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
|
import ctypes
import math
import itertools
import functools
import operator
import collections.abc
import numpy
class NumpyArray:
def __init__(self, array):
assert len(array.shape) == len(array.strides)
minpos, nbytes = 0, 0
for i in range(len(array.shape)):
if array.strides[i] < 0:
minpos += (array.shape[i] - 1)*array.strides[i]
nbytes -= array.shape[i]*array.strides[i]
else:
nbytes += array.shape[i]*array.strides[i]
self.ptr = numpy.ctypeslib.as_array(ctypes.cast(array.ctypes.data + minpos, ctypes.POINTER(ctypes.c_uint8)), (nbytes,))
self.shape = array.shape
self.strides = array.strides
self.itemsize = array.itemsize
self.dtype = array.dtype
self.byteoffset = -minpos
def copy(self, ptr=None, shape=None, strides=None, itemsize=None, dtype=None, byteoffset=None):
out = type(self).__new__(type(self))
out.ptr = self.ptr
out.shape = self.shape
out.strides = self.strides
out.itemsize = self.itemsize
out.dtype = self.dtype
out.byteoffset = self.byteoffset
if ptr is not None:
out.ptr = ptr
if shape is not None:
out.shape = shape
if strides is not None:
out.strides = strides
if itemsize is not None:
out.itemsize = itemsize
if dtype is not None:
out.dtype = dtype
if byteoffset is not None:
out.byteoffset = byteoffset
return out
def __array__(self):
assert len(self.shape) == len(self.strides)
if len(self.shape) == 0:
return numpy.frombuffer(self.ptr[self.byteoffset : self.byteoffset + self.itemsize], dtype=self.dtype).reshape(())
else:
return numpy.lib.stride_tricks.as_strided(self.ptr[self.byteoffset : self.byteoffset + self.itemsize].view(self.dtype), self.shape, self.strides)
def tolist(self):
return numpy.array(self).tolist()
def __len__(self):
return self.shape[0]
def minmax_depth(self):
return len(self.shape), len(self.shape)
@property
def isscalar(self):
return len(self.shape) == 0
@property
def iscontiguous(self):
test = self.itemsize
for sh, st in zip(self.shape[::-1], self.strides[::-1]):
if st != test:
return False
test *= sh
return True # isscalar implies iscontiguous
def become_contiguous(self):
out = self.contiguous()
self.ptr = out.ptr
self.shape = out.shape
self.strides = out.strides
self.itemsize = out.itemsize
self.dtype = out.dtype
self.byteoffset = out.byteoffset
def contiguous(self):
if self.iscontiguous:
return self
else:
bytepos = numpy.arange(0, self.shape[0]*self.strides[0], self.strides[0])
return self.contiguous_next(bytepos)
def contiguous_next(self, bytepos):
if self.iscontiguous:
ptr = numpy.full(len(bytepos)*self.strides[0], 123, dtype=numpy.uint8)
for i in range(len(bytepos)):
ptr[i*self.strides[0] : (i + 1)*self.strides[0]] = self.ptr[self.byteoffset + bytepos[i] : self.byteoffset + bytepos[i] + self.strides[0]]
return self.copy(ptr=ptr, byteoffset=0)
elif len(self.shape) == 1:
ptr = numpy.full(len(bytepos)*self.itemsize, 123, dtype=numpy.uint8)
for i in range(len(bytepos)):
ptr[i*self.itemsize : (i + 1)*self.itemsize] = self.ptr[self.byteoffset + bytepos[i] : self.byteoffset + bytepos[i] + self.itemsize]
return self.copy(ptr=ptr, strides=(self.itemsize,), byteoffset=0)
else:
next = self.copy(shape=flatten_shape(self.shape), strides=flatten_strides(self.strides))
nextbytepos = numpy.full(len(bytepos)*self.shape[1], 999, dtype=int)
for i in range(len(bytepos)):
for j in range(self.shape[1]):
nextbytepos[i*self.shape[1] + j] = bytepos[i] + j*self.strides[1]
out = next.contiguous_next(nextbytepos)
return out.copy(shape=self.shape, strides=(self.shape[1]*out.strides[0],) + out.strides)
def __getitem__(self, where):
assert len(self.shape) != 0
if not isinstance(where, tuple):
where = (where,)
if where.count(Ellipsis) > 1:
raise ValueError("an index can only have a single ellipsis ('...')")
if len([x for x in where if x is not numpy.newaxis and x is not Ellipsis]) > len(self.shape):
raise ValueError("too many indexes for array")
if all(x is numpy.newaxis or x is Ellipsis or (isinstance(x, tuple) and len(x) == 0) or isinstance(x, (int, numpy.integer, slice)) for x in where):
next = self.copy(shape=(1,) + self.shape, strides=(self.shape[0]*self.strides[0],) + self.strides)
nexthead, nexttail = head_tail(where)
length = 1
out = next.getitem_bystrides(nexthead, nexttail, length)
return out.copy(shape=out.shape[1:], strides=out.strides[1:])
else:
where = sum([bool2int_arrays(x) for x in where], ())
broadcastable, broadcastable_j = [], []
for i, x in enumerate(where):
if not isinstance(x, tuple) and isinstance(x, (int, numpy.integer, collections.abc.Iterable)):
broadcastable_j.append(len(broadcastable))
broadcastable.append(x)
else:
broadcastable_j.append(None)
broadcasted = broadcast_arrays(*broadcastable)
where = tuple(x if broadcastable_j[i] is None else broadcasted[broadcastable_j[i]] for i, x in enumerate(where))
while broadcastable_j[0] is None:
broadcastable_j.pop(0)
while broadcastable_j[-1] is None:
broadcastable_j.pop()
if any(x is None for x in broadcastable_j) and any(isinstance(x, int) for x in broadcastable_j):
raise ValueError("awkward-array does not allow basic indexes (slices, etc.) between two advanced indexes (integer or array)")
self.become_contiguous() # on second thought, no in-place
next = self.copy(shape=(1,) + self.shape, strides=(self.shape[0]*self.strides[0],) + self.strides)
nexthead, nexttail = head_tail(where)
nextcarry = numpy.array([0])
nextadvanced = None
length = 1
out = next.getitem_next(nexthead, nexttail, nextcarry, nextadvanced, length, next.strides[0])
return out.copy(shape=out.shape[1:], strides=out.strides[1:])
def getitem_bystrides(self, head, tail, length):
assert len(self.shape) == len(self.strides)
if head is numpy.newaxis:
nexthead, nexttail = head_tail(tail)
out = self.getitem_bystrides(nexthead, nexttail, length)
shape = (length, 1) + out.shape[1:]
strides = (out.strides[0],) + out.strides
return out.copy(shape=shape, strides=strides)
elif head is Ellipsis:
mindepth, maxdepth = self.minmax_depth()
assert mindepth == maxdepth
if mindepth - 1 == sum(0 if x is numpy.newaxis else 1 for x in tail) or len(tail) == 0:
nexthead, nexttail = head_tail(tail)
return self.getitem_bystrides(nexthead, nexttail, length)
else:
return self.getitem_bystrides(slice(None), (Ellipsis,) + tail, length)
elif isinstance(head, tuple) and len(head) == 0:
return self
elif isinstance(head, (int, numpy.integer)):
assert len(self.shape) >= 2
nextbyteoffset = self.byteoffset + head*self.strides[1]
next = self.copy(shape=flatten_shape(self.shape), strides=flatten_strides(self.strides), byteoffset=nextbyteoffset)
nexthead, nexttail = head_tail(tail)
out = next.getitem_bystrides(nexthead, nexttail, length)
shape = (length,) + out.shape[1:]
return out.copy(shape=shape)
elif isinstance(head, slice):
assert len(self.shape) >= 2
start, stop, step = head.start, head.stop, head.step
if step is None:
step = 1
assert step != 0
if step > 0:
if start is None:
start = 0
if stop is None:
stop = self.shape[1]
else:
if start is None:
start = self.shape[1] - 1
if stop is None:
stop = -1
d, m = divmod(abs(start - stop), abs(step))
lenhead = d + (1 if m != 0 else 0)
nextbyteoffset = self.byteoffset + start*self.strides[1]
next = self.copy(shape=flatten_shape(self.shape), strides=flatten_strides(self.strides), byteoffset=nextbyteoffset)
nexthead, nexttail = head_tail(tail)
out = next.getitem_bystrides(nexthead, nexttail, length*lenhead)
shape = (length, lenhead) + out.shape[1:]
strides = (self.strides[0], self.strides[1] * step) + out.strides[1:]
return out.copy(shape=shape, strides=strides)
else:
raise TypeError("cannot use {0} as an index".format(head))
def getitem_next(self, head, tail, carry, advanced, length, stride):
assert len(self.shape) == len(self.strides)
if head is numpy.newaxis:
nexthead, nexttail = head_tail(tail)
out = self.getitem_next(nexthead, nexttail, carry, advanced, length, stride)
shape = (length, 1) + out.shape[1:]
strides = (out.strides[0],) + out.strides
return out.copy(shape=shape, strides=strides)
elif head is Ellipsis:
mindepth, maxdepth = self.minmax_depth()
assert mindepth == maxdepth
if mindepth - 1 == sum(0 if x is numpy.newaxis else 1 for x in tail) or len(tail) == 0:
nexthead, nexttail = head_tail(tail)
return self.getitem_next(nexthead, nexttail, carry, advanced, length, stride)
else:
return self.getitem_next(slice(None), (Ellipsis,) + tail, carry, advanced, length, stride)
elif isinstance(head, tuple) and len(head) == 0:
ptr = numpy.full(len(carry)*stride, 123, dtype=numpy.uint8)
for i in range(len(carry)):
ptr[i*stride : (i + 1)*stride] = self.ptr[self.byteoffset + carry[i]*stride : self.byteoffset + (carry[i] + 1)*stride]
shape = (len(carry),) + self.shape[1:]
strides = (stride,) + self.strides[1:]
return self.copy(ptr=ptr, shape=shape, strides=strides, byteoffset=0)
elif isinstance(head, (int, numpy.integer)):
raise Exception("these should now be broadcasted into arrays")
assert len(self.shape) >= 2
next = self.copy(shape=flatten_shape(self.shape), strides=flatten_strides(self.strides))
nexthead, nexttail = head_tail(tail)
nextcarry = numpy.full(len(carry), 999, dtype=int)
skip, remainder = divmod(self.strides[0], self.strides[1])
assert remainder == 0
for i in range(len(carry)):
nextcarry[i] = skip*carry[i] + head
out = next.getitem_next(nexthead, nexttail, nextcarry, advanced, length, next.strides[0])
shape = (length,) + out.shape[1:]
return out.copy(shape=shape)
elif isinstance(head, slice):
assert len(self.shape) >= 2
next = self.copy(shape=flatten_shape(self.shape), strides=flatten_strides(self.strides))
start, stop, step = head.start, head.stop, head.step
if step is None:
step = 1
assert step != 0
if step > 0:
if start is None:
start = 0
if stop is None:
stop = self.shape[1]
else:
if start is None:
start = self.shape[1] - 1
if stop is None:
stop = -1
d, m = divmod(abs(start - stop), abs(step))
lenhead = d + (1 if m != 0 else 0)
nexthead, nexttail = head_tail(tail)
nextcarry = numpy.full(len(carry)*lenhead, 999, dtype=int)
skip, remainder = divmod(self.strides[0], self.strides[1])
assert skip == self.shape[1]
assert remainder == 0
if advanced is None:
nextadvanced = None
for i in range(len(carry)):
for j in range(lenhead):
nextcarry[i*lenhead + j] = skip*carry[i] + start + j*step
else:
nextadvanced = numpy.full(len(carry)*lenhead, 999, dtype=int)
for i in range(len(carry)):
for j in range(lenhead):
nextcarry[i*lenhead + j] = skip*carry[i] + start + j*step
nextadvanced[i*lenhead + j] = advanced[i]
out = next.getitem_next(nexthead, nexttail, nextcarry, nextadvanced, length*lenhead, next.strides[0])
shape = (length, lenhead) + out.shape[1:]
strides = (shape[1]*out.strides[0],) + out.strides # FIXME: this 'shape[1]' could be 'lenhead'
return out.copy(shape=shape, strides=strides)
elif isinstance(head, numpy.ndarray) and issubclass(head.dtype.type, numpy.integer):
assert len(self.shape) >= 2
next = self.copy(shape=flatten_shape(self.shape), strides=flatten_strides(self.strides))
nexthead, nexttail = head_tail(tail)
skip, remainder = divmod(self.strides[0], self.strides[1])
assert skip == self.shape[1]
assert remainder == 0
flathead = head.ravel() # Zork!
if advanced is None:
nextcarry = numpy.full(len(carry)*len(flathead), 999, dtype=int)
nextadvanced = numpy.full(len(carry)*len(flathead), 999, dtype=int)
for i in range(len(carry)):
for j in range(len(flathead)):
nextcarry[i*len(flathead) + j] = skip*carry[i] + flathead[j]
nextadvanced[i*len(flathead) + j] = j
out = next.getitem_next(nexthead, nexttail, nextcarry, nextadvanced, length*len(flathead), next.strides[0])
shape = (length,) + head.shape + out.shape[1:]
strides = out.strides
for x in head.shape[::-1]:
strides = (x*strides[0],) + strides
return out.copy(shape=shape, strides=strides)
else:
nextcarry = numpy.full(len(carry), 999, dtype=int)
nextadvanced = numpy.full(len(carry), 999, dtype=int)
for i in range(len(carry)):
nextcarry[i] = skip*carry[i] + flathead[advanced[i]]
nextadvanced[i] = advanced[i]
out = next.getitem_next(nexthead, nexttail, nextcarry, nextadvanced, length*len(head), next.strides[0])
shape = (length,) + out.shape[1:]
return out.copy(shape=shape)
else:
raise TypeError("cannot use {0} as an index".format(head))
def head_tail(x):
head = () if len(x) == 0 else x[0]
tail = x[1:]
return head, tail
def product_shape(shape):
return functools.reduce(operator.mul, shape, 1)
def flatten_shape(shape):
if len(shape) == 1:
return ()
else:
return (shape[0]*shape[1],) + shape[2:]
def flatten_strides(strides):
return strides[1:]
def broadcast_arrays(*args):
return numpy.broadcast_arrays(*args)
def bool2int_arrays(whereitem):
if isinstance(whereitem, collections.abc.Iterable):
whereitem = numpy.asarray(whereitem)
if issubclass(whereitem.dtype.type, (numpy.bool, numpy.bool_)):
return numpy.nonzero(whereitem)
return (whereitem,)
# a = numpy.arange(10)[9::-2]
# print(a.tolist())
# b = NumpyArray(a)
# cut = (3,)
# acut = a[cut]
# print("should be shape", acut.shape, "strides", acut.strides)
# print(acut.tolist())
# bcut = b[cut]
# print(" is shape", bcut.shape, "strides", bcut.strides)
# print(bcut.tolist())
# if acut.tolist() != bcut.tolist():
# print("WRONG!!!")
# a = numpy.arange(7*5).reshape(7, 5)[6::-2, ::-1]
# b = NumpyArray(a)
# cut = (numpy.newaxis, numpy.newaxis, ..., slice(0, 3))
# acut = a[cut]
# print("should be shape", acut.shape, "strides", acut.strides)
# print(acut.tolist())
# bcut = b[cut]
# print(" is shape", bcut.shape, "strides", bcut.strides)
# print(bcut.tolist())
# if acut.tolist() != bcut.tolist():
# print("WRONG!!!")
# a = numpy.arange(7*5*6).reshape(7, 5, 6)
# b = NumpyArray(a)
# # cut = (slice(0, 5), numpy.array([[1, 0, 0, 1]]), numpy.array([[1], [0]]),)
# # cut = (slice(0, 5), numpy.array([[1, 0, 0, 1], [1, 0, 0, 1]]), numpy.array([[1, 1, 1, 1], [0, 0, 0, 0]]),)
# cut = (numpy.newaxis, numpy.newaxis, slice(1, 3), numpy.newaxis, slice(0, 2), numpy.newaxis, slice(2, 5))
# acut = a[cut]
# print("should be shape", acut.shape, "strides", acut.strides)
# print(acut.tolist())
# bcut = b[cut]
# print(" is shape", bcut.shape, "strides", bcut.strides)
# print(bcut.tolist())
# if acut.tolist() != bcut.tolist():
# print("WRONG!!!")
# a = numpy.arange(7*5*6*8).reshape(7, 5, 6, 8)
# b = NumpyArray(a)
# # cut = (slice(0, 5), numpy.array([[1, 0, 0, 1]]), numpy.array([[1], [0]]),)
# # cut = (slice(0, 5), numpy.array([[1, 0, 0, 1], [1, 0, 0, 1]]), numpy.array([[1, 1, 1, 1], [0, 0, 0, 0]]),)
# cut = (..., None)
# acut = a[cut]
# print("should be shape", acut.shape, "strides", acut.strides)
# print(acut.tolist())
# bcut = b[cut]
# print(" is shape", bcut.shape, "strides", bcut.strides)
# print(bcut.tolist())
# if acut.tolist() != bcut.tolist():
# print("WRONG!!!")
# a = numpy.arange(7*5*6*8).reshape(7, 5, 6, 8)[::2, ::3, ::-1, ::-2]
# b = NumpyArray(a)
# assert a.tolist() == b.tolist()
# b.become_contiguous()
# assert a.tolist() == b.tolist()
# a = numpy.arange(7*5).reshape(7, 5)
# a = numpy.arange(7*5*6).reshape(7, 5, 6)
a = numpy.arange(7*5*6*8).reshape(7, 5, 6, 8)
b = NumpyArray(a)
# for depth in 0, 1, 2:
# for cuts in itertools.permutations((0, 1, slice(0, 5), slice(1, 4), slice(2, 3)), depth):
# for depth in 0, 1, 2, 3:
# for cuts in itertools.permutations((0, 1, 2, slice(0, 5), slice(1, 4), slice(2, 3)), depth):
for depth in 0, 1, 2, 3, 4:
for cuts in itertools.permutations((0, 1, 2, 3, slice(0, 5), slice(1, 4), slice(1, 4), slice(1, 4), slice(2, 0, -1), slice(2, 0, -1), numpy.array([1, 0, 0, 1]), numpy.array([2, 2, 0, 1]), numpy.array([[1], [0]]), Ellipsis, numpy.newaxis), depth):
try:
print(cuts)
acut = a[cuts].tolist()
bcut = b[cuts].tolist()
# print(acut)
# print(bcut)
# print()
assert acut == bcut
except ValueError:
pass
|
