|View source on GitHub|
Evaluates a lattice using hypercube interpolation.
tfl.lattice_lib.evaluate_with_hypercube_interpolation( inputs, kernel, units, lattice_sizes, clip_inputs )
Lattice function is multi-linearly interpolated between the 2^d vertices of a hypercube. This interpolation method is typically slower than simplex interpolation, since each value is interpolated from 2^d hypercube corners, rather than d+1 simplex corners. For details, see e.g. "Dissection of the hypercube into simplices", D.G. Mead, Proceedings of the AMS, 76:2, Sep. 1979.
Tensor of shape:
||Lattice kernel of shape (num_params_per_lattice, units).|
||Output dimension of the lattice.|
||List or tuple of integers which represents lattice sizes of layer for which interpolation is being computed.|
||Whether inputs should be clipped to the input range of the lattice.|