def verify_sparse_to_dense(sparse_indices, sparse_values, default_value, output_shape, xpected):
sparse_indices_data = np.array(sparse_indices)
sparse_values_data = np.array(sparse_values)
default_value_data = np.array(default_value)
a = relay.var("a", relay.TensorType(sparse_indices_data.shape, str(sparse_indices_data.dtype)))
b = relay.var("b", relay.TensorType(sparse_values_data.shape, str(sparse_values_data.dtype)))
if default_value is None:
args = [a, b]
d = relay.sparse_to_dense(a, output_shape, b)
else:
c = relay.var("c", relay.TensorType(default_value_data.shape, str(default_value_data.dtype)))
args = [a, b, c]
d = relay.sparse_to_dense(a, output_shape, b, c)
zz = run_infer_type(d)
assert zz.checked_type == relay.ty.TensorType(output_shape, str(sparse_values_data.dtype))
func = relay.Function(args, d)
for target, ctx in tvm.testing.enabled_targets():
for kind in ["graph", "debug"]:
intrp = relay.create_executor(kind, ctx=ctx, target=target)
if default_value is None:
op_res = intrp.evaluate(func)(sparse_indices_data, sparse_values_data)
else:
op_res = intrp.evaluate(func)(
sparse_indices_data, sparse_values_data, default_value_data
)
tvm.testing.assert_allclose(op_res.asnumpy(), xpected, rtol=1e-5)
def verify_sparse_to_dense(sparse_indices, sparse_values, default_value,
output_shape, xpected):
sparse_indices_data = np.array(sparse_indices)
sparse_values_data = np.array(sparse_values)
default_value_data = np.array(default_value)
output_shape_data = np.array(output_shape)
a = relay.var(
"a",
relay.TensorType(sparse_indices_data.shape,
str(sparse_indices_data.dtype)))
b = relay.var(
"b",
relay.TensorType(sparse_values_data.shape,
str(sparse_values_data.dtype)))
output_shape_var = relay.var(
"output_shape",
relay.TensorType(output_shape_data.shape,
str(output_shape_data.dtype)))
if default_value is None:
args = [a, b, output_shape_var]
d = relay.sparse_to_dense(a, output_shape_var, b)
else:
c = relay.var(
"c",
relay.TensorType(default_value_data.shape,
str(default_value_data.dtype)))
args = [a, b, c, output_shape_var]
d = relay.sparse_to_dense(a, output_shape_var, b, c)
zz = run_infer_type(d)
assert len(zz.checked_type.shape) == len(output_shape)
func = relay.Function(args, d)
if default_value is None:
arguments = [
sparse_indices_data, sparse_values_data, output_shape_data
]
else:
arguments = [
sparse_indices_data,
sparse_values_data,
default_value_data,
output_shape_data,
]
verify_func(executor_kind, func, arguments, xpected)
def verify_sparse_to_dense(sparse_indices, sparse_values, default_value,
output_shape, xpected):
sparse_indices_data = np.array(sparse_indices)
sparse_values_data = np.array(sparse_values)
default_value_data = np.array(default_value)
output_shape_data = np.array(output_shape)
a = relay.var(
"a",
relay.TensorType(sparse_indices_data.shape,
str(sparse_indices_data.dtype)))
b = relay.var(
"b",
relay.TensorType(sparse_values_data.shape,
str(sparse_values_data.dtype)))
output_shape_const = relay.const(output_shape_data)
if default_value is None:
args = [a, b]
d = relay.sparse_to_dense(a, output_shape_const, b)
else:
c = relay.var(
"c",
relay.TensorType(default_value_data.shape,
str(default_value_data.dtype)))
args = [a, b, c]
d = relay.sparse_to_dense(a, output_shape_const, b, c)
zz = run_infer_type(d)
assert len(zz.checked_type.shape) == len(output_shape)
func = relay.Function(args, d)
func2 = run_opt_pass(run_opt_pass(func, transform.DynamicToStatic()),
transform.InferType())
assert isinstance(func2.body, relay.Call)
assert func2.body.op == relay.op.get("sparse_to_dense")
if default_value is None:
arguments = [sparse_indices_data, sparse_values_data]
else:
arguments = [
sparse_indices_data, sparse_values_data, default_value_data
]
verify_func(func2, arguments, xpected)
