Fully Connected Neural Network

class e3nn.nn.FullyConnectedNet(hs, act=None, variance_in: int = 1, variance_out: int = 1, out_act: bool = False)

Bases: Sequential

Fully-connected Neural Network

Parameters:

• hs (list of int) – input, internal and output dimensions

• act (function) –

  activation function \(\phi\), it will be automatically normalized by a scaling factor such that

  \[\int_{-\infty}^{\infty} \phi(z)^2 \frac{e^{-z^2/2}}{\sqrt{2\pi}} dz = 1\]