pathint_torch.nn.ControlNet

class pathint_torch.nn.ControlNet(x_size, get_score_mu, T, L_max, emb_dim, emb_hidden_widths, hidden_widths, norm_layer=None, activation_layer=<class 'torch.nn.modules.activation.LeakyReLU'>, scalar_coeff_net=False, dropout=0.0)

Bases: Module

Affine transformation of score parametrized by two neural networks, using the architecture close to the one from the path integral sampler paper. This is of the form \(a_\theta(t, x) + b_\theta(t, x) \nabla \log \mu(x)\), with the output initialized to zero using an overall learn multiplicative factor.

forward(t, x)

Runs the network.

Return type

Tensor

training: bool