Sync master and dev

.github/workflows/docker.yml

@@ -2,19 +2,59 @@ name: Docker Images
 
 on:
   push:
+    branches:
+      - dev
     tags: ['v*']
+  workflow_dispatch:
 
 permissions:
   contents: read
 
 env:
-  ORG: ${{ secrets.DOCKER_HUB_USERNAME }}
+  # Image consumers (Compose and Apptainer) pull from the public srbench
+  # namespace. Keep this independent of the personal Docker Hub account used
+  # to authenticate the workflow.
+  IMAGE_ORG: srbench
 
 jobs:
   base:
+    # Temporarily disabled
+    if: ${{ false }}
     runs-on: ubuntu-latest
     steps:
       - uses: actions/checkout@v4
+      # A dev push gets the development tag. A vX.Y tag is a release: publish
+      # its numeric version and promote it to latest.
+      - name: Select image tags
+        id: image_tags
+        shell: bash
+        run: |
+          set -euo pipefail
+          if [[ "$GITHUB_REF_TYPE" == "tag" ]]; then
+            version="${GITHUB_REF_NAME#v}"
+            if [[ -z "$version" || "$version" == "$GITHUB_REF_NAME" ]]; then
+              echo "Release tags must use the v<version> form (for example v0.1)." >&2
+              exit 1
+            fi
+            echo "base_tag=$version" >> "$GITHUB_OUTPUT"
+            {
+              echo "tags<<EOF"
+              echo "${IMAGE_ORG}/base:latest"
+              echo "${IMAGE_ORG}/base:${version}"
+              echo "EOF"
+            } >> "$GITHUB_OUTPUT"
+          else
+            if [[ "$GITHUB_REF_NAME" != "dev" ]]; then
+              echo "Development images may only be published from the dev branch." >&2
+              exit 1
+            fi
+            echo "base_tag=dev" >> "$GITHUB_OUTPUT"
+            {
+              echo "tags<<EOF"
+              echo "${IMAGE_ORG}/base:dev"
+              echo "EOF"
+            } >> "$GITHUB_OUTPUT"
+          fi
       - uses: docker/login-action@v3
         with:
           username: ${{ secrets.DOCKER_HUB_USERNAME }}
@@ -25,13 +65,13 @@ jobs:
           context: .
           file: baseDockerfile
           push: true
-          tags: |
-            ${{ env.ORG }}/base:latest
-            ${{ env.ORG }}/base:${{ github.ref_name }}
+          tags: ${{ steps.image_tags.outputs.tags }}
           cache-from: type=gha,scope=base
           cache-to: type=gha,scope=base,mode=max
 
   algorithms:
+    # Temporarily disabled
+    if: ${{ false }}
     needs: base
     runs-on: ubuntu-latest
     strategy:
@@ -49,12 +89,16 @@ jobs:
           - ffx
           - geneticengine
           - gpgomea
+          - gp-elite
           - gplearn
           - gpzgd
           - itea
+          - keplearn
           - lightgbm
           - nesymres
           - operon
+          - pantara
+          - pir
           - ps-tree
           - pysr
           - qlattice
@@ -66,6 +110,36 @@ jobs:
           - xgboost
     steps:
       - uses: actions/checkout@v4
+      - name: Select image tags
+        id: image_tags
+        shell: bash
+        run: |
+          set -euo pipefail
+          if [[ "$GITHUB_REF_TYPE" == "tag" ]]; then
+            version="${GITHUB_REF_NAME#v}"
+            if [[ -z "$version" || "$version" == "$GITHUB_REF_NAME" ]]; then
+              echo "Release tags must use the v<version> form (for example v0.1)." >&2
+              exit 1
+            fi
+            echo "base_tag=$version" >> "$GITHUB_OUTPUT"
+            {
+              echo "tags<<EOF"
+              echo "${IMAGE_ORG}/${{ matrix.algorithm }}:latest"
+              echo "${IMAGE_ORG}/${{ matrix.algorithm }}:${version}"
+              echo "EOF"
+            } >> "$GITHUB_OUTPUT"
+          else
+            if [[ "$GITHUB_REF_NAME" != "dev" ]]; then
+              echo "Development images may only be published from the dev branch." >&2
+              exit 1
+            fi
+            echo "base_tag=dev" >> "$GITHUB_OUTPUT"
+            {
+              echo "tags<<EOF"
+              echo "${IMAGE_ORG}/${{ matrix.algorithm }}:dev"
+              echo "EOF"
+            } >> "$GITHUB_OUTPUT"
+          fi
       - uses: docker/login-action@v3
         with:
           username: ${{ secrets.DOCKER_HUB_USERNAME }}
@@ -85,10 +159,8 @@ jobs:
           file: ${{ steps.dockerfile.outputs.path }}
           build-args: |
             ALGORITHM=${{ matrix.algorithm }}
-            BASE_IMAGE=${{ env.ORG }}/base:${{ github.ref_name }}
+            BASE_IMAGE=${{ env.IMAGE_ORG }}/base:${{ steps.image_tags.outputs.base_tag }}
           push: true
-          tags: |
-            ${{ env.ORG }}/${{ matrix.algorithm }}:latest
-            ${{ env.ORG }}/${{ matrix.algorithm }}:${{ github.ref_name }}
+          tags: ${{ steps.image_tags.outputs.tags }}
           cache-from: type=gha,scope=${{ matrix.algorithm }}
           cache-to: type=gha,scope=${{ matrix.algorithm }},mode=max

README.md

@@ -22,19 +22,21 @@ To handle the lack of a unified framework, we've specified minimal requirements
 The current edition of the benchmark (SRBench 2025, reported in our [_call for action_ paper](#call-for-action)) evaluates **25** symbolic regression methods under a unified experimental setup: every method runs from a docker container, with hyperparameter tuning and **30** independent runs per dataset, on **24** datasets from [PMLB](https://github.com/EpistasisLab/penn-ml-benchmarks) plus a set of first-principles regression problems.
 This roster includes the 14 methods from the original SRBench together with the methods staged since then.
 
-Methods currently benchmarked:
+Methods currently benchmarked or staged for the next benchmark update:
 
 | Method |  |  |
 |:--|:--|:--|
 | **AFP** - [paper](http://dx.doi.org/10.1007/978-1-4419-7747-2_8) | **AFP_fe** | **AFP_ehc** - [paper](https://www.sciencedirect.com/science/article/abs/pii/S0952197616301294) |
 | **Bingo** - [paper](https://dl.acm.org/doi/10.1145/3520304.3534031) | **Brush** - [paper](https://royalsocietypublishing.org/rsta/article/384/2317/20240588/481208/Towards-symbolic-regression-for-interpretable) | **BSR** - [paper](https://arxiv.org/abs/1910.08892) |
 | **E2E** - [paper](https://papers.neurips.cc/paper_files/paper/2022/file/42eb37cdbefd7abae0835f4b67548c39-Paper-Conference.pdf) | **EPLEX** - [paper](https://direct.mit.edu/evco/article-pdf/27/3/377/1858632/evco_a_00224.pdf) | **EQL** - [paper](http://proceedings.mlr.press/v80/sahoo18a/sahoo18a.pdf) |
 | **FEAT** - [paper](https://openreview.net/pdf?id=Hke-JhA9Y7) | **FFX** - [paper](https://link.springer.com/chapter/10.1007/978-1-4614-1770-5_13) | **Genetic Engine** - [paper](https://dl.acm.org/doi/10.1145/3564719.3568697) |
-| **GPGomea** - [paper](http://dx.doi.org/10.1162/evco_a_00278) | **GPlearn** - [paper]() | **GPZGD** - [paper](https://doi.org/10.1145/3377930.3390237) |
-| **ITEA** - [paper](https://direct.mit.edu/evco/article-pdf/29/3/367/1959462/evco_a_00285.pdf) | **NeSymRes** - [paper](http://proceedings.mlr.press/v139/biggio21a/biggio21a.pdf) | **Operon** - [paper](https://link.springer.com/article/10.1007/s10710-019-09371-3) |
-| **Ps-Tree** - [paper](https://www.sciencedirect.com/science/article/pii/S2210650222000335) | **PySR** - [paper](https://arxiv.org/abs/2305.01582) | **Qlattice** - [paper](https://arxiv.org/abs/2104.05417) |
-| **Rils-rols** - [paper](http://dx.doi.org/10.1186/s40537-023-00743-2) | **TIR** - [paper](https://doi.org/10.1145/3597312) | **TPSR** - [paper](https://openreview.net/forum?id=0rVXQEeFEL) |
-| **uDSR** - [paper](https://proceedings.neurips.cc/paper_files/paper/2022/file/dbca58f35bddc6e4003b2dd80e42f838-Paper-Conference.pdf) |  |  |
+| **GPGomea** - [paper](http://dx.doi.org/10.1162/evco_a_00278) | **GP-ELITE** - [code](https://github.com/ariel95500-create/gp-elite) | **GPlearn** - [paper]() |
+| **GPZGD** - [paper](https://doi.org/10.1145/3377930.3390237) | **ITEA** - [paper](https://direct.mit.edu/evco/article-pdf/29/3/367/1959462/evco_a_00285.pdf) | **NeSymRes** - [paper](http://proceedings.mlr.press/v139/biggio21a/biggio21a.pdf) |
+| **Operon** - [paper](https://link.springer.com/article/10.1007/s10710-019-09371-3) | **Pantara** - [code](https://github.com/Yapock22/pantara) | **Ps-Tree** - [paper](https://www.sciencedirect.com/science/article/pii/S2210650222000335) |
+| **PySR** - [paper](https://arxiv.org/abs/2305.01582) | **Qlattice** - [paper](https://arxiv.org/abs/2104.05417) | **Rils-rols** - [paper](http://dx.doi.org/10.1186/s40537-023-00743-2) |
+| **TIR** - [paper](https://doi.org/10.1145/3597312) | **TPSR** - [paper](https://openreview.net/forum?id=0rVXQEeFEL) | **uDSR** - [paper](https://proceedings.neurips.cc/paper_files/paper/2022/file/dbca58f35bddc6e4003b2dd80e42f838-Paper-Conference.pdf) |
+
+GP-ELITE and Pantara are staged for benchmarking and are covered by the container build and test workflows; they are not included in the SRBench 2025 result figures above.
 
 The full experiment code and results for this edition live on the [`srbench_2025`](https://github.com/cavalab/srbench/tree/srbench_2025) branch (raw results in [`results/`](https://github.com/cavalab/srbench/tree/srbench_2025/results)).
 If you are choosing baselines for a new symbolic regression paper, please use this roster and these results rather than the 2021 tables below.
@@ -177,4 +179,3 @@ v1.0 was reported in our GECCO 2018 paper:
 # Contact
 
 William La Cava ([@lacava](https://github.com/lacava)), william dot lacava at childrens dot harvard dot edu
-

algorithms/keplearn/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Chandler Freeman
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

algorithms/keplearn/environment.yml

@@ -0,0 +1,9 @@
+channels:
+  - conda-forge
+dependencies:
+  - numpy
+  - sympy
+  - scikit-learn
+  - rust
+  - compilers
+  - git

algorithms/keplearn/install.sh

@@ -0,0 +1,19 @@
+#!/bin/bash
+set -euxo pipefail
+# Build keplearn from the pinned public release tag (no vendored source or
+# binaries in the srbench tree, per CONTRIBUTING). The crate has no
+# dependencies, so the only network fetches are this clone and nothing else;
+# rust/cargo come from conda-forge via environment.yml.
+TAG=v0.1.0
+git clone --depth 1 --branch "$TAG" https://github.com/owls-on-wires/keplearn.git keplearn-src
+cd keplearn-src
+cargo build --release
+install -m 0755 target/release/keplearn "${CONDA_PREFIX}/bin/keplearn"
+cd ..
+rm -rf keplearn-src
+# The toolchain is only needed for this build. Removing it here keeps the
+# image layer near the base size instead of ~3.2 GB, because the environment
+# install and this script run inside the same Docker RUN layer.
+micromamba remove -y -n base rust compilers git
+keplearn --version
+echo "keplearn: installed ${TAG} from source at ${CONDA_PREFIX}/bin/keplearn"

algorithms/keplearn/metadata.yml

@@ -0,0 +1,16 @@
+authors:
+  - Chandler Freeman
+email: [email protected]
+name: keplearn
+description: |
+  Keplearn is a deterministic symbolic regression engine written in Rust. It
+  recovers closed-form equations by best-first recursive reduction: the
+  target is reduced through chains of invertible operations (divide by a
+  fitted factor, target-side shells, fitted inner-affine nodes) until a
+  terminal monomial or composite fit explains it, and candidates are chosen
+  on an internal holdout with an MDL complexity penalty. The engine has no
+  randomness; runs are reproducible byte for byte. install.sh builds it from
+  a pinned release tag of the public repository, and this Python wrapper
+  shells out to the binary (training data as a TSV on stdin, closed-form
+  model as JSON on stdout).
+url: https://github.com/owls-on-wires/keplearn