feat(benchmark): add get_version() method to Benchmark class

browsergym/experiments/src/browsergym/experiments/benchmark/base.py

@@ -1,4 +1,5 @@
 import fnmatch
+import importlib.metadata
 import logging
 import random
 import typing
@@ -64,6 +65,31 @@ def make_action_set(self):
     "knows",
 ]
 
+_BACKEND_TO_DISTRIBUTIONS: dict[BenchmarkBackend, tuple[str, ...]] = {
+    "miniwob": ("browsergym.miniwob", "browsergym-miniwob"),
+    "workarena": ("browsergym.workarena", "browsergym-workarena"),
+    "webarena": ("browsergym.webarena", "browsergym-webarena"),
+    "webarena_verified": ("browsergym.webarena_verified", "browsergym-webarena-verified"),
+    "visualwebarena": ("browsergym.visualwebarena", "browsergym-visualwebarena"),
+    "assistantbench": ("browsergym.assistantbench", "browsergym-assistantbench"),
+    "weblinx": ("weblinx_browsergym", "weblinx-browsergym"),
+    "timewarp": ("browsergym.timewarp", "browsergym-timewarp"),
+    "knows": ("browsergym.knows", "browsergym-knows"),
+}
+
+_PREFIX_TO_DISTRIBUTIONS: list[tuple[str, tuple[str, ...]]] = [
+    ("webarena_verified", ("browsergym.webarena_verified", "browsergym-webarena-verified")),
+    ("visualwebarena", ("browsergym.visualwebarena", "browsergym-visualwebarena")),
+    ("webarena", ("browsergym.webarena", "browsergym-webarena")),
+    ("webarenalite", ("browsergym.webarena", "browsergym-webarena")),
+    ("miniwob", ("browsergym.miniwob", "browsergym-miniwob")),
+    ("workarena", ("browsergym.workarena", "browsergym-workarena")),
+    ("assistantbench", ("browsergym.assistantbench", "browsergym-assistantbench")),
+    ("weblinx", ("weblinx_browsergym", "weblinx-browsergym")),
+    ("timewarp", ("browsergym.timewarp", "browsergym-timewarp")),
+    ("knows", ("browsergym.knows", "browsergym-knows")),
+]
+
 
 @dataclass
 class Benchmark(DataClassJsonMixin):
@@ -104,6 +130,43 @@ def prepare_backends(self):
             prepare_backend(backend)
             logger.info(f"{backend} backend ready")
 
+    def get_version(self) -> str:
+        """
+        Return the version of the benchmark based on its backend or name.
+        """
+        dist_names = None
+        is_weblinx = False
+
+        for backend in self.backends:
+            if backend in _BACKEND_TO_DISTRIBUTIONS:
+                dist_names = _BACKEND_TO_DISTRIBUTIONS[backend]
+                if backend == "weblinx":
+                    is_weblinx = True
+                break
+
+        if dist_names is None:
+            name_lower = self.name.lower()
+            for prefix, dists in _PREFIX_TO_DISTRIBUTIONS:
+                if name_lower.startswith(prefix):
+                    dist_names = dists
+                    if prefix == "weblinx":
+                        is_weblinx = True
+                    break
+
+        if dist_names is None:
+            raise ValueError(f"Unknown benchmark {self.name}")
+
+        for dist_name in dist_names:
+            try:
+                return importlib.metadata.distribution(dist_name).version
+            except importlib.metadata.PackageNotFoundError:
+                continue
+
+        if is_weblinx:
+            return "0.0.1rc1"
+
+        raise importlib.metadata.PackageNotFoundError(dist_names[0])
+
     def subset_from_split(self, split: Literal["train", "valid", "test"]):
         split_column = "browsergym_split"
 

tests/experiments/test_benchmark.py

@@ -1,4 +1,5 @@
 import dataclasses
+import importlib.metadata
 import os
 import random
 import re
@@ -353,3 +354,74 @@ def test_dependency_graphs():
     assert [{k: set(v) for k, v in deps.items()} for deps in env_args_dependencies] == [
         {k: set(v) for k, v in {0: [], 1: [], 2: [1], 3: [0, 2], 4: [3], 5: [3, 4]}.items()}
     ]
+
+
+def test_benchmark_get_version():
+    from unittest.mock import MagicMock, patch
+
+    mock_dist = MagicMock()
+    mock_dist.version = "1.2.3"
+
+    # Test standard benchmarks with mocked distributions
+    with patch("importlib.metadata.distribution", return_value=mock_dist) as mock_metadata:
+        miniwob_bench = DEFAULT_BENCHMARKS["miniwob"]()
+        assert miniwob_bench.get_version() == "1.2.3"
+        mock_metadata.assert_called_with("browsergym.miniwob")
+
+    with patch("importlib.metadata.distribution", return_value=mock_dist) as mock_metadata:
+        webarena_bench = DEFAULT_BENCHMARKS["webarena"]()
+        assert webarena_bench.get_version() == "1.2.3"
+        mock_metadata.assert_called_with("browsergym.webarena")
+
+    workarena_bench = Benchmark(
+        name="workarena_l1",
+        high_level_action_set_args=DEFAULT_BENCHMARKS["miniwob_tiny_test"]().high_level_action_set_args,
+        is_multi_tab=False,
+        supports_parallel_seeds=True,
+        env_args_list=DEFAULT_BENCHMARKS["miniwob_tiny_test"]().env_args_list,
+        backends=["workarena"],
+    )
+    with patch("importlib.metadata.distribution", return_value=mock_dist) as mock_metadata:
+        assert workarena_bench.get_version() == "1.2.3"
+        mock_metadata.assert_called_with("browsergym.workarena")
+
+    with patch("importlib.metadata.distribution", return_value=mock_dist) as mock_metadata:
+        visualwebarena_bench = DEFAULT_BENCHMARKS["visualwebarena"]()
+        assert visualwebarena_bench.get_version() == "1.2.3"
+        mock_metadata.assert_called_with("browsergym.visualwebarena")
+
+    with patch("importlib.metadata.distribution", return_value=mock_dist) as mock_metadata:
+        assistantbench = DEFAULT_BENCHMARKS["assistantbench"]()
+        assert assistantbench.get_version() == "1.2.3"
+        mock_metadata.assert_called_with("browsergym.assistantbench")
+
+    # Test weblinx fallback to 0.0.1rc1 when not found
+    with patch(
+        "importlib.metadata.distribution",
+        side_effect=importlib.metadata.PackageNotFoundError("weblinx_browsergym"),
+    ):
+        weblinx_bench = DEFAULT_BENCHMARKS["weblinx"]()
+        assert weblinx_bench.get_version() == "0.0.1rc1"
+
+    # Test weblinx returns version when package is found
+    with patch("importlib.metadata.distribution", return_value=mock_dist):
+        weblinx_bench = DEFAULT_BENCHMARKS["weblinx"]()
+        assert weblinx_bench.get_version() == "1.2.3"
+
+    # Test unknown benchmark raises ValueError
+    fake_bench = Benchmark(
+        name="unknown_benchmark",
+        high_level_action_set_args=DEFAULT_BENCHMARKS["miniwob_tiny_test"]().high_level_action_set_args,
+        is_multi_tab=False,
+        supports_parallel_seeds=True,
+        env_args_list=DEFAULT_BENCHMARKS["miniwob_tiny_test"]().env_args_list,
+        backends=[],
+    )
+    with pytest.raises(ValueError, match="Unknown benchmark unknown_benchmark"):
+        fake_bench.get_version()
+
+    # Test benchmark subset preserves get_version
+    with patch("importlib.metadata.distribution", return_value=mock_dist):
+        sub_bench = miniwob_bench.subset_from_regexp(column="task_name", regexp="click")
+        assert sub_bench.get_version() == "1.2.3"
+