pmlayer.torch.layers.HLattice

Hierarchical lattice layer with partially monotone constraints for PyTorch.

class HLattice(
    num_input_dims,
    lattice_sizes,
    indices_increasing,
    neural_network=None)

Parameters

Args	Type	Description
num_input_dims	int	The length of input feature vectors.
lattice_sizes	Tensor (long)	The latice sizes of monotonically increasing features. Each entry corresponds to a monotonically increasing feature, and each entry must be at least 2.
indices_increasing	list of int	The list of indices of monotonically increasing features.
neural_network	neural network	Default: None. By setting this parameter, the sub-network internally used in this layer can be replaced.

Tensor Shape

I/O	Shape
Input	(N, num_input_dims)
Output	(N, 1)

N usually corresponds the batch size.

Note

Each value in the input tensor must be between 0 and 1.

Example

The following code transforms input tensor x into output tensor y, and the output (the shape of y) is (128,1).

sizes = torch.tensor([4,4], dtype=torch.long)
l = HLattice(10, sizes, [2,3])
x = torch.randn(128, 10)
y = l(x)
print(y.shape)