Port #539 to master (#553)

* Uncomment asserts

* Bump Quarto version in CI

* Mention Quarto min version in README

Author: penelopeysm

README.md

@@ -8,8 +8,10 @@ This repository is part of [Turing.jl's](https://turinglang.org/) website (i.e.
 
 To get started with the docs website locally, you'll need to have [Quarto](https://quarto.org/docs/download/) installed.
 Make sure you have at least version 1.5 of Quarto installed, as this is required to correctly run [the native Julia engine](https://quarto.org/docs/computations/julia.html#using-the-julia-engine).
+Ideally, you should use Quarto 1.6.31 or later as this version fixes [a bug which causes random number generation between different cells to not be deterministic](https://github.com/TuringLang/docs/issues/533).
+Note that as of October 2024, Quarto 1.6 is a pre-release version, so you may need to install it from source rather than via a package manager like Homebrew.
 
-Once you have the prerequisite installed, you can follow these steps:
+Once you have Quarto installed, you can follow these steps:
 
 1. Clone this repository:
 

tutorials/01-gaussian-mixture-model/index.qmd

@@ -142,8 +142,7 @@ let
         # μ[1] and μ[2] can switch places, so we sort the values first.
         chain = Array(chains[:, ["μ[1]", "μ[2]"], i])
         μ_mean = vec(mean(chain; dims=1))
-        # TODO: https://github.com/TuringLang/docs/issues/533
-        # @assert isapprox(sort(μ_mean), μ; rtol=0.1) "Difference between estimated mean of μ ($(sort(μ_mean))) and data-generating μ ($μ) unexpectedly large!"
+        @assert isapprox(sort(μ_mean), μ; rtol=0.1) "Difference between estimated mean of μ ($(sort(μ_mean))) and data-generating μ ($μ) unexpectedly large!"
     end
 end
 ```
@@ -208,8 +207,7 @@ let
         # μ[1] and μ[2] can no longer switch places. Check that they've found the mean
         chain = Array(chains[:, ["μ[1]", "μ[2]"], i])
         μ_mean = vec(mean(chain; dims=1))
-        # TODO: https://github.com/TuringLang/docs/issues/533
-        # @assert isapprox(sort(μ_mean), μ; rtol=0.4) "Difference between estimated mean of μ ($(sort(μ_mean))) and data-generating μ ($μ) unexpectedly large!"
+        @assert isapprox(sort(μ_mean), μ; rtol=0.4) "Difference between estimated mean of μ ($(sort(μ_mean))) and data-generating μ ($μ) unexpectedly large!"
     end
 end
 ```
@@ -349,8 +347,7 @@ let
         # μ[1] and μ[2] can no longer switch places. Check that they've found the mean
         chain = Array(chains[:, ["μ[1]", "μ[2]"], i])
         μ_mean = vec(mean(chain; dims=1))
-        # TODO: https://github.com/TuringLang/docs/issues/533
-        # @assert isapprox(sort(μ_mean), μ; rtol=0.4) "Difference between estimated mean of μ ($(sort(μ_mean))) and data-generating μ ($μ) unexpectedly large!"
+        @assert isapprox(sort(μ_mean), μ; rtol=0.4) "Difference between estimated mean of μ ($(sort(μ_mean))) and data-generating μ ($μ) unexpectedly large!"
     end
 end
 ```

tutorials/09-variational-inference/index.qmd

@@ -155,9 +155,8 @@ var(x), mean(x)
 #| echo: false
 let
     v, m = (mean(rand(q, 2000); dims=2)...,)
-    # TODO: Fix these as they randomly fail https://github.com/TuringLang/docs/issues/533
-    # @assert isapprox(v, 1.022; atol=0.1) "Mean of s (VI posterior, 1000 samples): $v"
-    # @assert isapprox(m, -0.027; atol=0.03) "Mean of m (VI posterior, 1000 samples): $m"
+    @assert isapprox(v, 1.022; atol=0.1) "Mean of s (VI posterior, 1000 samples): $v"
+    @assert isapprox(m, -0.027; atol=0.03) "Mean of m (VI posterior, 1000 samples): $m"
 end
 ```
 

tutorials/11-probabilistic-pca/index.qmd

@@ -246,13 +246,10 @@ heatmap(
 We can quantitatively check the absolute magnitudes of the column average of the gap between `mat_exp` and `mat_rec`:
 
 ```{julia}
-#| echo: false
-# let
-#     diff_matrix = mat_exp .- mat_rec
-#     @assert abs(mean(diff_matrix[:, 4])) <= 0.5 #0.327
-#     @assert abs(mean(diff_matrix[:, 5])) <= 0.5 #0.390
-#     @assert abs(mean(diff_matrix[:, 6])) <= 0.5 #0.326
-# end
+diff_matrix = mat_exp .- mat_rec
+for col in 4:6
+    @assert abs(mean(diff_matrix[:, col])) <= 0.5
+end
 ```
 
 We observe that, using posterior mean, the recovered data matrix `mat_rec` has values align with the original data matrix - particularly the same pattern in the first and last 3 gene features are captured, which implies the inference and p-PCA decomposition are successful.
@@ -281,12 +278,12 @@ Another way to put it: 2 dimensions is enough to capture the main structure of t
 A direct question arises from above practice is: how many principal components do we want to keep, in order to sufficiently represent the latent structure in the data?
 This is a very central question for all latent factor models, i.e. how many dimensions are needed to represent that data in the latent space.
 In the case of PCA, there exist a lot of heuristics to make that choice.
-For example, We can tune the number of principal components using empirical methods such as cross-validation based some criteria such as MSE between the posterior predicted (e.g. mean predictions) data matrix and the original data matrix or the percentage of variation explained [3].
+For example, We can tune the number of principal components using empirical methods such as cross-validation based some criteria such as MSE between the posterior predicted (e.g. mean predictions) data matrix and the original data matrix or the percentage of variation explained [^3].
 
 For p-PCA, this can be done in an elegant and principled way, using a technique called *Automatic Relevance Determination* (ARD).
-ARD can help pick the correct number of principal directions by regularizing the solution space using a parameterized, data-dependent prior distribution that effectively prunes away redundant or superfluous features [4].
+ARD can help pick the correct number of principal directions by regularizing the solution space using a parameterized, data-dependent prior distribution that effectively prunes away redundant or superfluous features [^4].
 Essentially, we are using a specific prior over the factor loadings $\mathbf{W}$ that allows us to prune away dimensions in the latent space. The prior is determined by a precision hyperparameter $\alpha$. Here, smaller values of $\alpha$ correspond to more important components.
-You can find more details about this in e.g. [5].
+You can find more details about this in, for example, Bishop (2006) [^5].
 
 ```{julia}
 @model function pPCA_ARD(X)
@@ -383,4 +380,4 @@ It can also thought as a matrix factorisation method, in which $\mathbf{X}=(\mat
 [^2]: Probabilistic PCA by TensorFlow, "https://www.tensorflow.org/probability/examples/Probabilistic_PCA".
 [^3]: Gareth M. James, Daniela Witten, Trevor Hastie, Robert Tibshirani, *An Introduction to Statistical Learning*, Springer, 2013.
 [^4]: David Wipf, Srikantan Nagarajan, *A New View of Automatic Relevance Determination*, NIPS 2007.
-[^5]: Christopher Bishop, *Pattern Recognition and Machine Learning*, Springer, 2006.
+[^5]: Christopher Bishop, *Pattern Recognition and Machine Learning*, Springer, 2006.