Forward pass

# This is only valid when the package is not installed
import sys
sys.path.append('../../') # two folders up
import DeepINN as dp
import torch

Using default backend: PyTorch

Using Pytorch:  2.0.1+cu117

activation = "tanh"
initialiser = "Xavier normal"
layer_size = [1] + [5] * 3 + [1]

net = dp.nn.FullyConnected(layer_size, activation, initialiser)
net.linears

ModuleList(
  (0): Linear(in_features=1, out_features=5, bias=True)
  (1-2): 2 x Linear(in_features=5, out_features=5, bias=True)
  (3): Linear(in_features=5, out_features=1, bias=True)
)

# net.linears[2].weight.data

# net.initialiser

input = torch.randn(3, 1)
input

tensor([[-1.0178],
        [-1.9623],
        [-0.3584]])

forward = net.forward(input)
forward

tensor([[0.0563],
        [0.0000],
        [0.0799]], grad_fn=<AddmmBackward0>)

net.activation(input)

tensor([[-0.7690],
        [-0.9613],
        [-0.3438]])

loss_metric = dp.backend.loss_metric("MSE")
loss_metric

MSELoss()

loss_metric(input, forward)

tensor(1.7321, grad_fn=<MseLossBackward0>)