You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
54 lines
1.6 KiB
54 lines
1.6 KiB
from typing import Callable, Optional, Tuple, TypeVar
|
|
|
|
from ..config import registry
|
|
from ..model import Model
|
|
from ..types import ArrayXd
|
|
from ..util import get_width
|
|
|
|
LayerT = TypeVar("LayerT")
|
|
SimT = TypeVar("SimT")
|
|
InT = Tuple[LayerT, LayerT]
|
|
OutT = TypeVar("OutT", bound=ArrayXd)
|
|
|
|
|
|
@registry.layers("siamese.v1")
|
|
def siamese(
|
|
layer: Model[LayerT, SimT], similarity: Model[Tuple[SimT, SimT], OutT]
|
|
) -> Model[InT, OutT]:
|
|
return Model(
|
|
f"siamese({layer.name}, {similarity.name})",
|
|
forward,
|
|
init=init,
|
|
layers=[layer, similarity],
|
|
dims={"nI": layer.get_dim("nI"), "nO": similarity.get_dim("nO")},
|
|
)
|
|
|
|
|
|
def forward(
|
|
model: Model[InT, OutT], X1_X2: InT, is_train: bool
|
|
) -> Tuple[OutT, Callable]:
|
|
X1, X2 = X1_X2
|
|
vec1, bp_vec1 = model.layers[0](X1, is_train)
|
|
vec2, bp_vec2 = model.layers[0](X2, is_train)
|
|
output, bp_output = model.layers[1]((vec1, vec2), is_train)
|
|
|
|
def finish_update(d_output: OutT) -> InT:
|
|
d_vec1, d_vec2 = bp_output(d_output)
|
|
d_input1 = bp_vec1(d_vec1)
|
|
d_input2 = bp_vec2(d_vec2)
|
|
return (d_input1, d_input2)
|
|
|
|
return output, finish_update
|
|
|
|
|
|
def init(
|
|
model: Model[InT, OutT], X: Optional[InT] = None, Y: Optional[OutT] = None
|
|
) -> None:
|
|
if X is not None:
|
|
model.layers[0].set_dim("nI", get_width(X[1]))
|
|
model.layers[0].initialize(X=X[0])
|
|
X = (model.layers[0].predict(X[0]), model.layers[0].predict(X[1]))
|
|
model.layers[1].initialize(X=X, Y=Y)
|
|
model.set_dim("nI", model.layers[0].get_dim("nI"))
|
|
model.set_dim("nO", model.layers[1].get_dim("nO"))
|