EFTTOptions

EFTTOptions(
    self,
    num_error_samples: int = 1000,
    fibre_method: str = 'random',
    tol_svd: float = 1e-12,
    num_aca: int = 50,
    tol_aca: float = 1e-10,
    max_fibres: int = 30,
    num_snapshots: int = 30,
)

Options for configuring the construction of an EFTT object.

Parameters

num_error_samples : int = 1000

The number of samples to use when estimating the \(L^{2}\) error of the FTT approximation to the target function at each iteration.

fibre_method : str = 'random'

The method used to compute a set of mode-\(k\) fibres in each dimension \(k \in \{1, \dots, d\}\). This can be "aca" (apply adaptive cross approximation, as in Strössner, Sun, and Kressner 2024), or "random" (choose a set of fibres at random).

tol_svd : float = 1e-12

The threshold to use when applying truncated SVD to compute an (approximate) orthogonal basis for the mode-\(k\) fibres in each dimension. The minimum number of singular values such that their sum exceeds (\(1-\) tol_svd) will be retained.

num_aca : int = 50

If fibre_method="aca", the number of elements of the fibre matrix to sample at each iteration when selecting a new pivot element.

tol_aca : float = 1e-10

If fibre_method="aca", the stopping tolerance, \(\epsilon\), to use. More concretely, if \(\mathcal{S}\) denotes a set of randomly-sampled indices of the mode-\(k\) fibre matrix \(\boldsymbol{M}\) (and \(\mathcal{I}\) and \(\mathcal{J}\) denote the current sets of row and column indices), the iteration is considered finished when \[ \max_{(i, j) \in \mathcal{S}}\|R_{ij}\| < \epsilon, \] where the residual matrix \(\boldsymbol{R}\) is given by \[ \boldsymbol{R} = \boldsymbol{M} - \boldsymbol{M}[:, \mathcal{J}] \boldsymbol{M}[\mathcal{I}, \mathcal{J}]^{-1} \boldsymbol{M}[\mathcal{I}, :]. \]

max_fibres : int = 30

If fibre_method="aca", the maximum number of fibres to sample.

num_snapshots : int = 30

If fibre_method="random", the number of snapshots to sample.

References

Strössner, Christoph, Bonan Sun, and Daniel Kressner. 2024. “Approximation in the Extended Functional Tensor Train Format.” Advances in Computational Mathematics 50 (3): 54. https://doi.org/10.1007/s10444-024-10140-9.