Input shape: B x L x D

x: (batch, lengths..., dim)

During training, call this function twice to compute closed-loop and open-loop -> minimize the closed-loop?

Initialize weights for differencing kernel - Assume u is shape B x D x L

Initialize weights for differencing kernel - Assume u is shape B x D x L

Constant matrix A A : (..., N, N) b : (..., N) c : (..., N) Returns if c: x : (..., L) x[i, l] = c[i] @ A^l @ b[i]

Specializes to lower triangular Toeplitz matrix A represented by its diagonals A : (..., N) b : (..., N) c : (..., N) Returns x

Padded version of krylov_toeplitz that saves some fft's TODO currently not faster than original version, not sure why

Compute A^L and the scan sum_i A^i v_i A: (..., N, N) v: (..., N, L)

Prints content of DictConfig using Rich library and its tree structure. Args: config (DictConfig): Configuration composed by Omegaconf

A version of X that's complex (i.e., last dimension is 2). Parameters: X: (...) tensor Return: X_complex: (..., 2) tensor

Convert Sequence or Mapping object to dict lists get converted to {0: x[0], 1: x[1], ...}

Convert an object to list. If Sequence (e.g. list, tuple, Listconfig): just return it Special case: If non-recursive and not a list, wrap in

Small function to trigger the compilation of a pykeops kernel Used in scenarios where we must manually control compilation, e.g. the multi-gpu c