def test_reshape_like_infer_type():
x = relay.var('x', relay.TensorType((1, 2, 3), 'float32'))
y = relay.var('y', relay.TensorType((1, 6), 'float32'))
z = relay.reshape_like(x, y)
zz = run_infer_type(z)
assert (zz.checked_type == relay.TensorType((1, 6), 'float32'))
(n, c, h, w) = (te.size_var('n'), 2, 3, te.size_var('w'))
x = relay.var('x', relay.TensorType((n, c, h, w), 'float32'))
y = relay.var('y', relay.TensorType((1, 8, 8), 'float32'))
z = relay.reshape_like(x, y)
zz = run_infer_type(z)
assert (zz.checked_type == relay.TensorType((1, 8, 8), 'float32'))
def test_reshape_like_infer_type():
# concrete shape
x = relay.var("x", relay.TensorType((1, 2, 3), "float32"))
y = relay.var("y", relay.TensorType((1,6), "float32"))
z = relay.reshape_like(x, y)
zz = relay.ir_pass.infer_type(z)
assert zz.checked_type == relay.TensorType((1, 6), "float32")
# symbolic shape
n, c, h, w = tvm.var("n"), 2, 3, tvm.var("w")
x = relay.var("x", relay.TensorType((n, c, h, w), "float32"))
y = relay.var("y", relay.TensorType((1, 8, 8), "float32"))
z = relay.reshape_like(x, y)
zz = relay.ir_pass.infer_type(z)
assert zz.checked_type == relay.TensorType((1, 8, 8), "float32")
def test_reshape_like_infer_type():
# concrete shape
x = relay.var("x", relay.TensorType((1, 2, 3), "float32"))
y = relay.var("y", relay.TensorType((1,6), "float32"))
z = relay.reshape_like(x, y)
zz = relay.ir_pass.infer_type(z)
assert zz.checked_type == relay.TensorType((1, 6), "float32")
# symbolic shape
n, c, h, w = tvm.var("n"), 2, 3, tvm.var("w")
x = relay.var("x", relay.TensorType((n, c, h, w), "float32"))
y = relay.var("y", relay.TensorType((1, 8, 8), "float32"))
z = relay.reshape_like(x, y)
zz = relay.ir_pass.infer_type(z)
assert zz.checked_type == relay.TensorType((1, 8, 8), "float32")
def test_concretize_reshape_like_attrs():
data = relay.var("data", shape=(2, 3, 4), dtype="float32")
shape_like = relay.var("shape_like", shape=(6, 2, 2), dtype="float32")
expr = relay.reshape_like(data, shape_like, lhs_begin=2, rhs_begin=1)
expected = run_infer_type(relay.reshape(data, (2, 3, 2, 2)))
actual = run_opt_pass(expr, relay.transform.SimplifyExpr())
assert tvm.ir.structural_equal(actual, expected)
def test_any_reshape_like():
mod = tvm.IRModule()
dtype = "float32"
data = relay.var("data", shape=(relay.Any(), 3, 10), dtype=dtype)
shape_like = relay.var("data", shape=(relay.Any(), 5, 6), dtype=dtype)
y = relay.reshape_like(data, shape_like)
mod["main"] = relay.Function([data, shape_like], y)
data_np = np.random.uniform(size=(3, 3, 10)).astype(dtype)
shape_like_np = np.random.uniform(size=(3, 5, 6)).astype(dtype)
check_result([data_np, shape_like_np], mod, shape_like_np.shape, assert_shape=True)
def test_any_reshape_like():
mod = tvm.IRModule()
dtype = "float32"
data = relay.var('data', shape=(relay.Any(), 3, 10), dtype=dtype)
shape_like = relay.var('data', shape=(relay.Any(), 5, 6), dtype=dtype)
y = relay.reshape_like(data, shape_like)
mod["main"] = relay.Function([data, shape_like], y)
data_np = np.random.uniform(size=(3, 3, 10)).astype(dtype)
shape_like_np = np.random.uniform(size=(3, 5, 6)).astype(dtype)
for kind in ["debug", "vm"]:
ex = relay.create_executor(kind, mod=mod, ctx=tvm.cpu(), target="llvm")
result = ex.evaluate()(data_np, shape_like_np)
assert result.asnumpy().shape == shape_like_np.shape, \
"Shape mismatch: expect %s but got %s." % (str(shape_like_np.shape), str(result.asnumpy().shape))
def verify_reshape_like(shape, oshape):
x_data = np.random.uniform(low=(- 1), high=1, size=shape).astype('float32')
y_data = np.random.uniform(low=(- 1), high=1, size=oshape).astype('float32')
ref_res = np.reshape(x_data, y_data.shape)
x = relay.var('x', relay.TensorType(shape, 'float32'))
y = relay.var('x', relay.TensorType(oshape, 'float32'))
z = relay.reshape_like(x, y)
zz = run_infer_type(z)
assert (zz.checked_type == relay.ty.TensorType(ref_res.shape, 'float32'))
func = relay.Function([x, y], z)
for (target, ctx) in tvm.testing.enabled_targets():
for kind in ['graph', 'debug']:
intrp = relay.create_executor(kind, ctx=ctx, target=target)
op_res = intrp.evaluate(func)(x_data, y_data)
tvm.testing.assert_allclose(op_res.asnumpy(), ref_res, rtol=1e-05)
def verify_reshape_like(shape, oshape):
x_data = np.random.uniform(low=-1, high=1, size=shape).astype("float32")
y_data = np.random.uniform(low=-1, high=1, size=oshape).astype("float32")
ref_res = np.reshape(x_data, y_data.shape)
x = relay.var("x", relay.TensorType(shape, "float32"))
y = relay.var("x", relay.TensorType(oshape, "float32"))
z = relay.reshape_like(x, y)
zz = relay.ir_pass.infer_type(z)
assert zz.checked_type == relay.ty.TensorType(ref_res.shape, "float32")
func = relay.Function([x, y], z)
for target, ctx in ctx_list():
for kind in ["graph", "debug"]:
intrp = relay.create_executor(kind, ctx=ctx, target=target)
op_res = intrp.evaluate(func)(x_data, y_data)
tvm.testing.assert_allclose(op_res.asnumpy(), ref_res, rtol=1e-5)
def verify_reshape_like(shape, oshape):
x_data = np.random.uniform(low=-1, high=1, size=shape).astype("float32")
y_data = np.random.uniform(low=-1, high=1, size=oshape).astype("float32")
ref_res = np.reshape(x_data, y_data.shape)
x = relay.var("x", relay.TensorType(shape, "float32"))
y = relay.var("x", relay.TensorType(oshape, "float32"))
z = relay.reshape_like(x, y)
zz = relay.ir_pass.infer_type(z)
assert zz.checked_type == relay.ty.TensorType(ref_res.shape, "float32")
func = relay.Function([x, y], z)
for target, ctx in ctx_list():
for kind in ["graph", "debug"]:
intrp = relay.create_executor(kind, ctx=ctx, target=target)
op_res = intrp.evaluate(func)(x_data, y_data)
tvm.testing.assert_allclose(op_res.asnumpy(), ref_res, rtol=1e-5)
def test_reshape_like_grad():
data = relay.var("data", shape=(2, 3, 4), dtype="float32")
shape_like = relay.var("shape_like", shape=(6, 2, 2), dtype="float32")
fwd_func = relay.Function([data, shape_like],
relay.reshape_like(data, shape_like))
check_grad(fwd_func)
def get_net(input_shape=(1, 3, 24, 12), dtype="float32", wtype=None):
"""Get synthetic testing network.
Parameters
----------
image_shape : tuple, optional
The input shape as (batch_size, channels, height, width).
dtype : str, optional
The data type for the input.
wtype : str, optional
The data type for weights. Defaults to `dtype`.
Returns
-------
net : relay.Function
The dataflow.
"""
if wtype is None:
wtype = dtype
data = relay.var("data", shape=input_shape, dtype=dtype)
dense_shape = [-1, input_shape[3]]
dense = relay.nn.relu(
relay.nn.dense(
relay.reshape(data, dense_shape),
relay.var("dense_weight",
shape=[input_shape[3], dense_shape[1]],
dtype=wtype),
))
dense = relay.reshape_like(dense, data)
conv_shape = [input_shape[1], input_shape[1], 3, 3]
conv = relay.nn.softmax(
relay.nn.conv2d(
data,
relay.var("conv_weight", shape=conv_shape, dtype=wtype),
padding=1,
kernel_size=3,
))
added = relay.add(dense, conv)
biased = layers.batch_norm_infer(relay.nn.bias_add(
added, relay.var("bias", dtype=wtype)),
name="batch_norm")
dense = relay.nn.relu(
relay.nn.dense(
relay.reshape(biased, dense_shape),
relay.var("dense2_weight",
shape=[input_shape[3], dense_shape[1]],
dtype=wtype),
))
dense = relay.reshape_like(dense, data)
conv = relay.nn.softmax(
relay.nn.conv2d(
biased,
relay.var("conv2_weight", shape=conv_shape, dtype=wtype),
padding=1,
kernel_size=3,
))
added = relay.add(dense, conv)
args = relay.analysis.free_vars(added)
return relay.Function(args, added)
