def _test_pairwise_distance(metric):
# Simulate some data, N.B., oriented such that we want to compute
# distance between columns.
data = np.random.randint(low=0, high=3, size=(100, 10), dtype=np.int8)
# Compute expected result, using scipy as reference implementation.
expect = spd.pdist(data.T, metric=metric)
# Test numpy array.
actual = pairwise_distance(data, metric=metric)
assert isinstance(actual, np.ndarray)
assert_allclose(expect, actual)
assert actual.dtype.kind == "f"
# Test cuda array.
data_cuda = cuda.to_device(data)
actual = pairwise_distance(data_cuda, metric=metric)
assert isinstance(actual, type(data_cuda))
assert_allclose(expect, actual.copy_to_host())
assert actual.dtype.kind == "f"
# Test dask array.
data_dask = da.from_array(data, chunks=(10, 5))
actual = pairwise_distance(data_dask, metric=metric)
assert isinstance(actual, da.Array)
ac = actual.compute(scheduler="single-threaded")
assert_allclose(expect, ac)
assert actual.dtype.kind == "f"
# Test dask array with cuda.
data_dask_cuda = data_dask.rechunk((10, -1)).map_blocks(cuda.to_device)
actual = pairwise_distance(data_dask_cuda, metric=metric)
assert isinstance(actual, da.Array)
ac = actual.compute(scheduler="single-threaded")
assert_allclose(expect, ac)
assert actual.dtype.kind == "f"
# Test zarr array.
data_zarr = zarr.array(data, chunks=(10, 5))
actual = pairwise_distance(data_zarr, metric=metric)
assert isinstance(actual, da.Array)
assert_allclose(expect, actual.compute())
assert actual.dtype.kind == "f"
def test_not_implemented():
data = np.random.randint(low=0, high=3, size=(100, 10), dtype=np.int8)
with pytest.raises(ValueError):
pairwise_distance(data, metric="foo")
data_dask = da.from_array(data, chunks=(10, 5))
with pytest.raises(ValueError):
pairwise_distance(data_dask, metric="foo")
data_cuda = cuda.to_device(data)
with pytest.raises(ValueError):
pairwise_distance(data_cuda, metric="foo")
def time_jaccard(self):
pairwise_distance(self.data_cuda, metric="jaccard")
cuda.synchronize()
def time_hamming(self):
pairwise_distance(self.data_cuda, metric="hamming")
cuda.synchronize()
def time_euclidean(self):
pairwise_distance(self.data_cuda, metric="euclidean")
cuda.synchronize()
def time_cityblock(self):
pairwise_distance(self.data_cuda, metric="cityblock")
cuda.synchronize()
def time_jaccard(self):
pairwise_distance(self.data_dask, metric="jaccard").compute()
def time_hamming(self):
pairwise_distance(self.data_dask, metric="hamming").compute()
def time_euclidean(self):
pairwise_distance(self.data_dask, metric="euclidean").compute()
def time_cityblock(self):
pairwise_distance(self.data_dask, metric="cityblock").compute()
def time_jaccard(self):
pairwise_distance(self.data, metric="jaccard")
def time_hamming(self):
pairwise_distance(self.data, metric="hamming")
def time_euclidean(self):
pairwise_distance(self.data, metric="euclidean")
def time_cityblock(self):
pairwise_distance(self.data, metric="cityblock")
