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
|
import numpy
import pytest
from hypothesis import given, settings
from hypothesis.strategies import composite, integers, lists
from numpy.testing import assert_allclose
from thinc.layers import Embed
from thinc.layers.uniqued import uniqued
ROWS = 10
# This test uses a newer hypothesis feature than the skanky flatmap-style
# I used previously. This is much nicer, although it still takes some getting
# used to. The key feature is this composite decorator. It injects a function,
# 'draw'.
@composite
def lists_of_integers(draw, columns=2, lo=0, hi=ROWS - 1):
# We call draw to get example values, which we can manipulate.
# Here we get a list of integers, where each member of the list
# should be between a min and max value.
int_list = draw(lists(integers(min_value=lo, max_value=hi)))
# Now we can use this int list to make an array, and it'll be the arrays
# that our functions receive.
# We trim the list, so we're of length divisible by columns.
int_list = int_list[len(int_list) % columns :]
# And make the array and reshape it.
array = numpy.array(int_list, dtype="uint64")
return array.reshape((-1, columns))
@pytest.fixture(scope="module")
def model(nO=128):
return Embed(nO, ROWS, column=0).initialize()
def test_uniqued_calls_init():
calls = []
embed = Embed(5, 5, column=0)
embed.init = lambda *args, **kwargs: calls.append(True)
embed.initialize()
assert calls == [True]
uembed = uniqued(embed)
uembed.initialize()
assert calls == [True, True]
@given(X=lists_of_integers(lo=0, hi=ROWS - 1))
@settings(deadline=None)
def test_uniqued_doesnt_change_result(model, X):
umodel = uniqued(model, column=model.attrs["column"]).initialize()
Y, bp_Y = model(X, is_train=True)
Yu, bp_Yu = umodel(X, is_train=True)
assert_allclose(Y, Yu)
dX = bp_Y(Y)
dXu = bp_Yu(Yu)
assert_allclose(dX, dXu)
if X.size:
pass
# TODO: This test is a problem, because we exceed the embedding table.
# Fix it with a better cap.
# Check that different inputs do give different results
# Z, bp_Z = model(X + 1, is_train=True)
# with pytest.raises(AssertionError):
# assert_allclose(Y, Z)
|
