def test_ir_builder_vector_argminmax(dense_input_tensor: np.ndarray,
engine: MlirJitEngine, aliases: AliasMap):
# Build Function
ir_builder = MLIRFunctionBuilder(
"vector_arg_min_and_max",
input_types=["tensor<?xi32, #CV64>"],
return_types=["i64", "i64"],
aliases=aliases,
)
(vec, ) = ir_builder.inputs
arg_min = ir_builder.graphblas.reduce_to_scalar(vec, "argmin")
arg_max = ir_builder.graphblas.reduce_to_scalar(vec, "argmax")
ir_builder.return_vars(arg_min, arg_max)
vector_arg_min_and_max = ir_builder.compile(engine=engine,
passes=GRAPHBLAS_PASSES)
# Test Results
input_tensor = sparsify_array(dense_input_tensor, [True])
assert input_tensor.verify()
res_min, res_max = vector_arg_min_and_max(input_tensor)
minimum = np.min(dense_input_tensor)
maximum = np.max(dense_input_tensor)
dwimmed_dense_input_tensor = np.copy(dense_input_tensor)
dwimmed_dense_input_tensor[dwimmed_dense_input_tensor == 0] = maximum + 1
assert res_min == np.argmin(dwimmed_dense_input_tensor)
dwimmed_dense_input_tensor = np.copy(dense_input_tensor)
dwimmed_dense_input_tensor[dwimmed_dense_input_tensor == 0] = minimum - 1
assert res_max == np.argmax(dwimmed_dense_input_tensor)
def test_scan_statistics(special_passes):
# Test Results
dense_input_tensor = np.array(
[
[0, 1, 0, 1, 1, 0, 0, 0],
[1, 0, 0, 1, 1, 0, 0, 0],
[0, 0, 0, 0, 1, 1, 1, 1],
[1, 1, 0, 0, 1, 0, 0, 0],
[1, 1, 1, 1, 0, 0, 0, 0],
[0, 0, 1, 0, 0, 0, 1, 0],
[0, 0, 1, 0, 0, 1, 0, 0],
[0, 0, 1, 0, 0, 0, 0, 0],
],
dtype=np.float64,
)
input_tensor = sparsify_array(dense_input_tensor, [False, True])
assert input_tensor.verify()
result = mlalgo.scan_statistics(input_tensor,
compile_with_passes=special_passes)
# valid results are in {0, 1, 3, 4}, but we choose the lowest index
expected_result = 0
assert result == expected_result
def test_connected_components(A_dense):
A = ss.csr_matrix(A_dense)
num_connected_components, expected_ans = ss.csgraph.connected_components(A)
A_sparse = sparsify_array(A_dense.astype(float), [False, True])
ans = mlalgo.connected_components(A_sparse)
ans = ans.toarray()
assert num_connected_components == len(np.unique(ans))
assert num_connected_components == len(set(zip(ans, expected_ans)))
def test_ir_select_random(engine: MlirJitEngine, aliases: AliasMap):
# Build Function
ir_builder = MLIRFunctionBuilder(
"test_select_random",
input_types=["tensor<?x?xf64, #CSR64>", "i64", "i64"],
return_types=["tensor<?x?xf64, #CSR64>"],
aliases=aliases,
)
M, n, context = ir_builder.inputs
filtered = ir_builder.graphblas.matrix_select_random(
M, n, context, choose_n="choose_first")
ir_builder.return_vars(filtered)
test_select_random = ir_builder.compile(engine=engine,
passes=GRAPHBLAS_PASSES)
# Test Results
dense_input_tensor = np.array(
[
[1, 0, 0, 0, 0],
[-9, 2, 3, 0, 0],
[0, 0, 4, 1, 1],
[0, 0, 5, 6, 0],
[0, 0, 0, -9, 0],
],
dtype=np.float64,
)
input_tensor = sparsify_array(dense_input_tensor, [False, True])
assert input_tensor.verify()
result = test_select_random(input_tensor, 2, 0xB00)
assert result.verify()
dense_result = result.toarray()
# choose_first always selects the first N elements on the row
expected_output_tensor = np.array(
[
[1, 0, 0, 0, 0],
[-9, 2, 0, 0, 0],
[0, 0, 4, 1, 0],
[0, 0, 5, 6, 0],
[0, 0, 0, -9, 0],
],
dtype=np.float64,
)
np.testing.assert_equal(expected_output_tensor, dense_result)
def test_ir_gt_thunk(engine: MlirJitEngine, aliases: AliasMap):
# Build Function
ir_builder = MLIRFunctionBuilder(
"gt_thunk",
input_types=["tensor<?x?xf64, #CSR64>", "f64"],
return_types=["tensor<?x?xf64, #CSR64>"],
aliases=aliases,
)
M, threshold = ir_builder.inputs
twelve_scalar = ir_builder.arith.constant(12, "f64")
thirty_four_scalar = ir_builder.arith.constant(34, "f64")
M2 = ir_builder.graphblas.apply(M, "div", left=twelve_scalar)
M3 = ir_builder.graphblas.apply(M2, "div", right=thirty_four_scalar)
filtered = ir_builder.graphblas.select(M3, "gt", threshold)
ir_builder.return_vars(filtered)
gt_thunk = ir_builder.compile(engine=engine, passes=GRAPHBLAS_PASSES)
# Test Results
dense_input_tensor = np.array(
[
[1, 0, 0, 0, 0],
[-9, 2, 3, 0, 0],
[0, 0, 4, 0, 0],
[0, 0, 5, 6, 0],
[0, 0, 0, -9, 0],
],
dtype=np.float64,
)
dense_input_tensor_mask = dense_input_tensor.astype(bool)
input_tensor = sparsify_array(dense_input_tensor, [False, True])
assert input_tensor.verify()
for threshold in np.unique(dense_input_tensor):
result = gt_thunk(input_tensor, threshold)
assert result.verify()
dense_result = result.toarray()
expected_dense_result = np.copy(dense_input_tensor)
expected_dense_result[dense_input_tensor_mask] /= 12.0
expected_dense_result[dense_input_tensor_mask] **= -1
expected_dense_result[dense_input_tensor_mask] /= 34.0
expected_dense_result[expected_dense_result <= threshold] = 0
assert np.all(dense_result == expected_dense_result)
def test_bipartite_project_and_filter(special_passes):
# Test Results
r"""
0 1 2 3
|\ | /|\ |\
| \|/ | \| \
5 6 7 8 9
"""
# fmt: off
dense_input_tensor = np.array(
[ # 0 1 2 3
[1, 0, 0, 0], # 5
[-9, 1, 1, 0], # 6
[0, 0, 1, 0], # 7
[0, 0, 1, 1], # 8
[0, 0, 0, -9], # 9
],
dtype=np.float64,
)
# fmt: on
input_tensor = sparsify_array(dense_input_tensor, [False, True])
assert input_tensor.verify()
# Test row projection
result = mlalgo.bipartite_project_and_filter(input_tensor)
assert result.verify()
dense_result = result.toarray()
expected_dense_result = dense_input_tensor @ dense_input_tensor.T
expected_dense_result[expected_dense_result < 0] = 0
assert np.all(dense_result == expected_dense_result)
# Test column projection
result2 = mlalgo.bipartite_project_and_filter(
input_tensor, "column", cutoff=1.0, compile_with_passes=special_passes)
assert result2.verify()
dense_result2 = result2.toarray()
expected_dense_result2 = dense_input_tensor.T @ dense_input_tensor
expected_dense_result2[expected_dense_result2 < 1] = 0
assert np.all(dense_result2 == expected_dense_result2)
def test_ir_select_random_uniform(engine: MlirJitEngine, aliases: AliasMap):
# Build Function
ir_builder = MLIRFunctionBuilder(
"test_select_random_uniform",
input_types=["tensor<?x?xf64, #CSR64>", "i64", "!llvm.ptr<i8>"],
return_types=["tensor<?x?xf64, #CSR64>"],
aliases=aliases,
)
M, n, context = ir_builder.inputs
filtered = ir_builder.graphblas.matrix_select_random(
M, n, context, choose_n="choose_uniform")
ir_builder.return_vars(filtered)
test_select_random_uniform = ir_builder.compile(engine=engine,
passes=GRAPHBLAS_PASSES)
# Test Results
dense_input_tensor = np.array(
[
[1, 0, 0, 0, 0],
[-9, 2, 3, 0, 0],
[0, 0, 4, 1, 1],
[0, 0, 5, 6, 0],
[0, 0, 0, -9, 0],
],
dtype=np.float64,
)
input_tensor = sparsify_array(dense_input_tensor, [False, True])
assert input_tensor.verify()
rng = ChooseUniformContext(seed=2)
result = test_select_random_uniform(input_tensor, 2, rng)
assert result.verify()
dense_result = result.toarray()
expected_row_count = np.minimum((dense_input_tensor != 0).sum(axis=1), 2)
actual_row_count = (dense_result != 0).sum(axis=1)
np.testing.assert_equal(expected_row_count, actual_row_count)
# check for correct truncation
assert len(result.indices[1]) == result.pointers[1][-1]
assert len(result.values) == result.pointers[1][-1]
def test_ir_transpose(
engine: MlirJitEngine,
aliases: AliasMap,
):
# Build Functions
ir_builder = MLIRFunctionBuilder(
"transpose_wrapper",
input_types=["tensor<?x?xf64, #CSR64>"],
return_types=["tensor<?x?xf64, #CSC64>"],
aliases=aliases,
)
(input_matrix, ) = ir_builder.inputs
output_matrix = ir_builder.graphblas.transpose(input_matrix,
"tensor<?x?xf64, #CSC64>")
ir_builder.return_vars(output_matrix)
transpose_wrapper = ir_builder.compile(engine=engine,
passes=GRAPHBLAS_PASSES)
# Test Results
dense_input_matrix = np.array(
[
[0, 7, 7, 0, 7],
[0, 1, 7, 0, 0],
],
dtype=np.float64,
)
input_matrix = sparsify_array(dense_input_matrix, [False, True])
assert input_matrix.verify()
output_matrix = transpose_wrapper(input_matrix)
assert output_matrix.verify()
output_matrix = output_matrix.toarray()
expected_output_matrix = dense_input_matrix.T
assert np.all(expected_output_matrix == output_matrix)
assert callable_func(*valid_args) == 1234
return callable_func, valid_args
##############
# Test Cases #
##############
BAD_INPUT_TEST_CASES = [ # elements are ( error_type, error_match_string, bad_arg_index, bad_arg )
pytest.param(
TypeError,
" is expected to have a sparsity of ",
3,
sparsify_array(np.arange(10, dtype=np.float32), [True]),
id="bad_sparse_rank",
),
pytest.param(
TypeError,
" is expected to have a sparsity of ",
3,
sparsify_array(
np.arange(10, dtype=np.float32).reshape([5, 2]), [True, True]),
id="bad_sparsity",
),
pytest.param(
TypeError,
" is expected to contain elements with type ",
3,
sparsify_array(
def test_graph_wave():
A = np.array([
[0, 1, 0, 1, 1, 0, 0, 0],
[1, 0, 0, 1, 1, 0, 0, 0],
[0, 0, 0, 0, 1, 1, 1, 1],
[1, 1, 0, 0, 1, 0, 0, 0],
[1, 1, 1, 1, 0, 0, 0, 0],
[0, 0, 1, 0, 0, 0, 1, 0],
[0, 0, 1, 0, 0, 1, 0, 0],
[0, 0, 1, 0, 0, 0, 0, 0],
])
A_sparse = sparsify_array(A.astype(np.float64), [False, True])
# fmt: off
expected_ans = np.array(
[[
1, 0, 0.87774496, 0.18258206, 0.73336003, 0.12847191, 0.81480474,
0.04483691, 0.92858125, 0.18957235, 1, 0, 0.88585881, 0.1907184,
0.73123909, 0.16473694, 0.77447817, 0.07729033, 0.89243752,
0.18318066
],
[
1, 0, 0.87774496, 0.18258206, 0.73336003, 0.12847191, 0.81480474,
0.04483691, 0.92858125, 0.18957235, 1, 0, 0.88585881, 0.1907184,
0.73123909, 0.16473694, 0.77447817, 0.07729033, 0.89243752,
0.18318066
],
[
1, 0, 0.87284429, 0.20949231, 0.72155559, 0.17678271, 0.79565621,
0.11923267, 0.88682803, 0.29206145, 1, 0, 0.87947901, 0.22133857,
0.71546402, 0.21813219, 0.75051113, 0.15777976, 0.84361165,
0.29583392
],
[
1, 0, 0.87774496, 0.18258206, 0.73336003, 0.12847191, 0.81480474,
0.04483691, 0.92858125, 0.18957235, 1, 0, 0.88585881, 0.1907184,
0.73123909, 0.16473694, 0.77447817, 0.07729033, 0.89243752,
0.18318066
],
[
1, 0, 0.87766593, 0.1933058, 0.73318096, 0.14998893, 0.81446235,
0.07760536, 0.92757334, 0.23415394, 1, 0, 0.88576381, 0.20334594,
0.73100612, 0.19011774, 0.77401683, 0.11605133, 0.8910841,
0.23619084
],
[
1, 0, 0.87663525, 0.1762012, 0.73095019, 0.11394647, 0.81154927,
0.02262734, 0.92156405, 0.15929394, 1, 0, 0.88447675, 0.18333647,
0.72823205, 0.1474896, 0.77106314, 0.05022246, 0.88602628,
0.14634493
],
[
1, 0, 0.87663525, 0.1762012, 0.73095019, 0.11394647, 0.81154927,
0.02262734, 0.92156405, 0.15929394, 1, 0, 0.88447675, 0.18333647,
0.72823205, 0.1474896, 0.77106314, 0.05022246, 0.88602628,
0.14634493
],
[
1, 0, 0.8796893, 0.15611312, 0.73895622, 0.07680882, 0.82548327,
-0.03400087, 0.95047066, 0.0820154, 1, 0, 0.88862453, 0.16028382,
0.7394665, 0.10587372, 0.78983456, -0.01201057, 0.92257343,
0.06079913
]])
# fmt: on
ans = mlalgo.graph_wave(
A_sparse,
[0.5, 0.6],
1.7099629140148158,
30,
sparsify_array(np.arange(0, 10, 2, dtype=np.float64), [True]),
)
assert np.all(np.isclose(ans, expected_ans))
def test_graph_sage():
# fmt: off
A = np.array([
[0, 1, 0, 1, 1, 0, 0, 0],
[1, 0, 0, 1, 1, 0, 0, 0],
[0, 0, 0, 0, 1, 1, 1, 1],
[1, 1, 0, 0, 1, 0, 0, 0],
[1, 1, 1, 1, 0, 0, 0, 0],
[0, 0, 1, 0, 0, 0, 1, 0],
[0, 0, 1, 0, 0, 1, 0, 0],
[0, 0, 1, 0, 0, 0, 0, 0],
],
dtype=np.float64)
A = sparsify_array(A, [False, True])
_num_nodes = A.shape[0]
_dims = [7, 11, 13]
# features = np.random.rand(_num_nodes, _dims[0])
features = np.array([
[
0.39108972, 0.5012513, 0.89723081, 0.27099058, 0.32275578,
0.49175124, 0.92462542
],
[
0.53783469, 0.9879627, 0.70174826, 0.93854928, 0.30000489,
0.63668296, 0.44688698
],
[
0.69605861, 0.27255265, 0.35099912, 0.55982677, 0.4273765,
0.71363344, 0.06490641
],
[
0.77244266, 0.88421098, 0.93914428, 0.0526303, 0.79569921,
0.86189287, 0.4751131
],
[
0.02654776, 0.7612383, 0.51814711, 0.08852528, 0.33581081,
0.34977708, 0.80279946
],
[
0.97218887, 0.8329417, 0.20214344, 0.39938109, 0.69338518,
0.37085116, 0.34679646
],
[
0.96758692, 0.19503829, 0.72639853, 0.16756625, 0.76714647,
0.16902978, 0.32243089
],
[
0.86073823, 0.75369185, 0.83310417, 0.97361141, 0.97555348,
0.07196352, 0.3626788
],
],
dtype=np.float64)
features = sparsify_array(features, [False, True])
# weight_matrices = [np.random.rand(2 * prev_dim, dim) for prev_dim, dim in zip(_dims, _dims[1:])]
# weight_matrices = [weight_matrix - np.mean(weight_matrix) for weight_matrix in weight_matrices]
weight_matrices = [
np.array([
0.44849969, -0.46665052, 0.38610724, -0.31954415, -0.31147248,
0.44301996, -0.17798382, 0.40529587, 0.05003699, 0.34199498,
0.01262532, -0.09585784, -0.00561395, -0.00705517, -0.09003662,
0.1275429, -0.22295171, -0.10000134, 0.39103172, 0.44716378,
0.39686884, -0.34738009, -0.19950878, 0.29104661, 0.11583818,
-0.29612347, -0.39153596, -0.46010402, 0.52668431, 0.40351219,
0.22675305, 0.4982137, -0.07524554, -0.12946731, 0.44637269,
-0.05820939, -0.30486384, -0.16480252, 0.03336091, 0.09315271,
-0.05037402, 0.17064608, -0.13781243, -0.28559323, -0.19857246,
0.23892015, -0.00151305, -0.01174612, -0.43757934, 0.14200969,
0.08152132, -0.18876494, -0.25516716, -0.11913041, -0.34402115,
-0.44478335, -0.03456947, 0.16198755, 0.22278569, -0.19652664,
-0.08282499, 0.30166882, -0.01572978, 0.10788459, 0.41259363,
0.1004477, -0.38695633, 0.22499054, -0.42970007, -0.05014973,
0.16016598, 0.38971812, -0.22680452, 0.44080774, -0.2054676,
-0.44422513, -0.18831138, 0.18761742, -0.2546238, -0.32898005,
-0.30363691, 0.51678939, 0.34565312, -0.18908647, -0.28406102,
0.10069261, 0.13913278, -0.13931456, 0.26840736, -0.26450718,
-0.09599091, 0.18030395, 0.07797712, -0.39424769, -0.22933512,
0.12896401, 0.21418343, -0.02390825, -0.43637196, 0.37204584,
-0.08008971, 0.50665893, 0.15342747, 0.09842373, -0.25118726,
0.49232265, -0.25966891, -0.18288796, 0.51514014, -0.29988576,
-0.28171506, -0.11972695, -0.1116426, 0.22407577, 0.07196866,
0.01189061, -0.39764054, 0.0846052, -0.01040355, 0.50315026,
-0.2239747, -0.1650933, -0.3489139, 0.4502217, 0.14315217,
-0.13754185, -0.4372288, -0.4174219, 0.23195491, -0.31930873,
0.00634386, -0.30075998, 0.31291442, -0.15781314, -0.03668325,
-0.01741184, 0.51845429, -0.19204506, -0.25458102, -0.29754166,
0.48897955, 0.39761734, 0.50640932, 0.36440904, -0.13967416,
-0.0590627, -0.14178824, -0.38033831, -0.06993286, -0.03272309,
-0.08665377, -0.37797494, 0.22340132, 0.41555362
]).reshape(14, 11),
np.array([
-0.22950825, 0.47356333, -0.23424293, -0.29770815, -0.34278594,
0.00926331, -0.15869387, -0.01557874, 0.47950994, -0.11531171,
0.50282882, 0.15465202, -0.13093073, 0.07591561, 0.08070483,
-0.2921942, -0.02377975, 0.37706882, 0.2608012, -0.13428392,
-0.08843151, 0.48539093, 0.19705366, 0.08918976, 0.08232219,
-0.16436965, -0.2884586, -0.39956644, 0.50318162, 0.37740941,
-0.01683036, -0.06273827, 0.35301233, 0.04761192, 0.12480188,
0.42263395, -0.38179371, 0.1269861, 0.03290487, -0.01790114,
-0.13740774, 0.11452844, 0.18328129, 0.28677584, -0.30561157,
-0.24498321, -0.34226796, 0.12621682, -0.35308995, -0.43483672,
-0.48333741, 0.41068243, 0.17499106, 0.4670842, -0.42290461,
0.17553893, -0.47494292, -0.43169526, 0.14721555, 0.08425418,
-0.18527037, 0.49794864, 0.19509773, 0.4071933, 0.03311219,
0.00488919, -0.41919659, 0.36778391, -0.04703998, 0.0398842,
-0.28625282, 0.41089371, -0.31845561, -0.15968473, -0.37156018,
0.28574549, 0.22923943, -0.28373254, 0.39101841, 0.18762113,
0.41920602, -0.13812974, 0.46620671, -0.36662375, -0.32503651,
-0.33775304, -0.22632346, -0.1268525, -0.48116324, 0.10788509,
-0.37015706, 0.0088477, -0.45482819, 0.50917172, 0.44409586,
-0.41619625, 0.1047773, 0.3254928, -0.18503631, 0.15787566,
0.40661986, -0.28784059, -0.40460377, 0.27253348, -0.27261771,
-0.1719656, 0.02250423, -0.42453753, 0.14162989, -0.20185391,
-0.27615418, 0.1258827, -0.36205247, -0.45835454, -0.30714659,
-0.0173512, 0.41241587, -0.42408194, 0.29607361, 0.14064099,
0.14000264, -0.34832738, -0.35163151, 0.16381133, -0.11631709,
-0.21180209, -0.18789508, 0.04168156, 0.33705247, 0.3608453,
-0.29178354, -0.26484368, 0.47700953, -0.06947477, -0.01467548,
-0.3786903, 0.43221556, -0.05880655, 0.47559545, -0.3807028,
0.19975792, 0.16071837, -0.38125686, -0.30645349, 0.01190725,
0.29984574, 0.03564541, -0.12862994, 0.14654883, -0.48636503,
-0.4102408, 0.48018389, -0.02150764, 0.34839799, -0.24232278,
0.23520199, -0.07027585, 0.00201372, -0.33555447, -0.03277968,
-0.15786746, 0.49581866, 0.12140629, 0.0545428, -0.06205085,
0.28415453, -0.05148018, 0.1314153, 0.2758047, -0.22779641,
0.21627012, -0.39168501, 0.10099208, -0.39645095, -0.16575748,
0.21427745, -0.48450068, 0.14938831, -0.45438177, -0.10204422,
-0.01765876, 0.37052772, -0.18384067, 0.08190951, 0.15128443,
0.06960123, -0.16301708, 0.03613785, 0.38059623, 0.05107006,
-0.42448801, 0.2226469, -0.33395037, 0.39704235, -0.19176086,
0.13343548, 0.32719678, 0.43761432, 0.44786207, -0.25254014,
-0.05925346, 0.40904021, -0.38186733, 0.31565752, -0.08247288,
0.30429126, -0.21306066, -0.12471994, -0.23898239, 0.08762914,
0.36601944, 0.12059623, -0.4629562, 0.4883836, 0.32175003,
0.0314467, 0.20712602, -0.06729656, 0.35388646, 0.30054174,
0.32922542, -0.32104118, 0.08844141, -0.22243073, -0.09338935,
-0.08056523, -0.34228469, -0.24185857, 0.0936732, 0.38480919,
0.49225913, -0.28794679, 0.45533206, 0.22958076, -0.2348411,
-0.20642301, 0.23596382, 0.02872514, -0.33251817, 0.17174844,
-0.03348297, -0.32859607, 0.10901442, -0.33836711, -0.16835414,
0.20867398, -0.19857043, -0.16663052, 0.07825061, -0.37368182,
-0.34076121, 0.40861486, -0.201939, -0.44015233, -0.29720221,
0.16881501, -0.19469514, 0.0423145, 0.34634573, 0.05801634,
0.35696215, 0.05604903, -0.11479683, -0.45094242, -0.40379434,
0.50817091, 0.18003263, -0.24748827, -0.25429576, 0.09361117,
-0.48118312, 0.32919184, 0.1726551, 0.50909515, 0.2381023,
0.24317162, -0.39750715, 0.08456536, 0.43103512, 0.21821163,
-0.16936465, 0.19771295, 0.06517018, -0.28561315, -0.05104375,
-0.04098526
]).reshape(22, 13)
]
weight_matrices = [
sparsify_array(weight_matrix, [False, True])
for weight_matrix in weight_matrices
]
expected = np.array([
[
0., 0.07379483, 0.26991188, 0., 0., 0.12508786, 0.34639293, 0., 0.,
0., 0., 0.55470501, 0.69164241
],
[
0., 0.07835575, 0.31145318, 0., 0., 0.0911438, 0.31115009, 0., 0.,
0., 0., 0.62758255, 0.63077402
],
[
0., 0.16404006, 0.23516657, 0., 0., 0.10430948, 0.33573165, 0., 0.,
0., 0., 0.69423177, 0.55877865
],
[
0., 0.06955507, 0.34183008, 0., 0., 0.09311781, 0.33595258, 0., 0.,
0., 0., 0.62122617, 0.60898052
],
[
0., 0.2550802, 0.01022494, 0., 0., 0., 0.2638225, 0., 0., 0., 0.,
0.64477487, 0.6704421
],
[
0., 0.10571722, 0.24756088, 0., 0., 0.10241749, 0.34497212, 0., 0.,
0., 0., 0.56359365, 0.69311223
],
[
0., 0.18235096, 0.17561458, 0., 0., 0.12238869, 0.33438618, 0., 0.,
0., 0., 0.55062202, 0.71128752
],
[
0., 0.08043667, 0.16902664, 0., 0., 0.07887828, 0.30559865, 0., 0.,
0., 0., 0.44908765, 0.81465815
],
],
dtype=np.float64)
# fmt: on
sample_count_per_layer = [999, 999]
rng_context = ChooseUniformContext(seed=1234)
ans = mlalgo.graph_sage(
A,
features,
weight_matrices,
sample_count_per_layer,
rng_context,
).toarray()
assert np.isclose(expected, ans).all()
def test_ir_select_random_weighted(engine: MlirJitEngine, aliases: AliasMap):
# Build Function
ir_builder = MLIRFunctionBuilder(
"test_select_random_weighted",
input_types=["tensor<?x?xf64, #CSR64>", "i64", "!llvm.ptr<i8>"],
return_types=["tensor<?x?xf64, #CSR64>"],
aliases=aliases,
)
M, n, context = ir_builder.inputs
filtered = ir_builder.graphblas.matrix_select_random(
M, n, context, choose_n="choose_weighted")
ir_builder.return_vars(filtered)
test_select_random_weighted = ir_builder.compile(engine=engine,
passes=GRAPHBLAS_PASSES)
# Test Results
# for weighted sampling to make sense, weights must all be >= 0
dense_input_tensor = np.array(
[
[1, 0, 0, 0, 0],
[1, 2, 4, 0, 0], # using this row for stats check below
[0, 0, 1, 100, 1],
[0, 0, 5, 6, 0],
[0, 0, 0, 1, 0],
],
dtype=np.float64,
)
input_tensor = sparsify_array(dense_input_tensor, [False, True])
assert input_tensor.verify()
# basic checks
rng = ChooseWeightedContext(seed=2)
result = test_select_random_weighted(input_tensor, 2, rng)
assert result.verify()
dense_result = result.toarray()
expected_row_count = np.minimum((dense_input_tensor != 0).sum(axis=1), 2)
actual_row_count = (dense_result != 0).sum(axis=1)
np.testing.assert_equal(expected_row_count, actual_row_count)
# rough statistical check of row 1
counts = defaultdict(lambda: 0)
n = 100
for i in range(n):
result = test_select_random_weighted(input_tensor, 1, rng)
assert result.verify()
dense_result = result.toarray()
choice = np.argmax(dense_result[1])
counts[choice] += 1
assert sorted(counts.keys()) == [0, 1, 2]
row_1 = dense_input_tensor[1]
row_1_sum = row_1.sum()
print(counts)
for key, actual_count in counts.items():
prob = row_1[key] / row_1_sum
expected_count = prob * n
# binomial standard deviation
stddev = (n * prob * (1 - prob))**0.5
assert abs(expected_count - actual_count) < (
2 * stddev
), f"key: {key}, expected: {expected_count}, actual: {actual_count}, stdev: {stddev}"
def test_ir_diag(
matrix_type_template: str,
vector_type_template: str,
mlir_type: str,
engine: MlirJitEngine,
aliases: AliasMap,
):
matrix_type = matrix_type_template.format(scalar_type=mlir_type)
vector_type = vector_type_template.format(scalar_type=mlir_type)
np_type = MLIR_TYPE_TO_NP_TYPE[mlir_type]
# Build Functions
ir_builder = MLIRFunctionBuilder(
f"diag_func_{mlir_type}",
input_types=[vector_type, matrix_type],
return_types=[
matrix_type,
vector_type,
],
aliases=aliases,
)
(input_vector, input_matrix) = ir_builder.inputs
output_matrix = ir_builder.graphblas.diag(input_vector, matrix_type)
output_vector = ir_builder.graphblas.diag(input_matrix, vector_type)
ir_builder.return_vars(output_matrix, output_vector)
diag_func = ir_builder.compile(engine=engine, passes=GRAPHBLAS_PASSES)
# Test Results
dense_input_vector = np.array(
[0, 0, 0, 1, 0, -2, 0, 0],
dtype=np_type,
)
input_vector = sparsify_array(dense_input_vector, [True])
assert input_vector.verify()
dense_input_matrix = np.array(
[
[0, 7, 7, 0, 7],
[0, 1, 7, 0, 0],
[0, 1, 0, 7, 0],
[0, 7, 0, 2, 0],
[7, 7, 0, 0, 0],
],
dtype=np_type,
)
input_matrix = sparsify_array(dense_input_matrix, [False, True])
assert input_matrix.verify()
matrix_type_is_csc = [1, 0] == SparseTensorType.parse(
matrix_type, aliases).encoding.ordering
if matrix_type_is_csc:
input_matrix = engine.csr_to_csc(input_matrix)
output_matrix, output_vector = diag_func(input_vector, input_matrix)
assert output_matrix.verify()
assert output_vector.verify()
output_matrix = output_matrix.toarray()
output_vector = output_vector.toarray()
expected_output_matrix = np.diagflat(dense_input_vector)
expected_output_vector = np.diag(dense_input_matrix)
assert np.all(expected_output_matrix == output_matrix)
assert np.all(expected_output_vector == output_vector)
def test_ir_reduce_to_vector(
input_type_template: str,
reduce_rows_output_type_template: str,
reduce_columns_output_type_template: str,
mlir_type: str,
engine: MlirJitEngine,
aliases: AliasMap,
):
input_type = input_type_template.format(scalar_type=mlir_type)
reduce_rows_output_type = reduce_rows_output_type_template.format(
scalar_type=mlir_type)
reduce_columns_output_type = reduce_columns_output_type_template.format(
scalar_type=mlir_type)
np_type = MLIR_TYPE_TO_NP_TYPE[mlir_type]
# Build Functions
ir_builder = MLIRFunctionBuilder(
f"reduce_func_{mlir_type}",
input_types=[input_type],
return_types=[
reduce_rows_output_type,
"tensor<?xi64, #CV64>",
reduce_rows_output_type,
"tensor<?xi64, #CV64>",
"tensor<?xi64, #CV64>",
"tensor<?xi64, #CV64>",
],
aliases=aliases,
)
(matrix, ) = ir_builder.inputs
reduced_rows = ir_builder.graphblas.reduce_to_vector(matrix, "plus", 1)
reduced_columns = ir_builder.graphblas.reduce_to_vector(matrix, "count", 0)
zero_scalar = ir_builder.arith.constant(0, mlir_type)
reduced_rows_clamped = ir_builder.graphblas.apply(reduced_rows,
"min",
right=zero_scalar)
reduced_rows_clamped = ir_builder.graphblas.apply(reduced_rows_clamped,
"identity")
reduced_columns_abs = ir_builder.graphblas.apply(reduced_columns, "abs")
reduced_columns_abs = ir_builder.graphblas.apply(reduced_columns_abs,
"identity")
reduced_columns_negative_abs = ir_builder.graphblas.apply(
reduced_columns, "ainv")
reduced_columns_negative_abs = ir_builder.graphblas.apply(
reduced_columns_negative_abs, "identity")
reduced_rows_argmin = ir_builder.graphblas.reduce_to_vector(
matrix, "argmin", 1)
reduced_columns_argmax = ir_builder.graphblas.reduce_to_vector(
matrix, "argmax", 0)
ir_builder.return_vars(
reduced_rows,
reduced_columns,
reduced_rows_clamped,
reduced_columns_negative_abs,
reduced_rows_argmin,
reduced_columns_argmax,
)
reduce_func = ir_builder.compile(engine=engine, passes=GRAPHBLAS_PASSES)
# Test Results
dense_input_tensor = np.array(
[
[1, 0, 0, 0],
[-2, 0, 3, -4],
[0, 0, 0, 0],
[0, 0, 5, -6],
[0, -7, 0, 8],
],
dtype=np_type,
)
input_tensor = sparsify_array(dense_input_tensor, [False, True])
input_type_is_csc = [1, 0
] == SparseTensorType.parse(input_type,
aliases).encoding.ordering
if input_type_is_csc:
input_tensor = engine.csr_to_csc(input_tensor)
(
reduced_rows,
reduced_columns,
reduced_rows_clamped,
reduced_columns_negative_abs,
reduced_rows_argmin,
reduced_columns_argmax,
) = reduce_func(input_tensor)
assert reduced_rows.verify()
assert reduced_columns.verify()
assert reduced_rows_clamped.verify()
assert reduced_columns_negative_abs.verify()
assert reduced_rows_argmin.verify()
assert reduced_columns_argmax.verify()
reduced_rows = reduced_rows.toarray()
reduced_columns = reduced_columns.toarray()
reduced_rows_clamped = reduced_rows_clamped.toarray()
reduced_columns_negative_abs = reduced_columns_negative_abs.toarray()
reduced_rows_argmin = reduced_rows_argmin.toarray()
reduced_columns_argmax = reduced_columns_argmax.toarray()
expected_reduced_rows = dense_input_tensor.sum(axis=1)
expected_reduced_columns = (dense_input_tensor.astype(bool).sum(
axis=0).astype(np_type))
expected_reduced_rows_clamped = np.copy(expected_reduced_rows)
expected_reduced_rows_clamped[expected_reduced_rows_clamped > 0] = 0
expected_reduced_columns_negative_abs = -np.abs(expected_reduced_columns)
M = dense_input_tensor.copy()
M[dense_input_tensor == 0] = dense_input_tensor.max() + 1
expected_reduced_rows_argmin = np.argmin(M, axis=1)
M = dense_input_tensor.copy()
M[dense_input_tensor == 0] = dense_input_tensor.min() - 1
expected_reduced_columns_argmax = np.argmax(M, axis=0)
assert np.all(reduced_rows == expected_reduced_rows)
assert np.all(reduced_columns == expected_reduced_columns)
assert np.all(reduced_rows_clamped == expected_reduced_rows_clamped)
assert np.all(
reduced_columns_negative_abs == expected_reduced_columns_negative_abs)
assert np.all(reduced_rows_argmin == expected_reduced_rows_argmin)
assert np.all(reduced_columns_argmax == expected_reduced_columns_argmax)
def test_ir_builder_dnn(
engine: MlirJitEngine,
array_initializer: Callable,
max_num_layers: int,
clamp_threshold: float,
):
sparsify_matrix = lambda matrix: sparsify_array(matrix, [False, True])
np.random.seed(hash(datetime.date.today()) % 2**32)
for num_layers in range(1, max_num_layers + 1):
dense_weight_matrices = [
array_initializer(64) for _ in range(num_layers)
]
dense_weight_matrices = [
m.reshape(8, 8) for m in dense_weight_matrices
]
dense_weight_matrices = [
m.astype(np.float64) for m in dense_weight_matrices
]
dense_bias_vectors = [array_initializer(8) for _ in range(num_layers)]
dense_bias_vectors = [
vec.astype(np.float64) for vec in dense_bias_vectors
]
dense_bias_matrices = [np.diag(vec) for vec in dense_bias_vectors]
dense_bias_matrices = [np.nan_to_num(m) for m in dense_bias_matrices]
dense_input_tensor = array_initializer(64).reshape(8, 8).astype(
np.float64)
numpy_dense_result = dense_input_tensor
for dense_weight_matrix, dense_bias_vector in zip(
dense_weight_matrices, dense_bias_vectors):
numpy_dense_result = numpy_dense_result @ dense_weight_matrix
numpy_dense_result = numpy_dense_result + dense_bias_vector
numpy_dense_result = numpy_dense_result * (
dense_bias_vector != 0).astype(int) # TODO is this correct?
numpy_dense_result = numpy_dense_result * (numpy_dense_result >
0).astype(int)
numpy_dense_result[
numpy_dense_result > clamp_threshold] = clamp_threshold
sparse_weight_matrices = [
sparsify_matrix(matrix) for matrix in dense_weight_matrices
]
assert all(e.verify() for e in sparse_weight_matrices)
sparse_bias_matrices = [
sparsify_matrix(matrix) for matrix in dense_bias_matrices
]
assert all(e.verify() for e in sparse_bias_matrices)
sparse_input_tensor = sparsify_matrix(dense_input_tensor)
assert sparse_input_tensor.verify()
sparse_result = dense_neural_network(
sparse_weight_matrices,
sparse_bias_matrices,
sparse_input_tensor,
clamp_threshold,
)
assert sparse_result.verify()
dense_result = sparse_result.toarray()
with np.printoptions(suppress=True):
assert np.isclose(dense_result, numpy_dense_result).all(), f"""
num_layers
{num_layers}
dense_input_tensor
{dense_input_tensor}
dense_result
{dense_result}
numpy_dense_result
{numpy_dense_result}
np.isclose(dense_result, numpy_dense_result)
{np.isclose(dense_result, numpy_dense_result)}
"""
return
