If a signal is sparse in the Fourier domain - for example, it is the sum of a small number of coherent sinusoids - you shouldn't be using the Fourier transform or Lomb-Scargle periodogram to analyse it! This is especially important for sparse, irregularly-sampled time series, as often occur in RV surveys, ground-based photometric surveys, or photometry from missions such as Hipparcos or Gaia. In these cases it would not ordinarily be possible to recover a good power spectrum of the astrophysical variability - but we know that for typical pulsating stars, or planets, there are only a few sinusoidal components to the signal.
Following the theory of compressed sensing, the L1 metric (sum of absolute values) turns out to be far better than the L2 metric (Euclidean distance) for inferring a sparse signal given noisy data. L1 is sometimes called the 'taxicab' or 'Manhattan' metric because in the discrete case, you can only move on a grid of streets rather than taking a diagonal - and because this code was written with the Empire State Building in view we'll take that as the repo name.
manhattan follows Chen & Donoho, "Application of Basis Pursuit in Spectrum Estimation", who built a compressed sensing periodogram with an astronomical context in mind. They considered a sum of sine, cosine and Dirac terms to fit astronomical radial velocity planet signals in collaboration with Scargle; we add some polynomial terms to help take care of long term trends, with the goal of also including cotrending basis vectors to make this more applicable to Kepler/K2 data. manhattan has a similar approach to Hara et al., 'Radial velocity data analysis with compressed sensing techniques', who did not publish their code.
This repo is a thin wrapper for the SPGL1 library, and we use basis pursuit to find our sparse periodogram. This is a work in progress and doesn't do everything I want it to do yet, or really work in every case!
To the best of my knowledge no such code is available open-source. I hope you find it useful!
Just type
python setup.py install
and give it a go.
Buy me a beer? And cite Chen & Donoho, and Hara et al.