pflm.utils.flatten_and_sort_data_matrices#

flatten_and_sort_data_matrices(y: list[~numpy.ndarray], t: list[~numpy.ndarray], input_dtype: str | ~numpy.dtype = <class 'numpy.float64'>, w: ~numpy.ndarray | None = None) → FlattenFunctionalData[source][source]#

Flatten per-sample 1D arrays into contiguous vectors and build indices.

This function concatenates lists of responses y and times t, expands per-sample weights w to observation level, drops NaNs, and constructs indexing helpers for the observation grid and subject ids.

Parameters:

ylist of np.ndarray: Each element is a 1D array of shape (nt_i,) with responses for sample i.
tlist of np.ndarray: Each element is a 1D array of shape (nt_i,) with time points for sample i.
input_dtypestr or np.dtype, default=np.float64: Target dtype for numeric arrays.
wnp.ndarray, optional: 1D array of length len(y) with per-sample weights. If None, uses ones.

Returns:

FlattenFunctionalData: Dataclass holding flattened arrays (y, t, w), grid/subject indices (tid, unique_tid, inverse_tid_idx, sid, unique_sid, sid_cnt).

Raises:

ValueError: If y/t are not lists of 1D arrays with matching lengths, if w is provided but invalid, or if all y values are NaN.