nigsp.operations.timeseries.graph_filter¶

nigsp.operations.timeseries.graph_filter(timeseries, eigenvec, freq_idx, keys=['low', 'high'])[source]¶

Filter a graph decomposition into two parts based on freq_idx.

Return the two eigenvector lists (high freq and low freq) that are equal to the original eigenvector list, but “low” is zero-ed for all frequencies >= of the given index, and “high” is zero-ed for all frequencies < to the given index. Also return their projection onto a timeseries.

Parameters:

timeseriesnumpy.ndarray: The input timeseries. It is assumed that the second dimension is time.
eigenvecnumpy.ndarray: The eigenvector resulting from the Laplacian decomposition.
freq_idxint or list: The index of the frequency that splits the spectral power into two (more or less) equal parts - i.e. the index of the first frequency in the “high” component.
keyslist, optional: The keys to call the split parts with.

Returns:

dict of numpy.ndarray: Return first the split eigenvectors.
dict of numpy.ndarray: Return second the projected split eigenvectors onto the timeseries.

Raises:

IndexError: If the given index is 0 (all “high”), the last possible index (all “low”), or higher than the last possible index (not applicable).