Binomial uncertainty

[javierggt]

Description

Add binomial uncertainty interval to star_probs.py

Interface impacts

Testing

Unit tests

• No unit tests
• Mac
• Linux
• Windows

Independent check of unit tests by [REVIEWER NAME]

• [PLATFORM]:

Functional tests

This notebook: coverage.ipynb

[taldcroft]

This test case should work:

In [7]: star_probs.binomial_confidence_interval(1, [1, 2])
---------------------------------------------------------------------------
IndexError                                Traceback (most recent call last)
Cell In[7], line 1
----> 1 star_probs.binomial_confidence_interval(1, [1, 2])

File ~/git/chandra_aca/chandra_aca/star_probs.py:1161, in binomial_confidence_interval(n_true, n_trials, coverage)
   1159 n_trials = np.ma.MaskedArray(n_trials, mask=(n_trials == 0))
   1160 ratio = np.ones_like(n_true) * np.nan
-> 1161 ratio[~n_trials.mask] = n_true[~n_trials.mask] / n_trials[~n_trials.mask]
   1163 # try:
   1164 #     float(coverage)
   1165 # except ValueError:
   1166 #     raise Exception("The coverage must be a float!") from None
   1168 alpha = (1 - coverage) / 2

IndexError: boolean index did not match indexed array along dimension 0; dimension is 1 but corresponding boolean dimension is 2

In [8]: np.broadcast_arrays(1, [1, 2])
Out[8]: [array([1, 1]), array([1, 2])]

In [9]: np.broadcast_arrays(1, 1)
Out[9]: [array(1), array(1)]

[javierggt]

@taldcroft I think I addressed all your comments.

[taldcroft]

@javierggt - thanks, looking much better. Just one more nitpick, returning a scalar array is technically OK but just not nice.

In [2]: star_probs.binomial_confidence_interval(0, 1, 0.6)
Out[2]: (array(0.), array(0.), array(0.47203609))

If you look at acq_success_prob there is a decent idiom for dealing with this. The chandra_aca.transform.broadcast_arrays() function does the atleast_1d for the inputs and returns an is_scalar bool. Then at the end if is_scalar=True then return value[0].

[taldcroft]

Suggested change

	"""Binomial error calculation using the Jeffreys prior.
	"""Binomial confidence interval calculation using the Jeffreys prior.

[taldcroft]

The first two can both be int.

[taldcroft]

Looks good thanks!