The default prior for σ is Exponential(residual), where residual = 1 / std(y). This follows the same approach as e.g. rstanarm. The Exponential in TuringGLM takes the scale as an argument, whereas stan takes the rate = 1/scale. The correct residual should be residual = std(y).
The default prior for
σ
isExponential(residual)
, whereresidual = 1 / std(y)
. This follows the same approach as e.g. rstanarm. TheExponential
in TuringGLM takes the scale as an argument, whereas stan takes therate = 1/scale
. The correct residual should beresidual = std(y)
.