Turing v0.36 and removal of Pathfinder code (#579)

* Bump Turing.jl to 0.36 and update all mentions of Gibbs to match * Replace Pathfinder example code with a link to Pathfinder docs * Update Manifest.toml * Bump Turing compat to 0.36.2 * Regenerate Manifest.toml with TuringBenchmarking.jl v0.5.8 * Update dependencies * Host golf.dat file ourselves --------- Co-authored-by: Penelope Yong <penelopeysm@gmail.com>
TuringLang · Jan 28, 2025 · 523428a · 523428a
1 parent c939949
commit 523428a
Show file tree

Hide file tree

Showing 12 changed files with 273 additions and 271 deletions.
diff --git a/Manifest.toml b/Manifest.toml
diff --git a/Project.toml b/Project.toml
@@ -7,7 +7,6 @@ AdvancedMH = "5b7e9947-ddc0-4b3f-9b55-0d8042f74170"
 Bijectors = "76274a88-744f-5084-9051-94815aaf08c4"
 CSV = "336ed68f-0bac-5ca0-87d4-7b16caf5d00b"
 ComponentArrays = "b0b7db55-cfe3-40fc-9ded-d10e2dbeff66"
-DataDeps = "124859b0-ceae-595e-8997-d05f6a7a8dfe"
 DataFrames = "a93c6f00-e57d-5684-b7b6-d8193f3e46c0"
 DataStructures = "864edb3b-99cc-5e75-8d2d-829cb0a9cfe8"
 DifferentialEquations = "0c46a032-eb83-5123-abaf-570d42b7fbaa"
@@ -39,7 +38,6 @@ Optimization = "7f7a1694-90dd-40f0-9382-eb1efda571ba"
 OptimizationNLopt = "4e6fcdb7-1186-4e1f-a706-475e75c168bb"
 OptimizationOptimJL = "36348300-93cb-4f02-beb5-3c3902f8871e"
 PDMats = "90014a1f-27ba-587c-ab20-58faa44d9150"
-Pathfinder = "b1d3bc72-d0e7-4279-b92f-7fa5d6d2d454"
 Plots = "91a5bcdd-55d7-5caf-9e0b-520d859cae80"
 RDatasets = "ce6b1742-4840-55fa-b093-852dadbb1d8b"
 Random = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
@@ -55,4 +53,4 @@ UnPack = "3a884ed6-31ef-47d7-9d2a-63182c4928ed"
 Zygote = "e88e6eb3-aa80-5325-afca-941959d7151f"
 
 [compat]
-Turing = "0.35"
+Turing = "0.36.2"
diff --git a/_quarto.yml b/_quarto.yml
@@ -32,7 +32,7 @@ website:
         text: Team
     right:
       # Current version
-      - text: "v0.35"
+      - text: "v0.36"
         menu:
           - text: Changelog
             href: https://turinglang.org/docs/changelog.html

diff --git a/core-functionality/index.qmd b/core-functionality/index.qmd
@@ -63,7 +63,7 @@ We can perform inference by using the `sample` function, the first argument of w
 c1 = sample(gdemo(1.5, 2), SMC(), 1000)
 c2 = sample(gdemo(1.5, 2), PG(10), 1000)
 c3 = sample(gdemo(1.5, 2), HMC(0.1, 5), 1000)
-c4 = sample(gdemo(1.5, 2), Gibbs(PG(10, :m), HMC(0.1, 5, :s²)), 1000)
+c4 = sample(gdemo(1.5, 2), Gibbs(:m => PG(10), :s² => HMC(0.1, 5)), 1000)
 c5 = sample(gdemo(1.5, 2), HMCDA(0.15, 0.65), 1000)
 c6 = sample(gdemo(1.5, 2), NUTS(0.65), 1000)
 ```
@@ -452,7 +452,7 @@ end
 
 simple_choice_f = simple_choice([1.5, 2.0, 0.3])
 
-chn = sample(simple_choice_f, Gibbs(HMC(0.2, 3, :p), PG(20, :z)), 1000)
+chn = sample(simple_choice_f, Gibbs(:p => HMC(0.2, 3), :z => PG(20)), 1000)
 ```
 
 The `Gibbs` sampler can be used to specify unique automatic differentiation backends for different variable spaces. Please see the [Automatic Differentiation]({{<meta using-turing-autodiff>}}) article for more.

diff --git a/developers/compiler/design-overview/index.qmd b/developers/compiler/design-overview/index.qmd
@@ -291,8 +291,7 @@ not. Let `md` be an instance of `Metadata`:
     `md.vns`, `md.ranges`, `md.dists`, `md.orders` and `md.flags`.
   - `md.vns[md.idcs[vn]] == vn`.
   - `md.dists[md.idcs[vn]]` is the distribution of `vn`.
-  - `md.gids[md.idcs[vn]]` is the set of algorithms used to sample `vn`. This is used in
-    the Gibbs sampling process.
+  - `md.gids[md.idcs[vn]]` is the set of algorithms used to sample `vn`. This was used by the Gibbs sampler. Since Turing v0.36 it is unused and will eventually be deleted.
   - `md.orders[md.idcs[vn]]` is the number of `observe` statements before `vn` is sampled.
   - `md.ranges[md.idcs[vn]]` is the index range of `vn` in `md.vals`.
   - `md.vals[md.ranges[md.idcs[vn]]]` is the linearized vector of values of corresponding to `vn`.

diff --git a/tutorials/gaussian-mixture-models/index.qmd b/tutorials/gaussian-mixture-models/index.qmd
@@ -112,7 +112,7 @@ model = gaussian_mixture_model(x);
 ```
 
 We run a MCMC simulation to obtain an approximation of the posterior distribution of the parameters $\mu$ and $w$ and assignments $k$.
-We use a `Gibbs` sampler that combines a [particle Gibbs](https://www.stats.ox.ac.uk/%7Edoucet/andrieu_doucet_holenstein_PMCMC.pdf) sampler for the discrete parameters (assignments $k$) and a Hamiltonion Monte Carlo sampler for the continuous parameters ($\mu$ and $w$).
+We use a `Gibbs` sampler that combines a [particle Gibbs](https://www.stats.ox.ac.uk/%7Edoucet/andrieu_doucet_holenstein_PMCMC.pdf) sampler for the discrete parameters (assignments $k$) and a Hamiltonian Monte Carlo sampler for the continuous parameters ($\mu$ and $w$).
 We generate multiple chains in parallel using multi-threading.
 
 ```{julia}
@@ -123,7 +123,7 @@ setprogress!(false)
 
 ```{julia}
 #| output: false
-sampler = Gibbs(PG(100, :k), HMC(0.05, 10, :μ, :w))
+sampler = Gibbs(:k => PG(100), (:μ, :w) => HMC(0.05, 10))
 nsamples = 150
 nchains = 4
 burn = 10

diff --git a/tutorials/gaussian-processes-introduction/golf.dat b/tutorials/gaussian-processes-introduction/golf.dat
@@ -0,0 +1,20 @@
+distance n y
+2 1443 1346 
+3 694 577 
+4 455 337 
+5 353 208 
+6 272 149 
+7 256 136 
+8 240 111 
+9 217 69 
+10 200 67 
+11 237 75 
+12 202 52 
+13 192 46 
+14 174 54 
+15 167 28 
+16 201 27 
+17 195 31 
+18 191 33 
+19 147 20 
+20 152 24
diff --git a/tutorials/gaussian-processes-introduction/index.qmd b/tutorials/gaussian-processes-introduction/index.qmd
@@ -30,20 +30,9 @@ given distance.
 ### Let's download the data and take a look at it:
 
 ```{julia}
-using CSV, DataDeps, DataFrames
-
-ENV["DATADEPS_ALWAYS_ACCEPT"] = true
-register(
-    DataDep(
-        "putting",
-        "Putting data from BDA",
-        "http://www.stat.columbia.edu/~gelman/book/data/golf.dat",
-        "fc28d83896af7094d765789714524d5a389532279b64902866574079c1a977cc",
-    ),
-)
-
-fname = joinpath(datadep"putting", "golf.dat")
-df = CSV.read(fname, DataFrame; delim=' ', ignorerepeated=true)
+using CSV, DataFrames
+
+df = CSV.read("golf.dat", DataFrame; delim=' ', ignorerepeated=true)
 df[1:5, :]
 ```
 

diff --git a/tutorials/hidden-markov-models/index.qmd b/tutorials/hidden-markov-models/index.qmd
@@ -137,7 +137,7 @@ setprogress!(false)
 ```
 
 ```{julia}
-g = Gibbs(HMC(0.01, 50, :m, :T), PG(120, :s))
+g = Gibbs((:m, :T) => HMC(0.01, 50), :s => PG(120))
 chn = sample(BayesHmm(y, 3), g, 1000);
 ```
 

diff --git a/usage/automatic-differentiation/index.qmd b/usage/automatic-differentiation/index.qmd
@@ -57,8 +57,8 @@ end
 c = sample(
     gdemo(1.5, 2),
     Gibbs(
-        HMC(0.1, 5, :m; adtype=AutoForwardDiff(; chunksize=0)),
-        HMC(0.1, 5, :s²; adtype=AutoReverseDiff(false)),
+        :m => HMC(0.1, 5; adtype=AutoForwardDiff(; chunksize=0)),
+        :s² => HMC(0.1, 5; adtype=AutoReverseDiff(false)),
     ),
     1000,
     progress=false,

diff --git a/usage/external-samplers/index.qmd b/usage/external-samplers/index.qmd
@@ -75,30 +75,10 @@ As previously mentioned, the Turing wrappers can often limit the capabilities of
 However, the native HMC sampler within Turing only allows the user to specify the type of the mass matrix despite the two options being possible within `AdvancedHMC`.
 Thankfully, we can use Turing's support for external samplers to define an HMC sampler with a custom mass matrix in `AdvancedHMC` and then use it to sample our Turing model.
 
-We will use the library `Pathfinder`[^2] ((`Pathfinder`'s GitHub)[https://github.com/mlcolab/Pathfinder.jl]) to construct our estimate of mass matrix.
+We can use the library `Pathfinder`[^2] ([`Pathfinder`'s GitHub](https://github.com/mlcolab/Pathfinder.jl)) to construct our estimate of mass matrix.
 `Pathfinder` is a variational inference algorithm that first finds the maximum a posteriori (MAP) estimate of a target posterior distribution and then uses the trace of the optimization to construct a sequence of multivariate normal approximations to the target distribution.
 In this process, `Pathfinder` computes an estimate of the mass matrix the user can access.
-
-The code below shows this can be done in practice.
-
-```{julia}
-using AdvancedHMC, Pathfinder
-# Running pathfinder
-draws = 1_000
-result_multi = multipathfinder(model, draws; nruns=8)
-
-# Estimating the metric
-inv_metric = result_multi.pathfinder_results[1].fit_distribution.Σ
-metric = DenseEuclideanMetric(Matrix(inv_metric))
-
-# Creating an AdvancedHMC NUTS sampler with the custom metric.
-n_adapts = 1000 # Number of adaptation steps
-tap = 0.9 # Large target acceptance probability to deal with the funnel structure of the posterior
-nuts = AdvancedHMC.NUTS(tap; metric=metric)
-
-# Sample
-chain = sample(model, externalsampler(nuts), 10_000; n_adapts=1_000)
-```
+You can see an example of how to use `Pathfinder` with Turing in [`Pathfinder`'s docs](https://mlcolab.github.io/Pathfinder.jl/stable/examples/turing/).
 
 ## Using new inference methods
 

diff --git a/usage/sampler-visualisation/index.qmd b/usage/sampler-visualisation/index.qmd
@@ -117,7 +117,7 @@ setprogress!(false)
 Gibbs sampling tends to exhibit a "jittery" trajectory. The example below combines `HMC` and `PG` sampling to traverse the posterior.
 
 ```{julia}
-c = sample(model, Gibbs(HMC(0.01, 5, :s²), PG(20, :m)), 1000)
+c = sample(model, Gibbs(:s² => HMC(0.01, 5), :m => PG(20)), 1000)
 plot_sampler(c)
 ```