed
183
            le_star = le_alpha1
184
185
            self.logger.error("The line search algorithm did not converge.")

186
        # extract the full energy from the line_energy
Martin Reinecke's avatar
Martin Reinecke committed
187
        energy_star = le_star.energy
Theo Steininger's avatar
Theo Steininger committed
188
189
        direction_length = pk.norm()
        step_length = alpha_star * direction_length
190
        return step_length, phi_star, energy_star
191
192

    def _zoom(self, alpha_lo, alpha_hi, phi_0, phiprime_0,
Martin Reinecke's avatar
Martin Reinecke committed
193
              phi_lo, phiprime_lo, phi_hi):
194
        """Performs the second stage of the line search algorithm.
195
196
197

        If the first stage was not successful then the Zoom algorithm tries to
        find a suitable step length by using bisection, quadratic, cubic
198
        interpolation.
199

200
201
202
        Parameters
        ----------
        alpha_lo : float
Martin Reinecke's avatar
Martin Reinecke committed
203
204
            A boundary for the step length interval.
            Fulfills Wolfe condition 1.
Martin Reinecke's avatar
Martin Reinecke committed
205
        alpha_hi : float
Martin Reinecke's avatar
Martin Reinecke committed
206
            The other boundary for the step length interval.
207
        phi_0 : float
208
            Value of the energy at the starting point of the line search
209
210
211
212
            algorithm.
        phiprime_0 : Field
            Gradient at the starting point of the line search algorithm.
        phi_lo : float
213
            Value of the energy if we perform a step of length alpha_lo in
214
215
            descent direction.
        phiprime_lo : Field
216
            Gradient at the nwe position if we perform a step of length
217
218
            alpha_lo in descent direction.
        phi_hi : float
219
            Value of the energy if we perform a step of length alpha_hi in
220
            descent direction.
221

222
223
224
225
226
227
228
229
        Returns
        -------
        alpha_star : float
            The optimal step length in the descent direction.
        phi_star : float
            Value of the energy after the performed descent.
        energy_star : Energy object
            The new Energy object on the new position.
230

231
        """
232
233
234
235
        max_iterations = self.max_zoom_iterations
        # define the cubic and quadratic interpolant checks
        cubic_delta = 0.2  # cubic
        quad_delta = 0.1  # quadratic
Martin Reinecke's avatar
Martin Reinecke committed
236
        phiprime_alphaj = 0.
237

Martin Reinecke's avatar
Martin Reinecke committed
238
239
        assert phi_lo <= phi_0 + self.c1*alpha_lo*phiprime_0
        assert phiprime_lo*(alpha_hi-alpha_lo)<0.
240
        for i in xrange(max_iterations):
Martin Reinecke's avatar
Martin Reinecke committed
241
242
            #assert phi_lo <= phi_0 + self.c1*alpha_lo*phiprime_0
            #assert phiprime_lo*(alpha_hi-alpha_lo)<0.
243
            delta_alpha = alpha_hi - alpha_lo
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
            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 successful, 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
265
266

            # Check if the current value of alpha_j is already sufficient
Martin Reinecke's avatar
Martin Reinecke committed
267
268
            le_alphaj = self.line_energy.at(alpha_j)
            phi_alphaj = le_alphaj.value
269

270
271
            # If the first Wolfe condition is not met replace alpha_hi
            # by alpha_j
Martin Reinecke's avatar
Martin Reinecke committed
272
            if (phi_alphaj > phi_0 + self.c1*alpha_j*phiprime_0) or\
273
               (phi_alphaj >= phi_lo):
274
                alpha_recent, phi_recent = alpha_hi, phi_hi
275
276
                alpha_hi, phi_hi = alpha_j, phi_alphaj
            else:
277
                phiprime_alphaj = le_alphaj.directional_derivative
278
                # If the second Wolfe condition is met, return the result
Martin Reinecke's avatar
Martin Reinecke committed
279
                if abs(phiprime_alphaj) <= -self.c2*phiprime_0:
280
281
                    alpha_star = alpha_j
                    phi_star = phi_alphaj
Martin Reinecke's avatar
Martin Reinecke committed
282
                    le_star = le_alphaj
283
284
285
                    break
                # If not, check the sign of the slope
                if phiprime_alphaj*delta_alpha >= 0:
286
                    alpha_recent, phi_recent = alpha_hi, phi_hi
287
                    alpha_hi, phi_hi = alpha_lo, phi_lo
288
289
                else:
                    alpha_recent, phi_recent = alpha_lo, phi_lo
290
291
292
293
294
                # Replace alpha_lo by alpha_j
                (alpha_lo, phi_lo, phiprime_lo) = (alpha_j, phi_alphaj,
                                                   phiprime_alphaj)

        else:
Martin Reinecke's avatar
Martin Reinecke committed
295
296
            alpha_star, phi_star, le_star = \
                alpha_j, phi_alphaj, le_alphaj
297
298
299
            self.logger.error("The line search algorithm (zoom) did not "
                              "converge.")

Martin Reinecke's avatar
Martin Reinecke committed
300
        return alpha_star, phi_star, le_star
301
302

    def _cubicmin(self, a, fa, fpa, b, fb, c, fc):
303
        """Estimating the minimum with cubic interpolation.
304

305
        Finds the minimizer for a cubic polynomial that goes through the
306
307
        points ( a,f(a) ), ( b,f(b) ), and ( c,f(c) ) with derivative at point
        a of fpa.
308
        f(x) = A *(x-a)^3 + B*(x-a)^2 + C*(x-a) + D
309
        If no minimizer can be found return None
310

311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
        Parameters
        ----------
        a : float
            Selected point.
        fa : float
            Value of polynomial at point a.
        fpa : Field
            Derivative at point a.
        b : float
            Selected point.
        fb : float
            Value of polynomial at point b.
        c : float
            Selected point.
        fc : float
            Value of polynomial at point c.
327

328
329
330
331
        Returns
        -------
        xmin : float
            Position of the approximated minimum.
332

333
334
335
336
337
338
339
        """

        with np.errstate(divide='raise', over='raise', invalid='raise'):
            try:
                C = fpa
                db = b - a
                dc = c - a
340
                denom = db * db * dc * dc * (db - dc)
341
                d1 = np.empty((2, 2))
342
343
344
345
                d1[0, 0] = dc * dc
                d1[0, 1] = -(db*db)
                d1[1, 0] = -(dc*dc*dc)
                d1[1, 1] = db*db*db
346
347
348
349
350
351
352
353
354
355
356
357
358
                [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):
359
        """Estimating the minimum with quadratic interpolation.
360

361
        Finds the minimizer for a quadratic polynomial that goes through
362
363
        the points ( a,f(a) ), ( b,f(b) ) with derivative at point a of fpa.
        f(x) = B*(x-a)^2 + C*(x-a) + D
364

365
366
367
368
369
370
371
372
373
374
375
376
        Parameters
        ----------
        a : float
            Selected point.
        fa : float
            Value of polynomial at point a.
        fpa : Field
            Derivative at point a.
        b : float
            Selected point.
        fb : float
            Value of polynomial at point b.
377

378
379
380
        Returns
        -------
        xmin : float
381
            Position of the approximated minimum.
382
383
384
385
386
387
388
389
390
391
392
393
394
395
        """
        # 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