Might be a way to parse out all the information and get the plotting mixins for free

Parameters: dist._param_info()
Continuous or discrete: from subclass

Not imported. Only used in the documentation

https://en.wikipedia.org/wiki/Normal-Wishart_distribution

Can anything be down for this type of model?

Includes:

• gamma with known rate
  inference on the shape is

$$\propto {\frac {a^{\alpha -1}\beta ^{\alpha c}}{\Gamma (\alpha )^{b}}}$$

• Gamma
  inference on the shape and inverse scale is

$$\propto {\frac {p^{\alpha -1}e^{-\beta q}}{\Gamma (\alpha )^{r}\beta ^{-\alpha s}}}$$

• Beta
  inference on alpha and beta is

$${\displaystyle \propto {\frac {\Gamma (\alpha +\beta )^{k},p^{\alpha },q^{\beta }}{\Gamma (\alpha )^{k},\Gamma (\beta )^{k}}}}$$

helper to plot this distribution

potential use the contour or contourf from matplotlib

same as normal
paper here

https://en.wikipedia.org/wiki/Normal-inverse-Wishart_distribution

inverse gamma prior is already supported

Is this possible given the pyproject toml? This would restrict functionality

mean and sigma exchangeability model already supported

https://en.wikipedia.org/wiki/Wishart_distribution

Consider https://pypi.org/project/pytest-mpl/

These aren't vectorized in scipy

https://en.wikipedia.org/wiki/Inverse-Wishart_distribution

Example using pypika to implement a model with a prior in just SQL

• Linear regression from pandas-bootstrap

Resources:

• https://gregorygundersen.com/blog/2020/02/04/bayesian-linear-regression/

x_total = sum(x)
x2_total = sum(x**2)
n = len(x)
etc

Shouldn't this just check for numeric types? How does it check against pytensor? Is it supposed to with with pytensor?

I.e. linear regression with multivariate student T posterior predictive at a given point is also student T. Does it make sense to index in and get this distribution out

• known covariance matrix
• known precision matrix
• known mean vector

The sampling doesn't seem right here based on the generation defined here:
https://en.wikipedia.org/wiki/Normal-inverse-gamma_distribution

Might be in scipy soon: scipy/scipy#17330

Can help with the negative binomial posterior predictive

Use the dirichlet_multinomial distribution from scipy