Draft: Prevent Overflow in pAstro calculation

Motivation

During LLPIC, an event from the injection channel got uploaded with pAstro as NaNs - pAstro file. The logs were checked and they revealed that there was overflow in the pAstro calculation for this event.

2025-05-21 13:38:13,507 | ll_inspiral_pastro_uploader : DEBUG : Processing G2989487...
/usr/local/lib/python3.6/site-packages/pastro/pastro.py:355: RuntimeWarning: overflow encountered in exp
  weights[c]=np.exp(np.nan_to_num(PPoly(self.template_weight_coefficients[c][:,:,index],self.template_weight_SNR)(snr)))
/usr/local/lib/python3.6/site-packages/pastro/pastro.py:356: RuntimeWarning: invalid value encountered in double_scalars
  return {c:weights[c]/np.sum(list(weights.values())) for c in self.categories}
2025-05-21 13:38:14,223 | ll_inspiral_pastro_uploader : DEBUG : Successfully uploaded gstlal.p_astro.json to G2989487.

Fix

NaNs make no sense for pAstro values. This MR prevents NaNs from being uploaded. It changes the behaviour of the the function such that

• Clips SNR such that maximum value passed to PPoly is 80
• It fails out if PPoly returns +inf
• NaN is mapped to -inf
• The PPoly output is normalized using scipy.special.logsumexp

Tests

Same Event

For the same event, the output is now more reasonable

SNR: 167.3338762340001
[WARNING] SNR=167.3338762340001 exceeds 80. Clipping to 80.
Poly Vals: {'BBH': -9.901634801062755, 'BNS': -3204.3943249001168, 'NSBH': -1981.145652845502}
Weights: {'BBH': 1.0, 'BNS': 0.0, 'NSBH': 0.0}
Total: 1.0
P(astro) = {"BBH": 1.0, "BNS": 0.0, "NSBH": 0.0, "Terrestrial": 0.0}

poly_val = inf

Manually set output of PPoly to inf

AssertionError: poly_val for category 'BBH' is +inf for template_id=1809086.0, snr=80.0.

poly_val = NaN

Manually set output of PPoly to NaN for BBH

SNR: 167.3338762340001
[WARNING] SNR=167.3338762340001 exceeds 80. Clipping to 80.
Poly Vals: {'BBH': -inf, 'BNS': -3204.3943249001168, 'NSBH': -1981.145652845502}
Weights: {'BBH': 0.0, 'BNS': 0.0, 'NSBH': 1.0}
Total: 1.0
P(astro) = {"BBH": 0.0, "BNS": 0.0, "NSBH": 1.0, "Terrestrial": 0.0}

All -inf

Manually set output of PPoly to NaN for all

SNR: 167.3338762340001
[WARNING] SNR=167.3338762340001 exceeds 80. Clipping to 80.
Poly Vals: {'BBH': -inf, 'BNS': -inf, 'NSBH': -inf}
Traceback (most recent call last):
  File "test.py", line 46, in <module>
    p_astro_value = uploader.calculate_pastro(mock_event)
  File "/ligo/home/ligo.org/shomik.adhicary/Projects/2025_Pastro/gstlal_ll_inspiral_pastro_uploader.py", line 226, in calculate_pastro
    pa = self.model(data,inj=self.is_injection_job)
  File "/usr/local/lib/python3.6/site-packages/pastro/pastro.py", line 402, in __call__
    template_weights=self.calculate_template_weights(template_id,snr)
  File "/usr/local/lib/python3.6/site-packages/pastro/pastro.py", line 382, in calculate_template_weights
    f"logsumexp returned a non-finite value (lse = {lse}) for template_id={template_id}, snr={snr}"
AssertionError: logsumexp returned a non-finite value (lse = -inf) for template_id=1809086.0, snr=80.0

SNR below 80

SNR: 79.99999
Poly Vals: {'BBH': -9.901634652452396, 'BNS': -3204.3935259286513, 'NSBH': -1981.145539405018}
Weights: {'BBH': 1.0, 'BNS': 0.0, 'NSBH': 0.0}
Total: 1.0
P(astro) = {"BBH": 1.0, "BNS": 0.0, "NSBH": 0.0, "Terrestrial": 0.0}

Notes

Have noticed

/usr/lib64/python3.6/site-packages/scipy/integrate/quadpack.py:364: IntegrationWarning: The occurrence of roundoff error is detected, which prevents 
  the requested tolerance from being achieved.  The error may be 
  underestimated.
  warnings.warn(msg, IntegrationWarning)

might be an underlying issue causing overflow