resample trajectory and hyp.tools.resample

[andrewheusser]

While animated line plots are resampled, static plot are not. We could add a resample flag to the plot function that allows a user to upsample (or downsample) their timeseries data. Then, we could expose the function as hyp.tools.resample.

questions:

• this only makes sense for timeseries data (I think), so do we allow the user to resample non-timeseries data?

• how should the user interact with the flag? i.e. does 10 mean upsample by a factor of 10? what about downsampling?

[andrewheusser]

one more thought (maybe worth a separate issue): it looks like the degree of sub-sampling (using pchip) changes with duration, but it shouldn't-- duration should change the rotation rate (i.e. how many degrees the camera moves with each frame), not the appearance of the data.

this will also fix the issue of static plots not being smoothed whereas animated plots are.

note: if a non-connected plot style is used (for animation or otherwise), no interpolation should be done. we'll have to parse out each line style to see if it's one of the continuous variants, which could get tricky (e.g. if there are some groups with non-connected styles and others with connected styles).

[jeremymanning]

This is now fully wired up in the HyperTools 1.0 update (PR #272, #272): hyp.plot() now accepts a resample= kwarg that runs the existing hyp.manip Resample (PCHIP) manipulator on each dataset right after format_data and before the normalize/reduce/align pipeline, so the resampled row count is what everything downstream sees.

Before: no way to resample a dataset's row count from hyp.plot() directly -- you had to call hyp.manip(..., model='Resample', ...) yourself and re-plot the result.

After:

import numpy as np, hypertools as hyp

rng = np.random.default_rng(0)
data = np.cumsum(rng.standard_normal((100, 3)), axis=0)

expected = np.asarray(hyp.manip(data, model='Resample', n_samples=500))

fig = hyp.plot([data], 'o', show=False, resample=500)
lines = fig.axes[0].get_lines()
assert lines[0].get_data_3d()[0].shape[0] == 500  # 500-vertex data, pre-interp

result = hyp.plot([data], 'o', show=False, return_model=True, resample=500,
                  normalize=None)
assert np.allclose(result['xform_data'][0], expected)  # matches hyp.manip exactly

resample= also composes cleanly with the (separately fixed, GH #141) line-smoothing interpolation: a line-style plot with resample=200 first resamples to 200 rows, then still gets further densified for smooth line rendering. Bad values (anything other than False/None or an integer >= 2) raise a clear ValueError.

Closing as addressed in the 1.0 update -- please reopen if you hit anything else.