Merge branch 'master' into feature/logistic-offset

Author: quietlychris

.github/workflows/checking.yml

@@ -10,7 +10,7 @@ jobs:
       fail-fast: false
       matrix:
         toolchain:
-          - 1.87.0
+          - 1.88.0
           - stable
           - nightly
         os:
@@ -39,4 +39,25 @@ jobs:
         run: cargo check --workspace --all-targets
 
       - name: Run cargo check (with serde)
-        run: cargo check --workspace --all-targets --features "linfa-clustering/serde linfa-ica/serde linfa-kernel/serde linfa-reduction/serde linfa-svm/serde linfa-elasticnet/serde linfa-pls/serde linfa-trees/serde linfa-nn/serde linfa-linear/serde linfa-preprocessing/serde linfa-bayes/serde linfa-logistic/serde linfa-ftrl/serde"
+        run: cargo check --workspace --all-targets --features "linfa-clustering/serde linfa-ica/serde linfa-kernel/serde linfa-reduction/serde linfa-svm/serde linfa-elasticnet/serde linfa-pls/serde linfa-trees/serde linfa-nn/serde linfa-linear/serde linfa-preprocessing/serde linfa-bayes/serde linfa-logistic/serde linfa-ftrl/serde linfa-ensemble/serde"
+
+  check-wasm-browser:
+    name: check-wasm-browser
+    runs-on: ubuntu-latest
+
+    steps:
+      - name: Checkout sources
+        uses: actions/checkout@master
+
+      - name: Install toolchain
+        uses: dtolnay/rust-toolchain@master
+        with:
+          toolchain: stable
+          targets: wasm32-unknown-unknown
+
+      - name: Run browser WASM checks
+        run: |
+          cargo check -p linfa --lib --target wasm32-unknown-unknown --features wasm-bindgen
+          cargo check -p linfa-linear --lib --target wasm32-unknown-unknown --features wasm-bindgen
+          cargo check -p linfa-logistic --lib --target wasm32-unknown-unknown --features wasm-bindgen
+          cargo check -p linfa-ftrl --lib --target wasm32-unknown-unknown --features wasm-bindgen

.github/workflows/docs.yml

@@ -10,14 +10,12 @@ name: Build website with Zola, build rust docs and publish to GH pages
 jobs:
   build:
     runs-on: ubuntu-latest
-    if: github.ref != 'refs/heads/master' && github.repository == 'rust-ml/linfa'
+    if: github.ref != 'refs/heads/master' || github.repository != 'rust-ml/linfa'
     steps:
-      - name: 'Checkout'
+      - name: Checkout
         uses: actions/checkout@master
 
-
-
-      - name: 'Build only' 
+      - name: Build only 
         uses: shalzz/zola-deploy-action@v0.22.1
         env:
           BUILD_DIR: docs/website/
@@ -31,18 +29,16 @@ jobs:
 
   build_and_deploy:
     runs-on: ubuntu-latest
-    if: github.ref == 'refs/heads/master' || github.repository != 'rust-ml/linfa'
+    if: github.ref == 'refs/heads/master' && github.repository == 'rust-ml/linfa'
     steps:
-      - name: 'Checkout'
+      - name: Checkout
         uses: actions/checkout@master
 
       - name: Install Rust toolchain
         uses: dtolnay/rust-toolchain@stable
         with:
           components: rustfmt, rust-src
 
-
-
       - name: Build Documentation
         run: cargo doc --workspace --no-deps
         env:
@@ -51,7 +47,7 @@ jobs:
       - name: Copy Rust Documentation to Zola
         run: cp -R "target/doc/" "docs/website/static/rustdocs/"
 
-      - name: 'Build and deploy'
+      - name: Build and deploy website
         uses: shalzz/zola-deploy-action@master
         env:
           PAGES_BRANCH: gh-pages

.github/workflows/testing.yml

@@ -10,7 +10,7 @@ jobs:
       fail-fast: false
       matrix:
         toolchain:
-          - 1.87.0
+          - 1.88.0
           - stable
         os:
           - ubuntu-latest
@@ -37,7 +37,7 @@ jobs:
       fail-fast: false
       matrix:
         toolchain:
-          - 1.87.0
+          - 1.88.0
           - stable
         os:
           - ubuntu-latest

Cargo.toml

@@ -34,11 +34,13 @@ intel-mkl-system = ["blas", "ndarray-linalg/intel-mkl-system"]
 blas = ["ndarray/blas"]
 
 serde = ["serde_crate", "ndarray/serde"]
+wasm-bindgen = ["dep:getrandom"]
 
 [dependencies]
 num-traits = "0.2"
 rand = { version = "0.8", features = ["small_rng"] }
 approx = "0.5"
+getrandom = { version = "0.2", optional = true, features = ["js"] }
 
 ndarray = { version = "0.16", features = ["approx"] }
 ndarray-linalg = { version = "0.17", optional = true }
@@ -75,6 +77,7 @@ pprof = { version = "0.15", features = [
 
 [workspace]
 members = ["algorithms/*", "datasets"]
+resolver = "2"
 
 [profile.release]
 opt-level = 3

README.md

@@ -53,6 +53,10 @@ We believe that only a significant community effort can nurture, build, and sust
 
 If this strikes a chord with you, please take a look at the [roadmap](https://github.com/rust-ml/linfa/issues/7) and get involved!
 
+## Browser WASM
+
+For browser-style WASM on `wasm32-unknown-unknown`, enable `linfa`'s `wasm-bindgen` feature.
+
 ## BLAS/Lapack backend
 
 Some algorithm crates need to use an external library for linear algebra routines. By default, we use a pure-Rust implementation. However, you can also choose an external BLAS/LAPACK backend library instead, by enabling the `blas` feature and a feature corresponding to your BLAS backend. Currently you can choose between the following BLAS/LAPACK backends: `openblas`, `netblas` or `intel-mkl`.

algorithms/linfa-bayes/src/multinomial_nb.rs

@@ -74,8 +74,6 @@ where
                 .entry(class.clone())
                 .or_insert_with(ClassHistogram::default)
                 .update_with_smoothing(xclass.view(), self.alpha(), false);
-
-            dbg!(&model.class_info.get(&class));
         }
 
         // update priors

algorithms/linfa-clustering/benches/k_means.rs

@@ -5,7 +5,7 @@ use criterion::{
 use linfa::benchmarks::config;
 use linfa::prelude::*;
 use linfa::DatasetBase;
-use linfa_clustering::{IncrKMeansError, KMeans, KMeansInit};
+use linfa_clustering::{IncrKMeansError, KMeans, KMeansAlgorithm, KMeansInit};
 use linfa_datasets::generate;
 use ndarray::Array2;
 use ndarray_rand::RandomExt;
@@ -36,9 +36,10 @@ impl Drop for Stats {
 fn k_means_bench(c: &mut Criterion) {
     let mut rng = Xoshiro256Plus::seed_from_u64(40);
     let cluster_sizes = [(100, 4), (400, 10), (3000, 10)];
+    let algorithms = [KMeansAlgorithm::Lloyd, KMeansAlgorithm::Hamerly];
     let n_features = 3;
 
-    let mut benchmark = c.benchmark_group("naive_k_means");
+    let mut benchmark = c.benchmark_group("k_means");
     config::set_default_benchmark_configs(&mut benchmark);
     benchmark.plot_config(PlotConfiguration::default().summary_scale(AxisScale::Logarithmic));
 
@@ -47,22 +48,29 @@ fn k_means_bench(c: &mut Criterion) {
         let centroids =
             Array2::random_using((n_clusters, n_features), Uniform::new(-30., 30.), rng);
         let dataset = DatasetBase::from(generate::blobs(cluster_size, &centroids, rng));
-        let mut stats = Stats::default();
+        for &algorithm in &algorithms {
+            let mut stats = Stats::default();
 
-        benchmark.bench_function(
-            BenchmarkId::new("naive_k_means", format!("{n_clusters}x{cluster_size}")),
-            |bencher| {
-                bencher.iter(|| {
-                    let m = KMeans::params_with_rng(black_box(n_clusters), black_box(rng.clone()))
-                        .init_method(KMeansInit::KMeansPlusPlus)
-                        .max_n_iterations(black_box(1000))
-                        .tolerance(black_box(1e-3))
-                        .fit(&dataset)
-                        .unwrap();
-                    stats.add(m.inertia());
-                });
-            },
-        );
+            benchmark.bench_function(
+                BenchmarkId::new(
+                    "k_means",
+                    format!("{algorithm:?}:{n_clusters}x{cluster_size}"),
+                ),
+                |bencher| {
+                    bencher.iter(|| {
+                        let m =
+                            KMeans::params_with_rng(black_box(n_clusters), black_box(rng.clone()))
+                                .init_method(KMeansInit::KMeansPlusPlus)
+                                .algorithm(algorithm)
+                                .max_n_iterations(black_box(1000))
+                                .tolerance(black_box(1e-3))
+                                .fit(&dataset)
+                                .unwrap();
+                        stats.add(m.inertia());
+                    });
+                },
+            );
+        }
     }
 
     benchmark.finish();

algorithms/linfa-clustering/src/dbscan/algorithm.rs

@@ -131,7 +131,7 @@ impl<F: Float, D: Data<Elem = F>, DF: Distance<F>, N: NearestNeighbour>
                 continue;
             }
             neighbors.iter().for_each(|&n| search_found[n] = true);
-            search_queue.extend(neighbors.into_iter());
+            search_queue.extend(neighbors);
 
             // Now go over the neighbours adding them to the cluster
             cluster_memberships[i] = Some(current_cluster_id);

algorithms/linfa-clustering/src/gaussian_mixture/algorithm.rs

@@ -341,10 +341,11 @@ impl<F: Float> GaussianMixtureModel<F> {
         observations: &ArrayBase<D, Ix2>,
     ) -> (Array1<F>, Array2<F>) {
         let weighted_log_prob = self.estimate_weighted_log_prob(observations);
-        let log_prob_norm = weighted_log_prob
-            .mapv(|x| x.exp())
-            .sum_axis(Axis(1))
-            .mapv(|x| x.ln());
+        // Log-sum-exp trick to avoid overflow: ln(Σ exp(xᵢ)) = max + ln(Σ exp(xᵢ - max))
+        let log_max = weighted_log_prob
+            .map_axis(Axis(1), |row| row.fold(F::neg_infinity(), |a, &b| a.max(b)));
+        let shifted = &weighted_log_prob - &log_max.clone().insert_axis(Axis(1));
+        let log_prob_norm = shifted.mapv(|x| x.exp()).sum_axis(Axis(1)).mapv(|x| x.ln()) + &log_max;
         let log_resp = weighted_log_prob - log_prob_norm.to_owned().insert_axis(Axis(1));
         (log_prob_norm, log_resp)
     }