Questions and limits¶

Which function do I need?¶

Task	Call
Plot a sequence	`hp.plot(values, curve=12)`
Draw the path	`hp.plot_curve(0, order=3)`
Draw a non-power-of-two path	`hp.plot_curve(0, shape=(15, 16))`
Make a data grid	`hp.data.to_grid(values, curve=12)`
Read an image	`hp.data.from_grid(image, curve=12)`
Reconstruct it	`hp.data.to_grid(sequence, curve=12, shape=image.shape)`
Transform without drawing	`hp.data.fourier(values)`
Get curve metadata	`hp.curves.info(12)`
Locate one index	`hp.curves.index_to_xy(index, order, curve=12)`

Why is nothing showing in my script?¶

The plotting functions return figures without opening windows. Call matplotlib.pyplot.show() or save with fig.savefig(...). For a machine without a display, set MPLBACKEND=Agg. Notebook display behavior depends on the selected Matplotlib backend.

Does square fitting discard values?¶

No. fit="square" uses side ceil(sqrt(N)), with padding if needed. It requires one of the eight generalizing curves. fit="pad" selects an enclosing power-of-two square. fit="truncate" explicitly allows dropping the tail. Examples

Can I use a general rectangle?¶

This version supports squares and quasisquares, whose side lengths differ by one. General rectangles raise an error. Arbitrary squares/quasisquares require a supporting curve; all forty curves support power-of-two squares.

Why did integer padding produce a masked array?¶

NaN cannot be represented as an integer. Masked padding preserves exact integer values, including uint64 values above the exact range of float64. Use a representable explicit fill=0 if an ordinary integer array is preferable. Keep track of the original length.

Does from_grid remove NaNs or masks?¶

No: those may be legitimate input values. Pass length=original_length to keep an explicit number of flattened values. For color images, that number includes channels. Do not trim padding before inverting a permutation created by using different curves.

Is the Fourier map the signal’s spectrum?¶

It is the mapped image’s 2-D FFT magnitude. It is neither the original sequence’s 1-D spectrum nor squared power. Curve choice, fitting, missing-value fill, mean subtraction and windowing affect it. Windowing changes amplitudes and has no gain correction; a two-cell Hann axis is all zeros. Exact steps

Why does the difference map ignore my values?¶

It measures spatial neighbours’ separation in visit index. It is a geometric property of the selected curve and grid, so it remains identical for different measurements on that grid. Granularity changes measurements, and therefore does not change the difference map.

Why do some curves look identical at low orders?¶

Distinct recursive constructions can share a low-order path. Their catalog classification refers to the construction across orders, not just one tiny picture.

Are the curves continuous mathematical objects?¶

The package materializes discrete cell paths at a chosen finite resolution. The catalog is based on forty Hilbert-type constructions from the cited paper. It does not implement every possible space-filling curve family.

Can I use complex values?¶

Raw array conversion preserves complex dtypes. Plotting and the public Fourier helper require real scalar sequences. Explicitly select .real, .imag, np.abs(...) or np.angle(...) when that is the quantity you want to display.

Is there a DOI?¶

Yes. 10.5281/zenodo.23858267 represents all versions; 1.0.1 has its own archived record. New GitHub releases are archived automatically. See CITATION.cff in the repository.