base.py
34.6 KB
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
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
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
# file is called AST to not collide with std lib module 'ast'
#
# It provides types to build ASTs in a simple lambda-notation style
#
from __future__ import print_function
from __future__ import absolute_import
from future import standard_library
import copy
import logging
import operator
import time
from collections import deque
import networkx as nx
# TODO: move things depending on numpy (among others too) to a library file
import numpy as np
import six
from six import StringIO
from six.moves import zip
from six.moves import map
from six.moves import range
standard_library.install_aliases()
logger = logging.getLogger(__name__)
np_versions = list(map(int, np.__version__.split(".")[:2]))
DEFAULT_MAX_PROGRAM_LEN = 100000
class PyllImportError(ImportError):
"""A pyll symbol was not defined in the scope """
class MissingArgument(object):
"""Object to represent a missing argument to a function application
"""
class SymbolTable(object):
"""
An object whose methods generally allocate Apply nodes.
_impls is a dictionary containing implementations for those nodes.
>>> self.add(a, b) # -- creates a new 'add' Apply node
>>> self._impl['add'](a, b) # -- this computes a + b
"""
def __init__(self):
# -- list and dict are special because they are Python builtins
self._impls = {
"list": list,
"dict": dict,
"range": range,
"len": len,
"int": int,
"float": float,
"map": map,
"max": max,
"min": min,
"getattr": getattr,
}
def _new_apply(self, name, args, kwargs, o_len, pure):
pos_args = [as_apply(a) for a in args]
named_args = [(k, as_apply(v)) for (k, v) in list(kwargs.items())]
named_args.sort()
return Apply(
name, pos_args=pos_args, named_args=named_args, o_len=o_len, pure=pure
)
def dict(self, *args, **kwargs):
# XXX: figure out len
return self._new_apply("dict", args, kwargs, o_len=None, pure=True)
def int(self, arg):
return self._new_apply("int", [as_apply(arg)], {}, o_len=None, pure=True)
def float(self, arg):
return self._new_apply("float", [as_apply(arg)], {}, o_len=None, pure=True)
def len(self, obj):
return self._new_apply("len", [obj], {}, o_len=None, pure=True)
def list(self, init):
return self._new_apply("list", [as_apply(init)], {}, o_len=None, pure=True)
def map(self, fn, seq, pure=False):
"""
pure - True is assertion that fn does not modify seq[i]
"""
return self._new_apply(
"map", [as_apply(fn), as_apply(seq)], {}, o_len=seq.o_len, pure=pure
)
def range(self, *args):
return self._new_apply("range", args, {}, o_len=None, pure=True)
def max(self, *args):
""" return max of args """
return self._new_apply(
"max", list(map(as_apply, args)), {}, o_len=None, pure=True
)
def min(self, *args):
""" return min of args """
return self._new_apply(
"min", list(map(as_apply, args)), {}, o_len=None, pure=True
)
def getattr(self, obj, attr, *args):
return self._new_apply(
"getattr",
[as_apply(obj), as_apply(attr)] + list(map(as_apply, args)),
{},
o_len=None,
pure=True,
)
def _define(self, f, o_len, pure):
name = f.__name__
entry = SymbolTableEntry(self, name, o_len, pure)
setattr(self, name, entry)
self._impls[name] = f
return f
def define(self, f, o_len=None, pure=False):
"""Decorator for adding python functions to self
"""
name = f.__name__
if hasattr(self, name):
raise ValueError("Cannot override existing symbol", name)
return self._define(f, o_len, pure)
def define_if_new(self, f, o_len=None, pure=False):
"""Pass silently if f matches the current implementation
for f.__name__"""
name = f.__name__
if hasattr(self, name) and self._impls[name] is not f:
raise ValueError("Cannot redefine existing symbol", name)
return self._define(f, o_len, pure)
def undefine(self, f):
if isinstance(f, six.string_types):
name = f
else:
name = f.__name__
del self._impls[name]
delattr(self, name)
def define_pure(self, f):
return self.define(f, o_len=None, pure=True)
def define_info(self, o_len=None, pure=False):
def wrapper(f):
return self.define(f, o_len=o_len, pure=pure)
return wrapper
def inject(self, *args, **kwargs):
"""
Add symbols from self into a dictionary and return the dict.
This is used for import-like syntax: see `import_`.
"""
rval = {}
for k in args:
try:
rval[k] = getattr(self, k)
except AttributeError:
raise PyllImportError(k)
for k, origk in list(kwargs.items()):
try:
rval[k] = getattr(self, origk)
except AttributeError:
raise PyllImportError(origk)
return rval
def import_(self, _globals, *args, **kwargs):
_globals.update(self.inject(*args, **kwargs))
class SymbolTableEntry(object):
"""A functools.partial-like class for adding symbol table entries.
"""
def __init__(self, symbol_table, apply_name, o_len, pure):
self.symbol_table = symbol_table
self.apply_name = apply_name
self.o_len = o_len
self.pure = pure
def __call__(self, *args, **kwargs):
return self.symbol_table._new_apply(
self.apply_name, args, kwargs, self.o_len, self.pure
)
scope = SymbolTable()
def as_apply(obj):
"""Smart way of turning object into an Apply
"""
if isinstance(obj, Apply):
rval = obj
elif isinstance(obj, tuple):
rval = Apply("pos_args", [as_apply(a) for a in obj], {}, len(obj))
elif isinstance(obj, list):
rval = Apply("pos_args", [as_apply(a) for a in obj], {}, None)
elif isinstance(obj, dict):
items = list(obj.items())
# -- should be fine to allow numbers and simple things
# but think about if it's ok to allow Applys
# it messes up sorting at the very least.
items.sort()
if all(isinstance(k, six.string_types) for k in obj):
named_args = [(k, as_apply(v)) for (k, v) in items]
rval = Apply("dict", [], named_args, len(named_args))
else:
new_items = [(k, as_apply(v)) for (k, v) in items]
rval = Apply("dict", [as_apply(new_items)], {}, o_len=None)
else:
rval = Literal(obj)
assert isinstance(rval, Apply)
return rval
class Apply(object):
"""
Represent a symbolic application of a symbol to arguments.
o_len - None or int if the function is guaranteed to return a fixed number
`o_len` of outputs if it returns successfully
pure - True only if the function has no relevant side-effects
"""
def __init__(
self, name, pos_args, named_args, o_len=None, pure=False, define_params=None
):
self.name = name
# -- tuples or arrays -> lists
self.pos_args = list(pos_args)
self.named_args = [[kw, arg] for (kw, arg) in named_args]
# -- o_len is attached this early to support tuple unpacking and
# list coersion.
self.o_len = o_len
self.pure = pure
# -- define_params lets us cope with stuff that may be in the
# SymbolTable on the master but not on the worker.
self.define_params = define_params
assert all(isinstance(v, Apply) for v in pos_args)
assert all(isinstance(v, Apply) for k, v in named_args)
assert all(isinstance(k, six.string_types) for k, v in named_args)
def __setstate__(self, state):
self.__dict__.update(state)
# -- On deserialization, update scope if need be.
if self.define_params:
scope.define_if_new(**self.define_params)
def eval(self, memo=None):
"""
Recursively evaluate an expression graph.
This method operates directly on the graph of extended inputs to this
node, making no attempt to modify or optimize the expression graph.
Caveats:
* If there are nodes in the graph that do not represent expressions,
(e.g. nodes that correspond to statement blocks or assertions)
then it's not clear what this routine should do, and you should
probably not call it.
* If there are Lambdas in the graph, this procedure will not evluate
them -- see rec_eval for that.
However, for many cases that are pure expression graphs, this
offers a quick and simple way to evaluate them.
"""
if memo is None:
memo = {}
if id(self) in memo:
return memo[id(self)]
else:
args = [a.eval() for a in self.pos_args]
kwargs = dict([(n, a.eval()) for (n, a) in self.named_args])
f = scope._impls[self.name]
memo[id(self)] = rval = f(*args, **kwargs)
return rval
def inputs(self):
# -- this function gets called a lot and it's not 100% safe to cache
# so the if/else is a small optimization
if self.named_args:
rval = self.pos_args + [v for (k, v) in self.named_args]
else:
rval = self.pos_args
return rval
@property
def arg(self):
# XXX: move this introspection to __init__, and change
# the basic data-structure to not use pos_args and named_args.
# XXX: think though... we want the binding to be updated if pos_args
# and named_args is modified... so maybe this is an ok way to do it?
#
# XXX: extend something to deal with Lambda objects instead of
# decorated python functions.
#
# http://docs.python.org/reference/expressions.html#calls
#
binding = {}
fn = scope._impls[self.name]
# XXX does not work for builtin functions
defaults = fn.__defaults__ # right-aligned default values for params
code = fn.__code__
extra_args_ok = bool(code.co_flags & 0x04)
extra_kwargs_ok = bool(code.co_flags & 0x08)
# -- assert that my understanding of calling protocol is correct
try:
if extra_args_ok and extra_kwargs_ok:
assert len(code.co_varnames) >= code.co_argcount + 2
param_names = code.co_varnames[: code.co_argcount + 2]
args_param = param_names[code.co_argcount]
kwargs_param = param_names[code.co_argcount + 1]
pos_params = param_names[: code.co_argcount]
elif extra_kwargs_ok:
assert len(code.co_varnames) >= code.co_argcount + 1
param_names = code.co_varnames[: code.co_argcount + 1]
kwargs_param = param_names[code.co_argcount]
pos_params = param_names[: code.co_argcount]
elif extra_args_ok:
assert len(code.co_varnames) >= code.co_argcount + 1
param_names = code.co_varnames[: code.co_argcount + 1]
args_param = param_names[code.co_argcount]
pos_params = param_names[: code.co_argcount]
else:
assert len(code.co_varnames) >= code.co_argcount
param_names = code.co_varnames[: code.co_argcount]
pos_params = param_names[: code.co_argcount]
except AssertionError:
print("YIKES: MISUNDERSTANDING OF CALL PROTOCOL:")
print(code.co_argcount)
print(code.co_varnames)
print("%x" % code.co_flags)
raise
if extra_args_ok:
binding[args_param] == []
if extra_kwargs_ok:
binding[kwargs_param] == {}
if len(self.pos_args) > code.co_argcount and not extra_args_ok:
raise TypeError("Argument count exceeds number of positional params")
# -- bind positional arguments
for param_i, arg_i in zip(param_names, self.pos_args):
binding[param_i] = arg_i
if extra_args_ok:
# XXX: THIS IS NOT BEING TESTED AND IS OBVIOUSLY BROKEN
# TODO: 'args' does not even exist at this point
binding[args_param].extend(args[code.co_argcount :])
# -- bind keyword arguments
for aname, aval in self.named_args:
try:
pos = pos_params.index(aname)
except ValueError:
if extra_kwargs_ok:
binding[kwargs_param][aname] = aval
continue
else:
raise TypeError("Unrecognized keyword argument", aname)
param = param_names[pos]
if param in binding:
raise TypeError("Duplicate argument for parameter", param)
binding[param] = aval
assert len(binding) <= len(param_names)
if len(binding) < len(param_names):
for p in param_names:
if p not in binding:
binding[p] = MissingArgument
return binding
def set_kwarg(self, name, value):
for ii, (key, val) in enumerate(self.named_args):
if key == name:
self.named_args[ii][1] = as_apply(value)
return
arg = self.arg
if name in arg and arg[name] != MissingArgument:
raise NotImplementedError("change pos arg to kw arg")
else:
self.named_args.append([name, as_apply(value)])
self.named_args.sort()
def clone_from_inputs(self, inputs, o_len="same"):
if len(inputs) != len(self.inputs()):
raise TypeError()
L = len(self.pos_args)
pos_args = list(inputs[:L])
named_args = [
[kw, inputs[L + ii]] for ii, (kw, arg) in enumerate(self.named_args)
]
# -- danger cloning with new inputs can change the o_len
if o_len == "same":
o_len = self.o_len
return self.__class__(self.name, pos_args, named_args, o_len)
def replace_input(self, old_node, new_node):
rval = []
for ii, aa in enumerate(self.pos_args):
if aa is old_node:
self.pos_args[ii] = new_node
rval.append(ii)
for ii, (nn, aa) in enumerate(self.named_args):
if aa is old_node:
self.named_args[ii][1] = new_node
rval.append(ii + len(self.pos_args))
return rval
def pprint(self, ofile, lineno=None, indent=0, memo=None):
if memo is None:
memo = {}
if lineno is None:
lineno = [0]
if self in memo:
print(lineno[0], " " * indent + memo[self], file=ofile)
lineno[0] += 1
else:
memo[self] = self.name + (" [line:%i]" % lineno[0])
print(lineno[0], " " * indent + self.name, file=ofile)
lineno[0] += 1
for arg in self.pos_args:
arg.pprint(ofile, lineno, indent + 2, memo)
for name, arg in self.named_args:
print(lineno[0], " " * indent + " " + name + " =", file=ofile)
lineno[0] += 1
arg.pprint(ofile, lineno, indent + 2, memo)
def __str__(self):
sio = StringIO()
self.pprint(sio)
return sio.getvalue()[:-1] # remove trailing '\n'
def __add__(self, other):
return scope.add(self, other)
def __radd__(self, other):
return scope.add(other, self)
def __sub__(self, other):
return scope.sub(self, other)
def __rsub__(self, other):
return scope.sub(other, self)
def __neg__(self):
return scope.neg(self)
def __mul__(self, other):
return scope.mul(self, other)
def __rmul__(self, other):
return scope.mul(other, self)
def __div__(self, other):
return scope.div(self, other)
def __rdiv__(self, other):
return scope.div(other, self)
def __truediv__(self, other):
return scope.truediv(self, other)
def __rtruediv__(self, other):
return scope.truediv(other, self)
def __floordiv__(self, other):
return scope.floordiv(self, other)
def __rfloordiv__(self, other):
return scope.floordiv(other, self)
def __pow__(self, other):
return scope.pow(self, other)
def __rpow__(self, other):
return scope.pow(other, self)
def __gt__(self, other):
return scope.gt(self, other)
def __ge__(self, other):
return scope.ge(self, other)
def __lt__(self, other):
return scope.lt(self, other)
def __le__(self, other):
return scope.le(self, other)
def __getitem__(self, idx):
if self.o_len is not None and isinstance(idx, int):
if idx >= self.o_len:
# -- this IndexError is essential for supporting
# tuple-unpacking syntax or list coersion of self.
raise IndexError()
return scope.getitem(self, idx)
def __len__(self):
if self.o_len is None:
raise TypeError("len of pyll.Apply either undefined or unknown")
return self.o_len
def __call__(self, *args, **kwargs):
return scope.call(self, args, kwargs)
def apply(name, *args, **kwargs):
pos_args = [as_apply(a) for a in args]
named_args = [(k, as_apply(v)) for (k, v) in list(kwargs.items())]
named_args.sort()
return Apply(name, pos_args=pos_args, named_args=named_args, o_len=None)
class Literal(Apply):
def __init__(self, obj=None):
try:
o_len = len(obj)
except TypeError:
o_len = None
Apply.__init__(self, "literal", [], {}, o_len, pure=True)
self._obj = obj
def eval(self, memo=None):
if memo is None:
memo = {}
return memo.setdefault(id(self), self._obj)
@property
def obj(self):
return self._obj
@property
def arg(self):
return {}
def pprint(self, ofile, lineno=None, indent=0, memo=None):
if lineno is None:
lineno = [0]
if memo is None:
memo = {}
if self in memo:
print(lineno[0], " " * indent + memo[self], file=ofile)
else:
# TODO: set up a registry for this
if isinstance(self._obj, np.ndarray):
msg = "Literal{np.ndarray,shape=%s,min=%f,max=%f}" % (
self._obj.shape,
self._obj.min(),
self._obj.max(),
)
else:
msg = "Literal{%s}" % str(self._obj)
memo[self] = "%s [line:%i]" % (msg, lineno[0])
print(lineno[0], " " * indent + msg, file=ofile)
lineno[0] += 1
def replace_input(self, old_node, new_node):
return []
def clone_from_inputs(self, inputs, o_len="same"):
return self.__class__(self._obj)
class Lambda(object):
# XXX: Extend Lambda objects to have a list of exception clauses.
# If the code of the expr() throws an error, these clauses convert
# that error to a return value.
def __init__(self, name, params, expr):
self.__name__ = name # like a python function
self.params = params # list of (name, symbol[, default_value]) tuples
self.expr = expr # pyll graph defining this Lambda
def __call__(self, *args, **kwargs):
# -- return `expr` cloned from given args and kwargs
if len(args) > len(self.params):
raise TypeError("too many arguments")
memo = {}
for arg, param in zip(args, self.params):
# print('applying with arg', param, arg)
memo[param[1]] = as_apply(arg)
if len(args) != len(self.params) or kwargs:
raise NotImplementedError("named / default arguments", (args, self.params))
rval = clone(self.expr, memo)
return rval
class UndefinedValue(object):
pass
# -- set up some convenience symbols to use as parameters in Lambda definitions
p0 = Literal(UndefinedValue)
p1 = Literal(UndefinedValue)
p2 = Literal(UndefinedValue)
p3 = Literal(UndefinedValue)
p4 = Literal(UndefinedValue)
@scope.define
def call(fn, args=(), kwargs={}):
""" call fn with given args and kwargs.
This is used to represent Apply.__call__
"""
return fn(*args, **kwargs)
@scope.define
def callpipe1(fn_list, arg):
"""
fn_list: a list lambdas that return either pyll expressions or python
values
arg: the argument to the first function in the list
return: `fn_list[-1]( ... (fn_list[1](fn_list[0](arg))))`
"""
# XXX: in current implementation, if fs are `partial`, then
# this loop will expand all functions f at once, so that they
# will all be evaluated in the same scope/memo by rec_eval.
# Normally programming languages would evaluate each f in a private
# scope
for f in fn_list:
arg = f(arg)
return arg
@scope.define
def partial(name, *args, **kwargs):
# TODO: introspect the named instruction, to retrieve the
# list of parameters *not* accounted for by args and kwargs
# then delete these stupid functions and just have one `partial`
try:
name = name.apply_name # to retrieve name from scope.foo methods
except AttributeError:
pass
my_id = len(scope._impls)
# -- create a function with this name
# the name is the string used index into scope._impls
temp_name = "partial_%s_id%i" % (name, my_id)
l = Lambda(temp_name, [("x", p0)], expr=apply(name, *(args + (p0,)), **kwargs))
scope.define(l)
# assert that the next partial will get a different id
# XXX; THIS ASSUMES THAT SCOPE ONLY GROWS
assert my_id < len(scope._impls)
rval = getattr(scope, temp_name)
return rval
def dfs(aa, seq=None, seqset=None):
if seq is None:
assert seqset is None
seq = []
seqset = {}
# -- seqset is the set of all nodes we have seen (which may be still on
# the stack)
# N.B. it used to be a stack, but now it's a dict mapping to inputs
# because that's an optimization saving us from having to call inputs
# so often.
if aa in seqset:
return
assert isinstance(aa, Apply)
seqset[aa] = aa.inputs()
for ii in seqset[aa]:
dfs(ii, seq, seqset)
seq.append(aa)
return seq
def toposort(expr):
"""
Return apply nodes of `expr` sub-tree as a list in topological order.
Raises networkx.NetworkXUnfeasible if subtree contains cycle.
"""
G = nx.DiGraph()
for node in dfs(expr):
G.add_edges_from([(n_in, node) for n_in in node.inputs()])
order = list(nx.topological_sort(G))
assert order[-1] == expr
return order
def clone(expr, memo=None):
if memo is None:
memo = {}
nodes = dfs(expr)
for node in nodes:
if node not in memo:
new_inputs = [memo[arg] for arg in node.inputs()]
new_node = node.clone_from_inputs(new_inputs)
memo[node] = new_node
return memo[expr]
def clone_merge(expr, memo=None, merge_literals=False):
nodes = dfs(expr)
if memo is None:
memo = {}
# -- args are somewhat slow to construct, so cache them out front
# XXX node.arg does not always work (builtins, weird co_flags)
node_args = [(node.pos_args, node.named_args) for node in nodes]
try:
del node
except:
pass
for ii, node_ii in enumerate(nodes):
if node_ii in memo:
continue
new_ii = None
if node_ii.pure:
for jj in range(ii):
node_jj = nodes[jj]
if node_ii.name != node_jj.name:
continue
if node_ii.name == "literal":
if not merge_literals:
continue
if node_ii._obj != node_jj._obj:
continue
else:
if node_args[ii] != node_args[jj]:
continue
logger.debug("clone_merge %s %i <- %i" % (node_ii.name, jj, ii))
new_ii = node_jj
break
if new_ii is None:
new_inputs = [memo[arg] for arg in node_ii.inputs()]
new_ii = node_ii.clone_from_inputs(new_inputs)
memo[node_ii] = new_ii
return memo[expr]
##############################################################################
##############################################################################
class GarbageCollected(object):
"""Placeholder representing a garbage-collected value """
def rec_eval(
expr,
deepcopy_inputs=False,
memo=None,
max_program_len=None,
memo_gc=True,
print_trace=False,
print_node_on_error=True,
):
"""
expr - pyll Apply instance to be evaluated
memo - optional dictionary of values to use for particular nodes
deepcopy_inputs - deepcopy inputs to every node prior to calling that
node's function on those inputs. If this leads to a different return
value, then some function (XXX add more complete DebugMode
functionality) in your graph is modifying its inputs and causing
mis-calculation. XXX: This is not a fully-functional DebugMode because
if the offender happens on account of the toposort order to be the last
user of said input, then it will not be detected as a potential
problem.
"""
if max_program_len == None:
max_program_len = DEFAULT_MAX_PROGRAM_LEN
if deepcopy_inputs not in (0, 1, False, True):
# -- I've been calling rec_eval(expr, memo) by accident a few times
# this error would have been appreciated.
raise ValueError("deepcopy_inputs should be bool", deepcopy_inputs)
node = as_apply(expr)
topnode = node
if memo is None:
memo = {}
else:
memo = dict(memo)
# -- hack for speed
# since the inputs are constant during rec_eval
# but not constant in general
node_inputs = {}
node_list = []
dfs(node, node_list, seqset=node_inputs)
# TODO: optimize dfs to not recurse past the items in memo
# this is especially important for evaluating Lambdas
# which cause rec_eval to recurse
#
# N.B. that Lambdas may expand the graph during the evaluation
# so that this iteration may be an incomplete
if memo_gc:
clients = {}
for aa in node_list:
clients.setdefault(aa, set())
for ii in node_inputs[aa]:
clients.setdefault(ii, set()).add(aa)
def set_memo(k, v):
assert v is not GarbageCollected
memo[k] = v
for ii in node_inputs[k]:
# -- if all clients of ii are already in the memo
# then we can free memo[ii] by replacing it
# with a dummy symbol
if all(iic in memo for iic in clients[ii]):
memo[ii] = GarbageCollected
else:
def set_memo(k, v):
memo[k] = v
todo = deque([topnode])
while todo:
if len(todo) > max_program_len:
raise RuntimeError("Probably infinite loop in document")
node = todo.pop()
if print_trace:
print("rec_eval:print_trace", len(todo), node.name)
if node in memo:
# -- we've already computed this, move on.
continue
# -- different kinds of nodes are treated differently:
if node.name == "switch":
# -- switch is the conditional evaluation node
switch_i_var = node.pos_args[0]
if switch_i_var in memo:
switch_i = memo[switch_i_var]
try:
int(switch_i)
except:
raise TypeError("switch argument was", switch_i)
if switch_i != int(switch_i) or switch_i < 0:
raise ValueError("switch pos must be positive int", switch_i)
rval_var = node.pos_args[int(switch_i) + 1]
if rval_var in memo:
set_memo(node, memo[rval_var])
continue
else:
waiting_on = [rval_var]
else:
waiting_on = [switch_i_var]
elif isinstance(node, Literal):
# -- constants go straight into the memo
set_memo(node, node.obj)
continue
else:
# -- normal instruction-type nodes have inputs
waiting_on = [v for v in node_inputs[node] if v not in memo]
if waiting_on:
# -- Necessary inputs have yet to be evaluated.
# push the node back in the queue, along with the
# inputs it still needs
todo.append(node)
todo.extend(waiting_on)
else:
# -- not waiting on anything;
# this instruction can be evaluated.
args = _args = [memo[v] for v in node.pos_args]
kwargs = _kwargs = dict([(k, memo[v]) for (k, v) in node.named_args])
if memo_gc:
for aa in args + list(kwargs.values()):
assert aa is not GarbageCollected
if deepcopy_inputs:
args = copy.deepcopy(_args)
kwargs = copy.deepcopy(_kwargs)
try:
rval = scope._impls[node.name](*args, **kwargs)
except Exception as e:
if print_node_on_error:
print("=" * 80)
print("ERROR in rec_eval")
print("EXCEPTION", type(e), str(e))
print("NODE")
print(node) # -- typically a multi-line string
print("=" * 80)
raise
if isinstance(rval, Apply):
# -- if an instruction returns a Pyll apply node
# it means evaluate that too. Lambdas do this.
#
# XXX: consider if it is desirable, efficient, buggy
# etc. to keep using the same memo dictionary
foo = rec_eval(rval, deepcopy_inputs, memo, memo_gc=memo_gc)
set_memo(node, foo)
else:
set_memo(node, rval)
return memo[topnode]
############################################################################
############################################################################
@scope.define_pure
def pos_args(*args):
return args
@scope.define_pure
def identity(obj):
return obj
# -- We used to define these as Python functions in this file, but the operator
# module already provides them, is slightly more efficient about it. Since
# searchspaces uses the same convention, we can more easily map graphs back
# and forth and reduce the amount of code in both codebases.
scope.define_pure(operator.getitem)
scope.define_pure(operator.add)
scope.define_pure(operator.sub)
scope.define_pure(operator.mul)
try:
scope.define_pure(operator.div)
except AttributeError:
pass # No more operator.div in Python3, but truediv also exists since Python2.2
scope.define_pure(operator.truediv)
scope.define_pure(operator.floordiv)
scope.define_pure(operator.neg)
scope.define_pure(operator.eq)
scope.define_pure(operator.lt)
scope.define_pure(operator.le)
scope.define_pure(operator.gt)
scope.define_pure(operator.ge)
@scope.define_pure
def exp(a):
return np.exp(a)
@scope.define_pure
def log(a):
return np.log(a)
@scope.define_pure
def pow(a, b):
return a ** b
@scope.define_pure
def sin(a):
return np.sin(a)
@scope.define_pure
def cos(a):
return np.cos(a)
@scope.define_pure
def tan(a):
return np.tan(a)
@scope.define_pure
def sum(x, axis=None):
if axis is None:
return np.sum(x)
else:
return np.sum(x, axis=axis)
@scope.define_pure
def sqrt(x):
return np.sqrt(x)
@scope.define_pure
def minimum(x, y):
return np.minimum(x, y)
@scope.define_pure
def maximum(x, y):
return np.maximum(x, y)
@scope.define_pure
def array_union1(args):
s = set()
for a in args:
s.update(a)
return np.asarray(sorted(s))
@scope.define_pure
def array_union(*args):
return array_union1(args)
@scope.define_pure
def asarray(a, dtype=None):
if dtype is None:
return np.asarray(a)
else:
return np.asarray(a, dtype=dtype)
@scope.define_pure
def str_join(s, seq):
return s.join(seq)
@scope.define_pure
def bincount(x, offset=0, weights=None, minlength=None, p=None):
y = np.asarray(x, dtype="int")
# hack for pchoice, p is passed as [ np.repeat(p, obs.size) ],
# so scope.len(p) gives incorrect #dimensions, need to get just the first one
if p is not None and p.ndim == 2:
assert np.all(p == p[0])
minlength = len(p[0])
return np.bincount(y - offset, weights, minlength)
@scope.define_pure
def repeat(n_times, obj):
return [obj] * n_times
@scope.define
def call_method(obj, methodname, *args, **kwargs):
method = getattr(obj, methodname)
return method(*args, **kwargs)
@scope.define_pure
def call_method_pure(obj, methodname, *args, **kwargs):
method = getattr(obj, methodname)
return method(*args, **kwargs)
@scope.define_pure
def copy_call_method_pure(obj, methodname, *args, **kwargs):
# -- this method copies object before calling the method
# so that in the case where args and kwargs are not modified
# the call_method can be done in a no-side-effect way.
#
# It is a mistake to use this method when args or kwargs are modified
# by the call to method.
method = getattr(copy.copy(obj), methodname)
return method(*args, **kwargs)
@scope.define_pure
def switch(pos, *args):
# switch is an unusual expression, in that it affects control flow
# when executed with rec_eval. args are not all evaluated, only
# args[pos] is evaluated.
# raise RuntimeError('switch is not meant to be evaluated')
#
# .. However, in quick-evaluation schemes it is handy that this be defined
# as follows:
return args[pos]
def _kwswitch(kw, **kwargs):
"""conditional evaluation according to string value"""
# Get the index of the string in kwargs to use switch
keys, values = list(zip(*sorted(kwargs.items())))
match_idx = scope.call_method_pure(keys, "index", kw)
return scope.switch(match_idx, *values)
scope.kwswitch = _kwswitch
@scope.define_pure
def Raise(etype, *args, **kwargs):
raise etype(*args, **kwargs)
@scope.define_info(o_len=2)
def curtime(obj):
return time.time(), obj
@scope.define
def pdb_settrace(obj):
import pdb
pdb.set_trace()
return obj