next manual (2.6.3++)

[bob-carpenter]

This is the issue for the next manual release. Please submit all bugs and feature requests for the manual here.

[randommm]

Notice: I just created the branch for this issue with some typos fixed.

[syclik]

From Andrew:

• We should always lead with the fast code and then explain how it’s equivalent to the non-vectorized version. For example on pp. 61-62 of manual we have multivariate regression with hier model and we start with non-vectorized. This is confusing because if I want to imitate this model myself I have to grab different code snippets.

[Ed. I only created the first optimization. I left the Cholesky one as an add-on.]

[bob-carpenter]

Sunil Nandihalli reported on stan-users:

on page 50 of stan-reference-2.6.2.pdf in section 5.7 it is said that C belongs R^(D-1)
should be read as C belongs to R^(K-1)

• fix it
• thank Sunil

[bob-carpenter]

Andy Choi brought this up on stan-users:

Under the Gaussian Process chapter of the STAN reference manual (Ch 14, version: 2.6.2, pg. 140, 142, 143), the footnote mentioned/suggested the code referenced in the text can be found at: http://mc-stan.org/examples.html

• point users to https://github.com/stan-dev/example-models/tree/master/misc/gaussian-process
• thank Andy

[bob-carpenter]

• add Alp Kucukelbir as a core developer (variational inference!)

[bob-carpenter]

From Sebastian Weber:

• add a reference to

https://software.intel.com/en-us/articles/consistency-of-floating-point-results-using-the-intel-compiler

in your manual concerning the Intel compiler. Most users won’t bother, but I certainly do.

[bob-carpenter]

From Heraa Hu in issue #1450

In the manual p. 273 it says: "Stan allows function overloading, where two functions have the same name. For example, an overloaded unit-normal log density function might be declared with an overload of the function unit_normal_log, giving it three signatures with one argument each."

• remove this and explicitly mention that overloading doesn't work

[jgabry]

On p. 59 (second paragraph of section 5.12) it says

"... and the outcome income, with predictors such as income and age...".

Income is listed both as the outcome and a predictor here. Maybe

• use education level (or something else) as an example predictor instead of income?

[bob-carpenter]

• add a disclaimer about local optima in mixture models (degenerate solutions where one component collapses into another)

[syclik]

• thank @dvukcevic for fixing a typo and creating a pull request: stan-reference: fix typo in HMC (56.1) #1491.
• thank @mbjoseph for fixing typos and creating pull request: fixed a few typos #1496.