That's the angular frequencies at which you want to evaluate the periodogram at. In your case (otherwise-regular time series but with missing values), I would recommend using the angular frequencies that you would have had if you had computed a normal periodogram using the FFT on the whole time series, e.g. `np.linspace(0, np.pi/300.0, 8928//2)` (assuming your `x` is in seconds). The running time is O(len(x)*len(freqs)), though, so that may take a long time. You may want to reduce the number of points you sample at first for visualization, then you can zoom in at the full frequency resolution to an area of interest if there is lots of dead space.