Use cesium for light curve feature generation?

[bfhealy]

Along with/in place of migrating ztfperiodic to scope (#189), we should consider using cesium for ZTF light curve feature generation going forward. It will be important to ensure a reasonable level of consistency between new and old scope feature sets.

Below are my initial findings for ztfperiodic features that appear to be missing (or different) in cesium after searching through descriptions here. This list may be updated if suitably similar features are identified in cesium.

Missing features from cesium

• Anderson-Darling (ad)
• Shapiro-Wilk (sw)
• welch_i
• f1_bic
• f1_a
• f1_b
• f1_phi0
• f1_power
• von Neumann statistic (inv_vonneumannratio)
• norm_excess_var
• roms
• significance
• smallkurt
• wstd

Different in cesium

• chi2red: the only chi2 features in cesium are qso_log_chi2_qsonu and qso_log_chi2nuNULL_chi2nu
• norm_peak_to_peak_amp: has some similar features; percent_amplitude and percent_difference_flux_percentile
• slight differences in percentile flux ratios (e.g. i60r is closest to flux_percentile_ratio_mid65)