pg.flatten() in star __init__ uses default filter width

[alexlyttle]

The default filter width of 0.01 used by pg.flatten() (added to star.__init__() following Issue #197) sometimes subtracts the signal, requiring a larger filter width (see below). When you pass a custom SNR to star, this resets and uses the default filter width. A quick work-around is to set star.pg = custom_snr following initialisation. However, I feel that either subtracting the background within PBjam or having kwargs to pass to flatten would be more intuitive.

# Given some predefined lightkurve.Periodogram object `pg`
flattened = []
filter_widths = [0.005, 0.01, 0.1, 0.2]

for fw in filter_widths:
    flattened.append(pg.flatten(filter_width=fw, return_trend=True))

ax = pg.plot(scale='log', alpha=0.3, label='No filter')

for (snr, bkd), fw in zip(flattened, filter_widths):
    bkd.plot(scale='log', ax=ax, label=f'FW = {fw}')

ax.axvline(numax[0], ymax=2e4, c='r', ls='--', alpha=0.5, label=r'$\nu_\max$')
ax.set_xlim(8, 300)
ax.set_ylim(10, 2e4)
ax.legend()

fig2, axs = plt.subplots(2, 1, figsize=(12, 9))
snr = [pg.flatten(), pg.flatten(filter_width=0.2)]

for ax, s in zip(axs, snr):
    ax.axvline(numax[0], ymax=100, c='r', ls='--', alpha=0.5, label=r'$\nu_\max$')
    s.plot(ax=ax);
    ax.set_xlim(0, 150)

plt.show()

[grd349]

Related but different to #225 - we should fix both at the same time.