def test_optimised_test_pattern(self):
ts = OTPS(
target=(1, 2),
pattern=TP({
(3, 4): +1,
(5, 6): -1,
}),
pattern_translation_multiple=(7, 8),
target_translation_multiple=(9, 10),
quantisation_index=11,
decoded_value=12,
num_search_iterations=13,
)
its = invert_test_pattern_specification(ts)
assert its == OTPS(
target=(1, 2),
pattern=TP({
(3, 4): -1,
(5, 6): +1,
}),
pattern_translation_multiple=(7, 8),
target_translation_multiple=(9, 10),
quantisation_index=11,
decoded_value=12,
num_search_iterations=13,
)
def test_serialise_test_pattern_specifications():
before = OrderedDict([
((1, "LH", 2, 3), OTPS(
target=(4, 5),
pattern=TP({(10, 20): 1}),
pattern_translation_multiple=(6, 7),
target_translation_multiple=(8, 9),
quantisation_index=30,
decoded_value=40,
num_search_iterations=50,
)),
((2, "HL", 3, 2), OTPS(
target=(1, 0),
pattern=TP({(10, 20): 1}),
pattern_translation_multiple=(6, 7),
target_translation_multiple=(8, 9),
quantisation_index=30,
decoded_value=40,
num_search_iterations=50,
)),
])
after = serialise_test_pattern_specifications(OTPS, before)
after = json_roundtrip(after)
assert after == [
{
"level": 1,
"array_name": "LH",
"phase": [2, 3],
"target": [4, 5],
"pattern": serialise_test_pattern(TP({(10, 20): 1})),
"pattern_translation_multiple": [6, 7],
"target_translation_multiple": [8, 9],
"quantisation_index": 30,
"decoded_value": 40,
"num_search_iterations": 50,
},
{
"level": 2,
"array_name": "HL",
"phase": [3, 2],
"target": [1, 0],
"pattern": serialise_test_pattern(TP({(10, 20): 1})),
"pattern_translation_multiple": [6, 7],
"target_translation_multiple": [8, 9],
"quantisation_index": 30,
"decoded_value": 40,
"num_search_iterations": 50,
},
]
assert deserialise_test_pattern_specifications(OTPS, after) == before
def test_normalise_dict(self):
t = TP({
(-5, 50): +1,
(10, -1): 0,
})
assert t.origin == (50, -5)
assert t.polarities.shape == (1, 1)
def test_len(self):
t = TP({
(-5, 50): +1,
(10, -1): -1,
(0, 0): 0,
})
assert len(t) == 2
def test_offset_picture(self):
test_patterns = OrderedDict([
(
(1, "Output", 0, 0),
TPS(
pattern=TP({
(2, 1): +1,
(1, 2): -1,
}),
pattern_translation_multiple=(4, 4),
target=(1, 1),
target_translation_multiple=(1, 1),
),
),
])
width, height = 20, 10
pictures, locations = pack_test_patterns(width, height, test_patterns)
assert len(pictures) == 1
assert list(locations) == list(test_patterns)
# Should have been assigned to the top-left most point possible
exp_picture = np.zeros((height, width), dtype=np.int8)
exp_picture[1, 2] = +1
exp_picture[2, 1] = -1
assert np.array_equal(pictures[0], exp_picture)
assert locations[(1, "Output", 0, 0)] == (0, 1, 1)
def test_conversion_to_array(self):
test_pattern, search_slice = convert_test_pattern_to_padded_picture_and_slice(
# NB: Coordinates are (x, y) here
test_pattern=TP({
(2, 3): +1,
(4, 5): -1,
}),
input_min=-512,
input_max=511,
dwt_depth=3,
dwt_depth_ho=0,
)
assert np.array_equal(
test_pattern,
np.array([
[0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 511, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, -512, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0],
]))
# NB: Coordinates are (y, x) here!
assert search_slice == (slice(3, 6), slice(2, 5))
def test_serialise_picture():
before = TP({
(10, 20): +1, (11, 20): -1, (12, 20): +1, (13, 20): -1, (14, 20): +1,
(10, 21): +1, (12, 21): -1, (14, 21): +1,
(10, 22): +1, (14, 22): -1,
})
after = serialise_test_pattern(before)
assert after["dx"] == 10
assert after["dy"] == 20
assert after["width"] == 5
assert after["height"] == 3
# positive array: mask array
# 10101 11111
# 1x0x1 10101
# 1xxx0 10001
# Packed positive array: 1010110001100000________
# '----''----''----''----'
# Decimal 43 6 0
# Base64 r G A =
assert after["positive"] == "rGA="
# Packed mask array: 1111110101100010________
# '----''----''----''----'
# Decimal 63 22 8
# Base64 / W I =
assert after["mask"] == "/WI="
after = json_roundtrip(after)
assert deserialise_test_pattern(after) == before
def test_eq(self):
t1 = TP({
(0, 0): +1,
(1, 0): -1,
})
t2 = TP({
(0, 0): +1,
(1, 0): -1,
})
t3 = TP({
(1, 1): +1,
(2, 1): -1,
})
assert t1 == t2
assert t1 != (-t2)
assert t1 != t3
def test_simple_packing(self):
test_patterns = OrderedDict([
(
(1, "Output", 0, 0),
TPS(
pattern=TP({
(2, 1): +1,
(1, 2): -1,
}),
pattern_translation_multiple=(4, 4),
target=(1, 1),
target_translation_multiple=(1, 1),
),
),
(
(1, "Output", 0, 1),
TPS(
pattern=TP({
(2, 1): +1,
(1, 2): +1,
}),
pattern_translation_multiple=(4, 4),
target=(1, 1),
target_translation_multiple=(1, 1),
),
),
])
width, height = 20, 10
pictures, locations = pack_test_patterns(width, height, test_patterns)
assert len(pictures) == 1
assert list(locations) == list(test_patterns)
exp_picture = np.zeros((height, width), dtype=np.int8)
# First pattern, top left
exp_picture[1, 2] = +1
exp_picture[2, 1] = -1
# Second pattern, just underneath
exp_picture[5, 2] = +1
exp_picture[6, 1] = +1
assert np.array_equal(pictures[0], exp_picture)
assert locations[(1, "Output", 0, 0)] == (0, 1, 1)
assert locations[(1, "Output", 0, 1)] == (0, 1, 2)
def test_from_dict(self):
t = TP({
(-5, 50): +1,
(10, -1): -1,
})
assert t.origin == (-1, -5)
assert t.polarities.shape == (52, 16)
assert t.polarities[51, 0] == +1
assert t.polarities[0, 15] == -1
def test_from_origin_and_polarities(self):
polarities = np.array([
[0, 0, 0, -1],
[1, 0, 0, 0],
])
t = TP((-1, -5), polarities)
assert t.origin == (-1, -5)
assert t.polarities.dtype == np.int8
assert np.array_equal(t.polarities, polarities)
def test_as_dict(self):
t = TP({
(-5, 50): +1,
(10, -1): -1,
})
assert t.as_dict() == {
(-5, 50): +1,
(10, -1): -1,
}
def test_normalise_origin_and_polarities(self):
polarities = np.array([
[0, 0, 0, 0],
[0, 0, 1, -1],
[0, 0, 0, 0],
])
t = TP((-1, -5), polarities)
assert t.origin == (0, -3)
assert t.polarities.shape == (1, 2)
assert np.array_equal(t.polarities, [[1, -1]])
def test_rounding_up_sizes(self, dwt_depth, dwt_depth_ho, exp_shape):
test_pattern, search_slice = convert_test_pattern_to_padded_picture_and_slice(
# NB: Coordinates are (x, y) here
test_pattern=TP({
(6, 6): +1,
}),
input_min=-512,
input_max=511,
dwt_depth=dwt_depth,
dwt_depth_ho=dwt_depth_ho,
)
assert test_pattern.shape == exp_shape
def test_test_pattern(self):
ts = TPS(
target=(1, 2),
pattern=TP({
(3, 4): +1,
(5, 6): -1,
}),
pattern_translation_multiple=(7, 8),
target_translation_multiple=(9, 10),
)
its = invert_test_pattern_specification(ts)
assert its == TPS(
target=(1, 2),
pattern=TP({
(3, 4): -1,
(5, 6): +1,
}),
pattern_translation_multiple=(7, 8),
target_translation_multiple=(9, 10),
)
def test_as_picture_and_slice(self):
p, s = TP({
(1, 2): +1,
(3, 4): -1,
}).as_picture_and_slice()
assert np.array_equal(p, [
[0, 0, 0, 0],
[0, 0, 0, 0],
[0, 1, 0, 0],
[0, 0, 0, 0],
[0, 0, 0, -1],
])
assert np.array_equal(p[s], [
[1, 0, 0],
[0, 0, 0],
[0, 0, -1],
])
def test_terminate_early(
self,
synthesis_intermediate_pyexps,
filter_params,
):
input_min = -512
input_max = 511
ts = TPS(
target=(0, 0),
pattern=TP({(0, 0): -1}),
pattern_translation_multiple=(0, 0),
target_translation_multiple=(0, 0),
)
synthesis_pyexp = Argument("coeffs")[0]["LL"][0, 0]
kwargs = {
"h_filter_params": filter_params,
"v_filter_params": filter_params,
"dwt_depth": 0,
"dwt_depth_ho": 0,
"quantisation_matrix": {
0: {
"LL": 0
}
},
"synthesis_pyexp": synthesis_pyexp,
"test_pattern_specification": ts,
"input_min": input_min,
"input_max": input_max,
"max_quantisation_index": 1,
"random_state": np.random.RandomState(1),
"base_iterations": 100,
"added_corruptions_per_iteration": 1,
"removed_corruptions_per_iteration": 0,
"added_iterations_per_improvement": 50,
"number_of_searches": 10,
}
# Check that with terminate early disabled the search runs the full
# base set of iterations but finds no improvement
new_ts = optimise_synthesis_maximising_test_pattern(
terminate_early=None, **kwargs)
assert new_ts.pattern == ts.pattern
assert new_ts.num_search_iterations == 100 * 10
# Check terminate early allows early termination after the specified
# number of iterations
new_ts = optimise_synthesis_maximising_test_pattern(terminate_early=3,
**kwargs)
assert new_ts.pattern == ts.pattern
assert new_ts.num_search_iterations == 100 * 3
# Should run all iterations if an improvement is found
kwargs["test_pattern_specification"] = TPS(
target=(0, 0),
pattern=TP({(0, 0): +1}),
pattern_translation_multiple=(0, 0),
target_translation_multiple=(0, 0),
)
new_ts = optimise_synthesis_maximising_test_pattern(terminate_early=3,
**kwargs)
assert new_ts.num_search_iterations >= 100 * 10
def test_target_translation(self):
# The test patterns should be allocated like so:
#
# +--+----+-----+
# |0 | 1 | |
# +--+ | |
# | +----+-+---+
# | | 2 | |
# | | | |
# +--+------+---+
#
test_patterns = OrderedDict([
(
(1, "Output", 0, 0),
TPS(
pattern=TP({
(2, 1): +1,
(1, 2): -1,
}),
pattern_translation_multiple=(4, 4),
target=(1, 1),
target_translation_multiple=(1, 1),
),
),
(
(1, "Output", 0, 1),
TPS(
pattern=TP({
(10, 1): +1,
(1, 5): +1,
}),
pattern_translation_multiple=(4, 4),
target=(1, 1),
target_translation_multiple=(1, 1),
),
),
(
(1, "Output", 0, 2),
TPS(
pattern=TP({
(5, 1): -1,
(1, 5): -1,
}),
pattern_translation_multiple=(4, 4),
target=(2, 3),
target_translation_multiple=(4, 5),
),
),
])
width, height = 20, 15
pictures, locations = pack_test_patterns(width, height, test_patterns)
assert len(pictures) == 1
assert list(locations) == list(test_patterns)
exp_picture = np.zeros((height, width), dtype=np.int8)
# First pattern, top left
exp_picture[1, 2] = +1
exp_picture[2, 1] = -1
# Second pattern, just to the right
exp_picture[1, 14] = +1
exp_picture[5, 5] = +1
# Third pattern, below-right of both
exp_picture[13, 5] = -1
exp_picture[9, 9] = -1
assert np.array_equal(pictures[0], exp_picture)
assert locations[(1, "Output", 0, 0)] == (0, 1, 1)
assert locations[(1, "Output", 0, 1)] == (0, 2, 1)
assert locations[(1, "Output", 0, 2)] == (0, 2 + 4, 3 + (2 * 5))
def test_as_picture_and_slice_invalid(self):
with pytest.raises(ValueError):
TP({(-1, 0): +1}).as_picture_and_slice()
with pytest.raises(ValueError):
TP({(0, -1): +1}).as_picture_and_slice()
def test_as_picture_and_slice_empty(self):
p, s = TP({}).as_picture_and_slice()
assert p.shape == (0, 0)
assert p[s].shape == (0, 0)
def test_overspill_pictures(self):
# Tests that allocation can overspill onto extra pictures and that
# later allocations can be inserted onto any available picture.
#
# +------+------+ +-------------+
# | | | | 1 |
# | 0 | 2 | +-------------+
# | | | | 3 |
# +------+------+ +-------------+
#
test_patterns = OrderedDict([
(
(1, "Output", 0, 0),
TPS(
pattern=TP({
(9, 0): -1,
(0, 9): -1,
}),
pattern_translation_multiple=(1, 1),
target=(1, 1),
target_translation_multiple=(1, 1),
),
),
(
(1, "Output", 0, 1),
TPS(
pattern=TP({
(19, 0): +1,
(0, 4): -1,
}),
pattern_translation_multiple=(1, 1),
target=(1, 1),
target_translation_multiple=(1, 1),
),
),
(
(1, "Output", 0, 2),
TPS(
pattern=TP({
(9, 0): +1,
(0, 9): +1,
}),
pattern_translation_multiple=(1, 1),
target=(1, 1),
target_translation_multiple=(1, 1),
),
),
(
(1, "Output", 0, 3),
TPS(
pattern=TP({
(19, 0): -1,
(0, 4): +1,
}),
pattern_translation_multiple=(1, 1),
target=(1, 1),
target_translation_multiple=(1, 1),
),
),
])
width, height = 20, 10
pictures, locations = pack_test_patterns(width, height, test_patterns)
assert len(pictures) == 2
assert list(locations) == list(test_patterns)
exp_picture_0 = np.zeros((height, width), dtype=np.int8)
exp_picture_1 = np.zeros((height, width), dtype=np.int8)
# Pattern 0
exp_picture_0[0, 9] = -1
exp_picture_0[9, 0] = -1
# Pattern 1
exp_picture_1[0, 19] = +1
exp_picture_1[4, 0] = -1
# Pattern 2
exp_picture_0[0, 19] = +1
exp_picture_0[9, 10] = +1
# Pattern 3
exp_picture_1[5, 19] = -1
exp_picture_1[9, 0] = +1
assert np.array_equal(pictures[0], exp_picture_0)
assert np.array_equal(pictures[1], exp_picture_1)
assert locations[(1, "Output", 0, 0)] == (0, 1, 1)
assert locations[(1, "Output", 0, 1)] == (1, 1, 1)
assert locations[(1, "Output", 0, 2)] == (0, 11, 1)
assert locations[(1, "Output", 0, 3)] == (1, 1, 6)
def test_large_test_patterns_skipped(self):
test_patterns = OrderedDict([
(
(1, "Output", 0, 0),
TPS(
pattern=TP({
(1, 0): +1,
(0, 1): -1,
}),
pattern_translation_multiple=(1, 1),
target=(1, 1),
target_translation_multiple=(1, 1),
),
),
(
(1, "Output", 0, 1),
TPS(
pattern=TP({
(100, 0): -1,
(0, 100): -1,
}),
pattern_translation_multiple=(1, 1),
target=(1, 1),
target_translation_multiple=(1, 1),
),
),
(
(1, "Output", 0, 2),
TPS(
pattern=TP({
(1, 0): -1,
(0, 1): +1,
}),
pattern_translation_multiple=(1, 1),
target=(1, 1),
target_translation_multiple=(1, 1),
),
),
])
width, height = 20, 10
pictures, locations = pack_test_patterns(width, height, test_patterns)
assert len(pictures) == 1
assert list(locations) == [
(1, "Output", 0, 0),
(1, "Output", 0, 2),
]
exp_picture = np.zeros((height, width), dtype=np.int8)
# First pattern, top left
exp_picture[0, 1] = +1
exp_picture[1, 0] = -1
# Third pattern, underneath
exp_picture[2, 1] = -1
exp_picture[3, 0] = +1
assert np.array_equal(pictures[0], exp_picture)
assert locations[(1, "Output", 0, 0)] == (0, 1, 1)
assert locations[(1, "Output", 0, 2)] == (0, 1, 3)
def test_neg(self):
t = TP({
(0, 0): +1,
(1, 0): -1,
})
assert np.array_equal((-t).polarities, [[-1, +1]])
def test_empty(self):
t = TP({})
assert t.origin == (0, 0)
assert t.polarities.shape == (0, 0)
assert len(t) == 0
def test_repr(self):
t = TP({
(-10, 20): -1,
})
assert repr(t) == "TestPattern({(-10, 20): -1})"