-
Notifications
You must be signed in to change notification settings - Fork 5
/
autodiff.py
229 lines (192 loc) · 10.9 KB
/
autodiff.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
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
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
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
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
import numpy as np
from dependent import dependent
from independent import independent
from fielddisplay import fielddisplay
from checkfield import checkfield
from WriteData import WriteData
from MatlabArray import *
class autodiff(object):
"""autodiff class definition
Usage:
autodiff = autodiff()
"""
def __init__(self, *args): # {{{
self.isautodiff = False
self.dependents = []
self.independents = []
self.driver = 'fos_forward'
self.obufsize = np.nan
self.lbufsize = np.nan
self.cbufsize = np.nan
self.tbufsize = np.nan
self.gcTriggerMaxSize = np.nan
self.gcTriggerRatio = np.nan
self.tapeAlloc = np.nan
self.outputTapeMemory = 0
self.outputTime = 0
self.enablePreaccumulation = 0
if not len(args):
self.setdefaultparameters()
else:
raise RuntimeError("constructor not supported")
# }}}
def __repr__(self): # {{{
s = ' automatic differentiation parameters:\n'
s += '{}\n'.format(fielddisplay(self, 'isautodiff', "indicates if the automatic differentiation is activated"))
s += '{}\n'.format(fielddisplay(self, 'dependents', "list of dependent variables"))
s += '{}\n'.format(fielddisplay(self, 'independents', "list of independent variables"))
s += '{}\n'.format(fielddisplay(self, 'driver', "ADOLC driver ('fos_forward' or 'fov_forward')"))
s += '{}\n'.format(fielddisplay(self, 'obufsize', "Number of operations per buffer (== OBUFSIZE in usrparms.h)"))
s += '{}\n'.format(fielddisplay(self, 'lbufsize', "Number of locations per buffer (== LBUFSIZE in usrparms.h)"))
s += '{}\n'.format(fielddisplay(self, 'cbufsize', "Number of values per buffer (== CBUFSIZE in usrparms.h)"))
s += '{}\n'.format(fielddisplay(self, 'tbufsize', "Number of taylors per buffer (<=TBUFSIZE in usrparms.h)"))
s += '{}\n'.format(fielddisplay(self, 'gcTriggerRatio', "free location block sorting / consolidation triggered if the ratio between allocated and used locations exceeds gcTriggerRatio"))
s += '{}\n'.format(fielddisplay(self, 'gcTriggerMaxSize', "free location block sorting / consolidation triggered if the allocated locations exceed gcTriggerMaxSize)"))
s += '{}\n'.format(fielddisplay(self, 'tapeAlloc', 'Iteration count of a priori memory allocation of the AD tape'))
s += '{}\n'.format(fielddisplay(self, 'outputTapeMemory', 'Write AD tape memory statistics to file ad_mem.dat'))
s += '{}\n'.format(fielddisplay(self, 'outputTime', 'Write AD recording and evaluation times to file ad_time.dat'))
s += '{}\n'.format(fielddisplay(self, 'enablePreaccumulation', 'Enable CoDiPack preaccumulation in augmented places'))
return s
# }}}
def setdefaultparameters(self): # {{{
self.obufsize = 524288
self.lbufsize = 524288
self.cbufsize = 524288
self.tbufsize = 524288
self.gcTriggerRatio = 2.0
self.gcTriggerMaxSize = 65536
self.tapeAlloc = 15000000
return self
# }}}
def checkconsistency(self, md, solution, analyses): # {{{
# Early return
if not self.isautodiff:
return md
md = checkfield(md, 'fieldname', 'autodiff.obufsize', '>=', 524288)
md = checkfield(md, 'fieldname', 'autodiff.lbufsize', '>=', 524288)
md = checkfield(md, 'fieldname', 'autodiff.cbufsize', '>=', 524288)
md = checkfield(md, 'fieldname', 'autodiff.tbufsize', '>=', 524288)
md = checkfield(md, 'fieldname', 'autodiff.gcTriggerRatio', '>=', 2.0)
md = checkfield(md, 'fieldname', 'autodiff.gcTriggerMaxSize', '>=', 65536)
md = checkfield(md, 'fieldname', 'autodiff.tapeAlloc', '>=', 0)
# Memory and time output
md = checkfield(md, 'fieldname', 'autodiff.outputTapeMemory', 'numel', [1], 'values', [0, 1])
md = checkfield(md, 'fieldname', 'autodiff.outputTime', 'numel', [1], 'values', [0, 1]
# Memory reduction options
md = checkfield(md, 'fieldname', 'autodiff.enablePreaccumulation', '>=', 0)
# Driver value
md = checkfield(md, 'fieldname', 'autodiff.driver', 'values', ['fos_forward', 'fov_forward', 'fov_forward_all', 'fos_reverse', 'fov_reverse', 'fov_reverse_all'])
# Go through our dependents and independents and check consistency
for dep in self.dependents:
dep.checkconsistency(md, solution, analyses)
for i, indep in enumerate(self.independents):
indep.checkconsistency(md, i, solution, analyses, self.driver)
return md
# }}}
def marshall(self, prefix, md, fid): # {{{
WriteData(fid, prefix, 'object', self, 'fieldname', 'isautodiff', 'format', 'Boolean')
WriteData(fid, prefix, 'object', self, 'fieldname', 'driver', 'format', 'String')
# Early return
if not self.isautodiff:
WriteData(fid, prefix, 'data', False, 'name', 'md.autodiff.mass_flux_segments_present', 'format', 'Boolean')
WriteData(fid, prefix, 'data', False, 'name', 'md.autodiff.keep', 'format', 'Boolean')
return
# Buffer sizes
WriteData(fid, prefix, 'object', self, 'fieldname', 'obufsize', 'format', 'Double')
WriteData(fid, prefix, 'object', self, 'fieldname', 'lbufsize', 'format', 'Double')
WriteData(fid, prefix, 'object', self, 'fieldname', 'cbufsize', 'format', 'Double')
WriteData(fid, prefix, 'object', self, 'fieldname', 'tbufsize', 'format', 'Double')
WriteData(fid, prefix, 'object', self, 'fieldname', 'gcTriggerRatio', 'format', 'Double')
WriteData(fid, prefix, 'object', self, 'fieldname', 'gcTriggerMaxSize', 'format', 'Double')
WriteData(fid, prefix, 'object', self, 'fieldname', 'tapeAlloc', 'format', 'Integer')
# Output of memory and time
WriteData(fid, prefix, 'object', self, 'fieldname', 'outputTapeMemory', 'format', 'Boolean')
WriteData(fid, prefix, 'object', self, 'fieldname', 'outputTime', 'format', 'Boolean')
# Memory reduction options
WriteData(fid, prefix, 'object', self, 'fieldname', 'enablePreaccumulation', 'format', 'Boolean')
# Process dependent variables
num_dependent_objects = len(self.dependents)
WriteData(fid, prefix, 'data', num_dependent_objects, 'name', 'md.autodiff.num_dependent_objects', 'format', 'Integer')
if num_dependent_objects:
names = []
for i, dep in enumerate(self.dependents):
names.append(dep.name)
WriteData(fid, prefix, 'data', names, 'name', 'md.autodiff.dependent_object_names', 'format', 'StringArray')
# Process independent variables
num_independent_objects = len(self.independents)
WriteData(fid, prefix, 'data', num_independent_objects, 'name', 'md.autodiff.num_independent_objects', 'format', 'Integer')
for indep in self.independents:
WriteData(fid, prefix, 'data', indep.name, 'name', 'md.autodiff.independent_name', 'format', 'String')
WriteData(fid, prefix, 'data', indep.typetoscalar(), 'name', 'md.autodiff.independent_type', 'format', 'Integer')
WriteData(fid, prefix, 'data', indep.min_parameters, 'name','md.autodiff.independent_min_parameters','format', 'DoubleMat', 'mattype', 3)
WriteData(fid, prefix, 'data', indep.max_parameters, 'name', 'md.autodiff.independent_max_parameters', 'format', 'DoubleMat', 'mattype', 3)
WriteData(fid, prefix, 'data', indep.control_scaling_factor, 'name', 'md.autodiff.independent_scaling_factor', 'format', 'Double')
WriteData(fid, prefix, 'data', indep.control_size, 'name', 'md.autodiff.independent_control_size', 'format', 'Integer')
# If driver is fos_forward, build index
if strcmpi(self.driver, 'fos_forward'):
index = 0
for indep in self.independents:
if not np.isnan(indep.fos_forward_index):
index += indep.fos_forward_index
break
else:
if strcmpi(indep.type, 'scalar'):
index += 1
else:
index += indep.nods
index -= 1 # get c-index numbering going
WriteData(fid, prefix, 'data', index, 'name', 'md.autodiff.fos_forward_index', 'format', 'Integer')
# If driver is fos_reverse, build index
if strcmpi(self.driver, 'fos_reverse'):
index = 0
for dep in self.dependents:
if not np.isnan(dep.fos_reverse_index):
index += dep.fos_reverse_index
break
else:
index += 1
index -= 1 # get c-index numbering going
WriteData(fid, prefix, 'data', index, 'name', 'md.autodiff.fos_reverse_index', 'format', 'Integer')
# If driver is fov_forward, build indices
if strcmpi(self.driver, 'fov_forward'):
indices = 0
for indep in self.independents:
if indep.fos_forward_index:
indices += indep.fov_forward_indices
break
else:
if strcmpi(indep.type, 'scalar'):
indices += 1
else:
indices += indep.nods
indices -= 1 # get c-indices numbering going
WriteData(fid, prefix, 'data', indices, 'name', 'md.autodiff.fov_forward_indices', 'format', 'IntMat', 'mattype', 3)
# Deal with mass fluxes
mass_flux_segments = [dep.segments for dep in self.dependents if strcmpi(dep.name, 'MassFlux')]
if mass_flux_segments:
WriteData(fid, prefix, 'data', mass_flux_segments, 'name', 'md.autodiff.mass_flux_segments', 'format', 'MatArray')
flag = True
else:
flag = False
WriteData(fid, prefix, 'data', flag, 'name', 'md.autodiff.mass_flux_segments_present', 'format', 'Boolean')
# Deal with trace keep on
keep = False
# From ADOLC userdoc:
# The optional integer argument keep of trace on determines whether the
# numerical values of all active variables are recorded in a buffered
# temporary array or file called the taylor stack. This option takes
# effect if keep = 1 and prepares the scene for an immediately
# following gradient evaluation by a call to a routine implementing the
# reverse mode as described in the Section 4 and Section 5.
#
if len(self.driver) <= 3:
keep = False # there is no "_reverse" string within the driver string
else:
if strncmpi(self.driver[3:], '_reverse', 8):
keep = True
else:
keep = False
WriteData(fid, prefix, 'data', keep, 'name', 'md.autodiff.keep', 'format', 'Boolean')
# }}}
return
# }}}