pos must be nonnegative and less than window_length

[retifrav]

This might be a duplicate of #76.

There is a problem in detrend_savgol() function on the line wl = max(wl, 5). Or at least I think it is a problem, because in my script it leads to the following error:

ValueError: pos must be nonnegative and less than window_length

Full error output, just in case

Detrending fake LC:

/usr/local/lib/python3.11/site-packages/numpy/lib/nanfunctions.py:1215: RuntimeWarning: Mean of empty slice
  return np.nanmean(a, axis, out=out, keepdims=keepdims)
Traceback (most recent call last):
  File "/path/to/some.py", line 23, in <module>
    flcd, fakeflc = flcd.sample_flare_recovery(
                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/path/to/AltaiPony/altaipony/flarelc.py", line 519, in sample_flare_recovery
    fake_lc = fake_lc.detrend(mode, func=func, **detrend_kwargs)
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/path/to/AltaiPony/altaipony/flarelc.py", line 342, in detrend
    new_lc =  detrend_savgol(new_lc, **kwargs)
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/path/to/AltaiPony/altaipony/altai.py", line 282, in detrend_savgol
    flux_model_i = savgol_filter(flux, wl, 3, mode='nearest')
                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.11/site-packages/scipy/signal/_savitzky_golay.py", line 341, in savgol_filter
    coeffs = savgol_coeffs(window_length, polyorder, deriv=deriv, delta=delta)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.11/site-packages/scipy/signal/_savitzky_golay.py", line 112, in savgol_coeffs
    raise ValueError("pos must be nonnegative and less than "
ValueError: pos must be nonnegative and less than window_length.

So wl can be NaN - value that is bigger than 5 - which is apparently what is causing the problem.

A script to reproduce the problem

from altaipony.lcio import from_mast
import pandas

flc = from_mast(
    "TRAPPIST-1",
    mode="LC",
    cadence="short",
    mission="K2"
)

for j in range(len(flc)):
    print(f"\n--- {j} ---\n")
    # detrend curve
    flcd = flc[j].detrend("savgol")
    # find flares
    flcd = flcd.find_flares(N1=3, N2=1, N3=3, minsep=3)
    flcdpanda = flcd.flares
    # print(flcdpanda)
    if not flcdpanda.empty:
        # here it fails with the error "pos must be nonnegative"
        flcd, fakeflc = flcd.sample_flare_recovery(
            inject_before_detrending=True,
            mode="savgol",
            iterations=20,
            fakefreq=2,
            ampl=[1e-4, 0.5],
            dur=[.001/6., 0.1/6.]
        )
    else:
        print("DataFrame is empty, no flares")

To fix this I applied the following patch:

diff --git a/altaipony/altai.py b/altaipony/altai.py
index 1d73d47..78ebb12 100755
--- a/altaipony/altai.py
+++ b/altaipony/altai.py
@@ -274,6 +274,12 @@ def detrend_savgol(lc, window_length=None, pad=3, printwl=False, **kwargs):
             wl = np.floor(.1 / dt)
             if wl % 2 == 0:
                 wl = wl + 1
+
+        # don't know what is a proper fallback in this situation,
+        # but since later it takes maximum of this value and 5,
+        # then 0 seems to be okay
+        if np.isnan(wl):
+            wl = 0
         
         # args are flux, window_length, polyorder, mode is 
         wl = max(wl, 5) #wl must be larger than polyorder

But I am not sure whether this is a correct was to fix it or not.

Tested with aa6ba8d revision.