nemos.convolve.tensor_convolve

nemos.convolve.tensor_convolve(array, eval_basis)

Apply a convolution on the given array with the evaluation basis and reshapes the result.

This function first flattens the input array across dimensions other than the first one, then performs a vectorized convolution with the evaluation basis. The result is reshaped back to the original array’s shape (except for the first dimension, which is adjusted based on the convolution).

Parameters:

• array (NDArray) – The input array to convolve. It is expected to be at least 1D. The first axis is expeted to be the sample axis, i.e. the shape of array is (num_samples, ...).

• eval_basis (NDArray) – The evaluation basis array for convolution. It should be 2D, where the first dimension represents the window size for convolution. Shape (window_size, n_basis_funcs).

Returns:

The convolved array, reshaped to maintain the original dimensions except for the first one, which is adjusted based on the window size of eval_basis.

Notes

The convolution implemented here is in mode "valid". This implies that the time axis shrinks num_samples - window_size + 1, where num_samples is the first size of the first axis of array and window_size is the size of the first axis in eval_basis.