File: test_lazy.py

package info (click to toggle)
python-autoray 0.7.2-1
  • links: PTS, VCS
  • area: main
  • in suites: forky, sid
  • size: 2,708 kB
  • sloc: python: 5,490; makefile: 20
file content (781 lines) | stat: -rw-r--r-- 21,848 bytes parent folder | download | duplicates (2)
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
import functools
import re

import pytest

from autoray import do, lazy, to_numpy, infer_backend, astype, shape
from numpy.testing import assert_allclose, assert_raises

from .test_autoray import BACKENDS, gen_rand


def test_manual_construct():
    def foo(a, b, c):
        a1, a2 = a
        b1 = b["1"]
        c1, c2 = c["sub"]
        return do("sum", do("stack", (a1, a2, b1, c1, c2)), axis=0)

    x = do("random.uniform", size=(5, 7), like="numpy")
    x0 = lazy.array(x[0, :])
    x1 = lazy.array(x[1, :])
    x2 = lazy.array(x[2, :])
    x3 = lazy.array(x[3, :])
    x4 = lazy.array(x[4, :])

    y = lazy.LazyArray(
        backend=infer_backend(x),
        fn=foo,
        args=((x0, x1), {"1": x2}),
        kwargs=dict(c={"sub": (x3, x4)}),
        shape=(7,),
    )

    assert y.deps == (x0, x1, x2, x3, x4)
    assert re.match(
        r"x\d+ = foo\d+\(\(x\d+, x\d+,\), "
        r"{1: x\d+}, c: {sub: \(x\d+, x\d+,\)}\)",
        y.get_source(),
    )
    assert_allclose(y.compute(), x.sum(0))


def modified_gram_schmidt(X):
    Q = []
    for j in range(0, shape(X)[0]):
        q = X[j, :]
        for i in range(0, j):
            rij = do("tensordot", do("conj", Q[i]), q, axes=1)
            q = q - rij * Q[i]
        rjj = do("linalg.norm", q, 2)
        Q.append(q / rjj)
    return do("stack", tuple(Q), axis=0)


def wrap_strict_check(larray):
    fn_orig = larray._fn

    @functools.wraps(fn_orig)
    def checked(*args, **kwargs):
        data = fn_orig(*args, **kwargs)
        assert shape(data) == shape(larray)
        assert infer_backend(data) == larray.backend
        return data

    return checked


def make_strict(larray):
    for node in larray.descend():
        larray._fn = wrap_strict_check(larray)


@pytest.mark.parametrize("backend", BACKENDS)
def test_lazy_mgs(backend):
    if backend == "sparse":
        pytest.xfail("Sparse doesn't support 'linalg.norm' yet...")
    x = gen_rand((5, 5), backend)
    lx = lazy.array(x)
    ly = modified_gram_schmidt(lx)
    ly.show()
    make_strict(ly)
    assert str(ly) == (
        f"<LazyArray(fn=stack, shape=(5, 5), backend='{backend}')>"
    )
    assert isinstance(ly, lazy.LazyArray)
    hmax = ly.history_max_size()
    hpeak = ly.history_peak_size()
    htot = ly.history_total_size()
    assert hmax == 25
    assert 25 < hpeak < htot
    assert ly.history_num_nodes() == 57
    assert len(ly.history_fn_frequencies()) == 9
    assert_allclose(to_numpy(ly.compute()), to_numpy(modified_gram_schmidt(x)))
    with lazy.shared_intermediates():
        ly = modified_gram_schmidt(lx)
        make_strict(ly)
    assert ly.history_num_nodes() == 51
    assert len(ly.history_fn_frequencies()) == 9
    assert_allclose(to_numpy(ly.compute()), to_numpy(modified_gram_schmidt(x)))


def test_partial_evaluation():
    la = lazy.array(gen_rand((10, 10), "numpy"))
    lb = lazy.array(gen_rand((10, 10), "numpy"))
    lc = lazy.array(gen_rand((10, 10), "numpy"))
    ld = lazy.array(gen_rand((10, 10), "numpy"))
    lab = do("tanh", la @ lb)
    lcd = lc @ ld
    ls = lab + lcd
    ld = do("abs", lab / lcd)
    le = do("einsum", "ab,ba->a", ls, ld)
    lf = do("sum", le)
    make_strict(lf)
    assert lf.history_num_nodes() == 12
    lf.compute_constants(variables=[lc, ld])  # constants = [la, lb]
    assert lf.history_num_nodes() == 9
    assert "tanh" not in {node.fn_name for node in lf.descend()}
    lf.compute()


def test_history_fn_frequencies():
    la = lazy.array(gen_rand((10, 10), "numpy"))
    lb = lazy.array(gen_rand((10, 10), "numpy"))
    lc = lazy.array(gen_rand((10, 10), "numpy"))
    ld = lazy.array(gen_rand((10, 10), "numpy"))
    lab = do("tanh", la @ lb)
    lcd = lc @ ld
    ls = lab + lcd
    ld = do("abs", lab / lcd)
    le = do("einsum", "ab,ba->a", ls, ld)
    lf = do("sum", le)
    assert lf.history_fn_frequencies() == {
        "None": 4,  # the inputs
        "tanh": 1,
        "matmul": 2,
        "add": 1,
        "absolute": 1,
        "truediv": 1,
        "einsum": 1,
        "sum": 1,
    }


def test_plot():
    pytest.importorskip("networkx")
    matplotlib = pytest.importorskip("matplotlib")
    matplotlib.use("Template")
    la = lazy.array(gen_rand((10, 10), "numpy"))
    lb = lazy.array(gen_rand((10, 10), "numpy"))
    lc = lazy.array(gen_rand((10, 10), "numpy"))
    ld = lazy.array(gen_rand((10, 10), "numpy"))
    lab = do("tanh", la @ lb)
    lcd = lc @ ld
    ls = lab + lcd
    ld = do("abs", lab / lcd)
    le = do("einsum", "ab,ba->a", ls, ld)
    lf = do("sum", le)
    lf.plot_graph()
    lf.plot_graph(initial_layout="layers")
    lf.plot_graph(variables=[lc, ld], color_by="variables")
    lf.plot_circuit()
    lf.plot_circuit(color_by="id")
    lf.plot_history_size_footprint()
    lf.plot_history_functions_scatter()
    lf.plot_history_functions_lines(log=2)
    lf.plot_history_functions_image(rasterize=True)
    lf.plot_history_stats()


def test_share_intermediates():
    la = lazy.array(gen_rand((10, 10), "numpy"))
    lb = lazy.array(gen_rand((10, 10), "numpy"))
    l1 = do("tanh", la @ lb)
    l2 = do("tanh", la @ lb)
    ly = l1 + l2
    assert ly.history_num_nodes() == 7
    y1 = ly.compute()
    with lazy.shared_intermediates():
        l1 = do("tanh", la @ lb)
        l2 = do("tanh", la @ lb)
        ly = l1 + l2
    assert ly.history_num_nodes() == 5
    y2 = ly.compute()
    assert_allclose(y1, y2)


@pytest.mark.parametrize("backend", BACKENDS)
def test_transpose_chain(backend):
    lx = lazy.array(gen_rand((2, 3, 4, 5, 6), backend))
    l1 = do("transpose", lx, (1, 0, 3, 2, 4))
    l2 = do("transpose", l1, (1, 0, 3, 2, 4))
    assert l2.args[0] is lx
    assert l2.deps == (lx,)
    assert l1.history_num_nodes() == 2
    assert l2.history_num_nodes() == 2
    assert_allclose(
        to_numpy(lx.compute()),
        to_numpy(l2.compute()),
    )


@pytest.mark.parametrize("backend", BACKENDS)
def test_reshape_chain(backend):
    lx = lazy.array(gen_rand((2, 3, 4, 5, 6), backend))
    l1 = do("reshape", lx, (6, 4, 30))
    l2 = do("reshape", l1, (-1,))
    assert l1.history_num_nodes() == 2
    assert l2.history_num_nodes() == 2
    assert l2.args[0] is lx
    assert l2.deps == (lx,)
    assert_allclose(
        to_numpy(lx.compute()).flatten(),
        to_numpy(l2.compute()),
        atol=1e-6,
    )


@pytest.mark.parametrize("backend", BACKENDS)
@pytest.mark.parametrize("dtype", ["float64", "complex128"])
def test_svd(backend, dtype):
    if backend == "sparse":
        pytest.xfail("Sparse doesn't support 'linalg.svd' yet...")

    if backend in ("paddle",) and "complex" in dtype:
        pytest.xfail(
            f"{backend} `linalg.solve` doesn't support complex dtype..."
        )

    x = lazy.array(gen_rand((4, 5), backend, dtype))
    U, s, VH = do("linalg.svd", x)
    assert shape(U) == (4, 4)
    assert shape(s) == (4,)
    assert shape(VH) == (4, 5)
    s = astype(s, dtype)
    ly = U @ (do("reshape", s, (-1, 1)) * VH)
    make_strict(ly)
    assert_allclose(
        to_numpy(x.compute()),
        to_numpy(ly.compute()),
    )


@pytest.mark.parametrize("backend", BACKENDS)
def test_qr(backend):
    if backend == "sparse":
        pytest.xfail("Sparse doesn't support 'linalg.qr' yet...")
    x = lazy.array(gen_rand((4, 5), backend))
    Q, R = do("linalg.qr", x)
    assert shape(Q) == (4, 4)
    assert shape(R) == (4, 5)
    ly = Q @ R
    make_strict(ly)
    assert_allclose(
        to_numpy(x.compute()),
        to_numpy(ly.compute()),
    )


@pytest.mark.parametrize("backend", BACKENDS)
@pytest.mark.parametrize("dtype", ["float64", "complex128"])
def test_eig_inv(backend, dtype):
    if backend in ("cupy", "dask", "torch", "mars", "sparse", "paddle"):
        pytest.xfail(f"{backend} doesn't support 'linalg.eig' yet...")

    # N.B. the prob that a real gaussian matrix has all real eigenvalues is
    # ``2**(-d * (d - 1) / 4)`` - see Edelman 1997 - so need ``d >> 5``
    d = 20
    x = lazy.array(gen_rand((d, d), backend, dtype))
    el, ev = do("linalg.eig", x)
    assert shape(el) == (d,)
    assert shape(ev) == (d, d)
    ly = ev @ (do("reshape", el, (-1, 1)) * do("linalg.inv", ev))
    make_strict(ly)
    assert_allclose(
        to_numpy(x.compute()),
        to_numpy(ly.compute()),
    )


@pytest.mark.parametrize("backend", BACKENDS)
@pytest.mark.parametrize("dtype", ["float64", "complex128"])
def test_eigh(backend, dtype):
    if backend in (
        "dask",
        "mars",
        "sparse",
    ):
        pytest.xfail(f"{backend} doesn't support 'linalg.eig' yet...")
    x = lazy.array(gen_rand((5, 5), backend, dtype))
    x = x + x.H
    el, ev = do("linalg.eigh", x)
    assert shape(el) == (5,)
    assert shape(ev) == (5, 5)
    ly = ev @ (do("reshape", el, (-1, 1)) * ev.H)
    make_strict(ly)
    assert_allclose(
        to_numpy(x.compute()),
        to_numpy(ly.compute()),
    )


@pytest.mark.parametrize("backend", BACKENDS)
@pytest.mark.parametrize("dtype", ["float64", "complex128"])
def test_cholesky(backend, dtype):
    if backend in ("sparse",):
        pytest.xfail(f"{backend} doesn't support 'linalg.cholesky' yet...")

    if backend in ("paddle",) and "complex" in dtype:
        pytest.xfail(f"{backend} doesn't support complex dtype...")

    x = lazy.array(gen_rand((5, 5), backend, dtype))
    x = x @ x.H
    C = do("linalg.cholesky", x)
    assert shape(C) == (5, 5)
    ly = C @ C.H
    make_strict(ly)
    assert_allclose(
        to_numpy(x.compute()),
        to_numpy(ly.compute()),
    )


@pytest.mark.parametrize("backend", BACKENDS)
@pytest.mark.parametrize("dtype", ["float64", "complex128"])
def test_solve(backend, dtype):
    if backend in ("sparse",):
        pytest.xfail(f"{backend} doesn't support 'linalg.solve' yet...")

    if backend in ("paddle",) and "complex" in dtype:
        pytest.xfail(
            f"{backend} `linalg.solve` doesn't support complex dtype..."
        )

    A = lazy.array(gen_rand((5, 5), backend, dtype))
    y = lazy.array(gen_rand((5,), backend, dtype))

    x = do("linalg.solve", A, y)
    assert shape(x) == (5,)
    # tensorflow e.g. doesn't allow ``A @ x`` for vector x ...
    ly = do("tensordot", A, x, axes=1)
    make_strict(ly)
    assert_allclose(
        to_numpy(y.compute()),
        to_numpy(ly.compute()),
    )


def test_dunder_magic():
    a = do("random.uniform", size=(), like="numpy")
    b = lazy.array(a)
    x, y, z = do("random.uniform", size=(3), like="numpy")
    a = x * a
    b = x * b
    a = a * y
    b = b * y
    a *= z
    b *= z
    assert_allclose(a, b.compute())

    a = do("random.uniform", size=(), like="numpy")
    b = lazy.array(a)
    x, y, z = do("random.uniform", size=(3), like="numpy")
    a = x + a
    b = x + b
    a = a + y
    b = b + y
    a += z
    b += z
    assert_allclose(a, b.compute())

    a = do("random.uniform", size=(), like="numpy")
    b = lazy.array(a)
    x, y, z = do("random.uniform", size=(3), like="numpy")
    a = x - a
    b = x - b
    a = a - y
    b = b - y
    a -= z
    b -= z
    assert_allclose(a, b.compute())

    a = do("random.uniform", size=(), like="numpy")
    b = lazy.array(a)
    x, y, z = do("random.uniform", size=(3), like="numpy")
    a = x / a
    b = x / b
    a = a / y
    b = b / y
    a /= z
    b /= z
    assert_allclose(a, b.compute())

    a = do("random.uniform", size=(), like="numpy")
    b = lazy.array(a)
    x, y, z = do("random.uniform", size=(3), like="numpy")
    a = x // a
    b = x // b
    a = a // y
    b = b // y
    a //= z
    b //= z
    assert_allclose(a, b.compute())

    a = do("random.uniform", size=(), like="numpy")
    b = lazy.array(a)
    x, y, z = do("random.uniform", size=(3), like="numpy")
    a = x**a
    b = x**b
    a = a**y
    b = b**y
    a **= z
    b **= z
    assert_allclose(a, b.compute())

    a = do("random.uniform", size=(3, 3), like="numpy")
    b = lazy.array(a)
    x, y, z = do("random.uniform", size=(3, 3, 3), like="numpy")
    a = x @ a
    b = x @ b
    a = a @ y
    b = b @ y
    a = a @ z
    b @= z
    assert_allclose(a, b.compute())


def test_indexing():
    a = do("random.uniform", size=(2, 3, 4, 5), like="numpy")
    b = lazy.array(a)

    for key in [0, (1, ..., -1), (0, 1, slice(None), -2)]:
        assert_allclose(a[key], b[key].compute())


@pytest.mark.parametrize("k", [-3, -1, 0, 2, 4])
@pytest.mark.parametrize(
    "shape",
    [
        (3,),
        (2, 2),
        (3, 4),
        (4, 3),
    ],
)
def test_diag(shape, k):
    a = do("random.uniform", size=shape, like="numpy")
    b = lazy.array(a)
    ad = do("diag", a, k)
    bd = do("diag", b, k)
    assert_allclose(ad, bd.compute())


def test_einsum():
    a = do("random.uniform", size=(2, 3, 4, 5), like="numpy")
    b = do("random.uniform", size=(4, 5), like="numpy")
    c = do("random.uniform", size=(6, 2, 3), like="numpy")
    eq = "abcd,cd,fab->fd"
    x1 = do("einsum", eq, a, b, c)
    la, lb, lc = map(lazy.array, (a, b, c))
    x2 = do("einsum", eq, la, lb, lc)
    assert_allclose(x1, x2.compute())


def test_tensordot():
    a = do("random.uniform", size=(7, 3, 4, 5), like="numpy")
    b = do("random.uniform", size=(5, 6, 3, 2), like="numpy")
    x1 = do("tensordot", a, b, axes=[(1, 3), (2, 0)])
    la, lb = map(lazy.array, (a, b))
    x2 = do("tensordot", la, lb, axes=[(1, 3), (2, 0)])
    assert_allclose(x1, x2.compute())


def test_use_variable_to_trace_function():
    a = lazy.Variable(shape=(2, 3), backend="numpy")
    b = lazy.Variable(shape=(3, 4), backend="numpy")
    c = do("tanh", a @ b)
    f = c.get_function([a, b])
    x = do("random.uniform", size=(2, 3), like="numpy")
    y = do("random.uniform", size=(3, 4), like="numpy")
    z = f([x, y])
    assert shape(z) == (2, 4)


def test_can_pickle_traced_function():
    import pickle

    a = lazy.Variable(shape=(2, 3), backend="numpy")
    b = lazy.Variable(shape=(3, 4), backend="numpy")
    c = do("tanh", a @ b)
    f = c.get_function([a, b])
    x = do("random.uniform", size=(2, 3), like="numpy")
    y = do("random.uniform", size=(3, 4), like="numpy")
    z = f([x, y])
    assert shape(z) == (2, 4)

    s = pickle.dumps(f)
    g = pickle.loads(s)
    z = g([x, y])
    assert shape(z) == (2, 4)


def test_where():
    a = lazy.Variable(shape=(4,), backend="numpy")
    b = lazy.Variable(shape=(4,), backend="numpy")
    c = do("where", *(a > 0, b, 1))
    f = c.get_function([a, b])
    x = do("asarray", [-0.5, -0.5, 1, 2], like="numpy")
    y = do("asarray", [1, 2, 3, 4], like="numpy")
    z = f(x, y)
    assert_allclose(z, [1, 1, 3, 4])


def test_lazy_function_pytree_input_and_output():
    inputs = {
        "a": lazy.Variable(shape=(2, 3), backend="numpy"),
        "b": lazy.Variable(shape=(3, 4), backend="numpy"),
    }
    outputs = {
        "outa": do("tanh", inputs["a"] @ inputs["b"]),
        "outb": [inputs["a"] - 1, inputs["b"] - 1],
    }
    f = lazy.Function(inputs, outputs)

    a = do("random.uniform", size=(2, 3), like="numpy")
    b = do("random.uniform", size=(3, 4), like="numpy")

    outs = f({"a": a, "b": b})

    assert_allclose(outs["outa"], do("tanh", a @ b))
    assert_allclose(outs["outb"][0], a - 1)
    assert_allclose(outs["outb"][1], b - 1)


@pytest.mark.parametrize(
    "indices",
    [
        [0, 1],
        [[0, 1], [1, 2]],
        [[[0, 1], [1, 2]], [[1, 1], [2, 2]]],
        [[[[0, 1, 2, 3]]]],
        [[[[0], [1]]], [[[2], [3]]]],
    ],
)
@pytest.mark.parametrize(
    "shape",
    [
        (4,),
        (4, 5),
        (4, 5, 6),
        (4, 5, 6, 7),
    ],
)
def test_take(indices, shape):
    a = do("random.uniform", size=shape, like="numpy")
    b = lazy.Variable(shape=shape, backend="numpy")
    np_shape = do("take", a, indices).shape
    lazy_shape = do("take", b, indices).shape

    fn = do("take", b, indices).get_function([b])
    lazy_func_shape = fn([a]).shape
    assert_allclose(np_shape, lazy_shape)
    assert_allclose(np_shape, lazy_func_shape)


@pytest.mark.parametrize(
    "indices",
    [
        [0, 1],
        [[0, 1], [1, 2]],
        [[[0, 1], [1, 2]], [[1, 1], [2, 2]]],
        [[[[0, 1, 2, 3]]]],
        [[[[0], [1]]], [[[2], [3]]]],
    ],
)
@pytest.mark.parametrize(
    "shape",
    [
        (4,),
        (4, 5),
        (4, 5, 6),
        (4, 5, 6, 7),
    ],
)
def test_getitem(indices, shape):
    a = do("random.uniform", size=shape, like="numpy")
    b = lazy.Variable(shape=shape, backend="numpy")
    np_shape = a[indices].shape
    lazy_shape = b[indices].shape

    fn = b[indices].get_function([b])
    lazy_func_shape = fn([a]).shape
    assert_allclose(np_shape, lazy_shape)
    assert_allclose(np_shape, lazy_func_shape)


def random_indexer(ndim_min=0, ndim_max=10, d_min=1, d_max=5, seed=None):
    """Generate a random shape and valid indexing object into that shape."""
    import numpy as np

    rng = np.random.default_rng(seed=seed)

    ndim = rng.integers(ndim_min, ndim_max + 1)

    # if we have a advanced indexing arrays, the shape of the array
    adv_ix_ndim = rng.integers(1, 4)
    adv_ix_shape = tuple(rng.integers(d_min, d_max + 1, size=adv_ix_ndim))

    def rand_adv_ix_broadcastable_shape():
        # get a random shape that broadcast matches adv_ix_shape
        ndim = rng.integers(1, adv_ix_ndim + 1)
        matching_shape = adv_ix_shape[-ndim:]
        return tuple(rng.choice([d, 1]) for d in matching_shape)

    shape = []
    indexer = []
    choices = ["index", "slice", "ellipsis", "array", "list", "newaxis"]

    i = 0
    while i < ndim:
        kind = rng.choice(choices)

        if kind == "newaxis":
            indexer.append(None)
            continue

        d = rng.integers(d_min, d_max + 1)
        shape.append(d)

        if kind == "index":
            ix = rng.integers(-d, d)
            if rng.random() > 0.5:
                # randomly supply integers and numpy ints
                ix = int(ix)

        elif kind == "ellipsis":
            # only one ellipsis allowed
            ix = ...
            choices.remove("ellipsis")
            # how many dims ellipsis should expand to
            i += rng.integers(0, 4)

        elif kind == "slice":
            start = rng.integers(-d - 2, d + 2)
            stop = rng.integers(-d - 2, d - 2)
            step = rng.choice([-3, -2, -1, 1, 2, 3])
            ix = slice(start, stop, step)

        elif kind == "array":
            ai_shape = rand_adv_ix_broadcastable_shape()
            ix = rng.integers(-d, d, size=ai_shape)

        elif kind == "list":
            ai_shape = rand_adv_ix_broadcastable_shape()
            ix = rng.integers(-d, d, size=ai_shape).tolist()

        indexer.append(ix)
        i += 1

    if (len(indexer) == 1) and (rng.random() > 0.5):
        # return the raw object
        (indexer,) = indexer
    else:
        indexer = tuple(indexer)

    return tuple(shape), indexer


@pytest.mark.parametrize("seed", range(1000))
def test_lazy_getitem_random(seed):
    shape, indexer = random_indexer()
    a = do("random.uniform", size=shape, like="numpy")
    ai = a[indexer]
    b = lazy.array(a)
    bi = b[indexer]
    assert bi.shape == ai.shape
    assert_allclose(bi.compute(), ai)


@pytest.mark.parametrize(
    "shape1, shape2",
    [
        ((3,), (3,)),
        ((3,), (3, 2)),
        ((6, 5, 4, 3), (3,)),
        ((7, 6, 5, 4), (7, 6, 4, 3)),
    ],
)
def test_matmul_shape(shape1, shape2):
    a = lazy.Variable(shape=shape1)
    b = lazy.Variable(shape=shape2)
    np_a = do("random.uniform", size=shape1, like="numpy")
    np_b = do("random.uniform", size=shape2, like="numpy")

    lazy_shape = (a @ b).shape
    np_shape = (np_a @ np_b).shape
    assert_allclose(lazy_shape, np_shape)


@pytest.mark.parametrize(
    "shape1, shape2",
    [
        ((3,), (1,)),
        ((3,), (4, 3)),
        ((3,), (3, 2, 1)),
        (
            (2, 2, 3, 4),
            (
                1,
                2,
                4,
                5,
            ),
        ),
        ((6, 5, 4), (6, 3, 3)),
    ],
)
def test_matmul_shape_error(shape1, shape2):
    a = lazy.Variable(shape=shape1)
    b = lazy.Variable(shape=shape2)

    def f(x, y):
        return x @ y

    assert_raises(ValueError, f, a, b)


def test_pytree_compute():
    x = do("random.uniform", size=(5, 6), like="numpy")
    lx = lazy.array(x)
    lu, ls, lv = do("linalg.svd", lx)
    lresults = {"u": lu, "s": ls, "v": lv}
    results = lazy.compute(lresults)
    assert isinstance(results, dict)
    assert infer_backend(results["s"]) == infer_backend(x)


def test_kron():
    x = do("random.uniform", size=(2, 3), like="numpy")
    y = do("random.uniform", size=(2, 3), like="numpy")
    xy = do("kron", x, y)

    lx = lazy.array(x)
    ly = lazy.array(y)
    lxy = do("kron", lx, ly)
    assert lxy.shape == xy.shape
    assert_allclose(lxy.compute(), xy)

    x = do("random.uniform", size=(3,), like="numpy")
    y = do("random.uniform", size=(3, 4, 5), like="numpy")
    xy = do("kron", x, y)

    lx = lazy.array(x)
    ly = lazy.array(y)
    lxy = do("kron", lx, ly)
    assert lxy.shape == xy.shape
    assert_allclose(lxy.compute(), xy)

    x = do("random.uniform", size=(3, 4, 5), like="numpy")
    y = do("random.uniform", size=(3,), like="numpy")
    xy = do("kron", x, y)

    lx = lazy.array(x)
    ly = lazy.array(y)
    lxy = do("kron", lx, ly)
    assert lxy.shape == xy.shape
    assert_allclose(lxy.compute(), xy)


def test_concatenate():
    x = do("random.uniform", size=(3, 4, 5), like="numpy")
    y = do("random.uniform", size=(3, 1, 5), like="numpy")
    z = do("random.uniform", size=(3, 7, 5), like="numpy")
    xyz = do("concatenate", (x, y, z), axis=1)

    lx = lazy.array(x)
    ly = lazy.array(y)
    lz = lazy.array(z)
    lxyz = do("concatenate", (lx, ly, lz), axis=1)

    assert lxyz.shape == xyz.shape
    assert_allclose(lxyz.compute(), xyz)