def _check_input(self) -> None:
"""Assert the input input tensor is 4 dimensional."""
check_tensor(self.x, shape_dim=4)
if self.x.shape[1] < self.height:
raise (
f"{self.name}: Input heights must be higher or equal to filter heights."
f"Input heights {self.x.shape[1]}, filter heights {self.height}"
)
if self.x.shape[2] < self.width:
raise (
f"{self.name}: Input heights must be higher or equal to filter heights."
f"Input heights {self.x.shape[2]}, filter heights {self.width}"
)
def check_input(self) -> None:
"""Check all input tensor types"""
check_tensor(self.y)
check_tensor(self.x_out)
[check_variable(w) for w in self.weights]
def _check_input(self) -> None:
"""Assert the input input tensor is 2 dimensional."""
check_tensor(self.x, shape_dim=2)
def _check_input(self) -> None:
"""Check if both input respect all dimension conditions."""
check_tensor(self.x, shape_dim=4)
check_tensor(self.x_lag, shape_dim=4)
if self.x.shape[3] < self.x_lag.shape[3]:
raise TypeError(
f"{self.name}: x must have a number of channel higher or equal than x_lag. "
f"Channel for x is {self.x.shape[3].value} whereas is {self.x_lag.shape[3].value} "
"for x_lag.")
if self.x.shape[1] > self.x_lag.shape[1]:
raise TypeError(
f"{self.name}: x must have a height lower or equal than x_lag. "
f"Height for x is {self.x.shape[1].value} whereas is {self.x_lag.shape[1].value}"
" for x_lag.")
elif self.x.shape[1] < self.x_lag.shape[1]:
if any([
x is None for x in
[self.pool.height, self.pool.stride, self.pool.padding]
]):
raise TypeError(
"x_lag reduction needed. You must feed value for height, stride and padding. "
)
dilation = 0 if self.pool.dilation is None else self.pool.dilation[
0]
new_dim = get_dim_reduction(self.x_lag.shape[1].value,
self.pool.height, dilation,
self.pool.stride[0], self.pool.padding)
if new_dim != self.x.shape[1]:
raise TypeError(
f"{self.name}: After transformation height of x must be equal than x_lag. "
f"Height for x_lag becomes {new_dim} whereas is {self.x.shape[1].value} for x."
)
if self.x.shape[2] > self.x_lag.shape[2]:
raise TypeError(
f"{self.name}: x must have width lower or equal than x_lag. "
f"Width for x is {self.x.shape[2].value} whereas is {self.x_lag.shape[2].value} for x_lag."
)
elif self.x.shape[2] < self.x_lag.shape[2]:
if any([
x is None for x in
[self.pool.width, self.pool.stride, self.pool.padding]
]):
raise TypeError(
"x_lag reduction needed. You must feed value for width, stride and padding. "
)
dilation = 0 if self.pool.dilation is None else self.pool.dilation[
1]
new_dim = get_dim_reduction(self.x_lag.shape[2].value,
self.pool.width, dilation,
self.pool.stride[1], self.pool.padding)
if new_dim != self.x.shape[2]:
raise TypeError(
f"{self.name}: After transformation width of x must be equal than x_lag. "
f"Width for x_lag becomes {new_dim} whereas is {self.x.shape[2].value} for x."
)