bayesplain.mean¶
-
bayesplain.mean(x, prior=
'conventional', reference=None, label='', unit='', log=False, n_draws=None, seed='unset', data=None)[source]¶ Estimate one unknown average from a sample.
The posterior is exact: with the reference prior, the mean of a normal sample has a scaled Student-t posterior with
n - 1degrees of freedom. No sampling, nothing to converge.- Parameters:¶
- x : array_like or str¶
Observations, or the name of a column in
data. Non-finite values are dropped, and the summary says how many.- prior : str, float, or EffectSizePrior, default 'conventional'¶
Cauchy prior on standardised effect size. Affects the Bayes factor only; the posterior and credible interval do not depend on it.
- reference : float, optional¶
The value to compare against: the null for the t-test, the threshold for the reported probability, and the point null for the Bayes factor. Defaults to 0. With
log=Trueit is required, on the original scale, since zero has no logarithm.- label : str, optional¶
What the quantity is, for output, e.g.
"commute time".- unit : str, optional¶
Display unit, e.g.
"minutes".- log : bool, default False¶
Work on the log scale and report the typical (geometric-mean) value on the original scale. For skewed, strictly positive data – prices, incomes – where the arithmetic mean is dragged around by a few large values.
referencestays on the original scale.- n_draws : int, optional¶
Number of draws for derived quantities. Defaults to the package setting.
- seed : int or None, optional¶
Seed for those draws.
- data : data frame, optional¶
Where to look
xup when it is a column name. The column name becomes the label unlesslabelis given.
- Returns:¶
Call
.summary()for the full report, and.predict()for a forecast of the next observation.- Return type:¶
Examples
>>> import numpy as np, bayesplain as bp >>> commutes = np.array([28, 35, 42, 31, 25, 38, 45, 33, 29, 40.0]) >>> res = bp.mean(commutes, reference=30, label="commute time", unit="minutes") >>> round(res.point("mean"), 2) 34.6 >>> res.exact True