Merge branch 'remove_bijectors' of github.com:TuringLang/AdvancedVI.jl into remove_bijectors

Author: Red-Portal

docs/src/tutorials/basic.md

@@ -50,7 +50,7 @@ end
 nothing
 ```
 
-Notice that the support of `σ` is constrained to be positive. 
+Notice that the support of `σ` is constrained to be positive.
 Since most VI algorithms assume an unconstrained Euclidean support, we will apply a change-of-variable by a bijective map (a *bijector*) to the posterior to make it unconstrained.
 For this, we will use the [`Bijectors`](https://github.com/TuringLang/Bijectors.jl) package.
 
@@ -221,7 +221,6 @@ Therefore, we finally need to apply a change-of-variable to `q_out` to make it a
 q_trans = Bijectors.TransformedDistribution(q_out, binv)
 ```
 
-
 ## Custom Callback
 
 The ELBO estimates above however, use only a handful of Monte Carlo samples.

docs/src/tutorials/subsampling.md

@@ -88,9 +88,7 @@ using LogDensityProblemsAD
 
 prob = LogReg(X, y, size(X, 1))
 prob_ad = LogDensityProblemsAD.ADgradient(
-    ADTypes.AutoReverseDiff(), 
-    prob, 
-    x=randn(LogDensityProblems.dimension(prob))
+    ADTypes.AutoReverseDiff(), prob; x=randn(LogDensityProblems.dimension(prob))
 )
 nothing
 ```