Merge branch 'main' into main

Makefile

@@ -33,7 +33,6 @@ clean-build: ## remove build artifacts
 	rm -fr dist/
 	rm -fr .eggs/
 	find . -name '*.egg-info' -exec rm -fr {} +
-	find . -name '*.egg' -exec rm -f {} +
 
 clean-pyc: ## remove Python file artifacts
 	find . -name '*.pyc' -exec rm -f {} +
@@ -51,11 +50,11 @@ clean-docs: ## remove docs artifacts
 	cd docs && make clean
 
 ruff: ## run ruff as a formatter
-	python -m ruff format hydra_plugins
-	python -m ruff check --silent --exit-zero --no-cache --fix hydra_plugins
-	python -m ruff check --exit-zero hydra_plugins
+	uvx ruff format hydra_plugins
+	uvx ruff check --silent --exit-zero --no-cache --fix hydra_plugins
+	uvx ruff check --exit-zero hydra_plugins
 isort:
-	python -m isort hydra_plugins tests
+	uvx isort hydra_plugins tests
 
 test: ## run tests quickly with the default Python
 	python -m pytest tests
@@ -99,13 +98,13 @@ dist: clean ## builds source and wheel package
 	python setup.py bdist_wheel
 	ls -l dist
 install: clean ## install the package to the active Python's site-packages
-	pip install -e . --config-settings editable_mode=compat
+	uv pip install -e . --config-settings editable_mode=compat
 
 install-dev: clean ## install the package to the active Python's site-packages
-	pip install -e ".[dev,examples,doc,all]" --config-settings editable_mode=compat
+	uv pip install -e ".[dev,examples,doc,all]" --config-settings editable_mode=compat
 
 check:
-	pre-commit run --all-files
+	uvx pre-commit run --all-files
 
 format:
 	make ruff

README.md

@@ -16,22 +16,15 @@ We recommend installing hypersweeper via a uv virtual environment:
 
 ```bash
 pip install uv
-uv init
-uv sync
+uv venv --python 3.10
+make install
 ```
 
-For extra dependencies, add them to the sync command like this:
+For extra dependencies, add them like this:
 ```bash
 uv sync --extra dev --extra carps
 ```
 
-Alternatively you can also install in a fresh conda environment:
-
-```bash
-conda create -n hypersweeper python=3.10
-make install
-```
-
 ## Basic Usage
 
 To use the sweeper, you need to specify a target function with a hydra interface (see our examples). 

examples/configs/branin_carps_hebo.yaml

@@ -16,6 +16,6 @@ hydra:
         task: ${carps_task}
     search_space: ${search_space}
   run:
-    dir: ./tmp/${now:%Y-%m-%d}/${now:%H-%M-%S}
+    dir: ./tmp/branin_carps_hebo/
   sweep:
-    dir: ./tmp/${now:%Y-%m-%d}/${now:%H-%M-%S}
+    dir: ./tmp/branin_carps_hebo/

examples/configs/branin_carps_smac.yaml

@@ -40,6 +40,6 @@ hydra:
               _partial_: true
               eta: 3
   run:
-    dir: ./tmp/${now:%Y-%m-%d}/${now:%H-%M-%S}
+    dir: ./tmp/branin_carps_smac/
   sweep:
-    dir: ./tmp/${now:%Y-%m-%d}/${now:%H-%M-%S}
+    dir: ./tmp/branin_carps_smac/

examples/configs/branin_nevergrad.yaml

@@ -17,6 +17,6 @@ hydra:
           budget: 100
     search_space: ${search_space}
   run:
-    dir: ./tmp/${now:%Y-%m-%d}/${now:%H-%M-%S}
+    dir: ./tmp/branin_nevergrad/
   sweep:
-    dir: ./tmp/${now:%Y-%m-%d}/${now:%H-%M-%S}
+    dir: ./tmp/branin_nevergrad/

examples/configs/branin_rs.yaml

@@ -13,6 +13,6 @@ hydra:
       max_budget: 100
     search_space: ${search_space}
   run:
-    dir: ./tmp/${now:%Y-%m-%d}/${now:%H-%M-%S}
+    dir: ./tmp/branin_rs/
   sweep:
-    dir: ./tmp/${now:%Y-%m-%d}/${now:%H-%M-%S}
+    dir: ./tmp/branin_rs/

examples/configs/branin_smac.yaml

@@ -24,6 +24,6 @@ hydra:
           output_directory: ${hydra.sweep.dir}
     search_space: ${search_space}
   run:
-    dir: ./tmp/${now:%Y-%m-%d}/${now:%H-%M-%S}
+    dir: ./tmp/branin_smac/
   sweep:
-    dir: ./tmp/${now:%Y-%m-%d}/${now:%H-%M-%S}
+    dir: ./tmp/branin_smac/

examples/configs/mlp_carps_smac.yaml

@@ -35,6 +35,6 @@ hydra:
               eta: 3
     search_space: ${search_space}
   run:
-    dir: ./tmp/${now:%Y-%m-%d}/${now:%H-%M-%S}
+    dir: ./tmp/mlp_carps_smac/
   sweep:
-    dir: ./tmp/${now:%Y-%m-%d}/${now:%H-%M-%S}
+    dir: ./tmp/mlp_carps_smac/

examples/configs/mlp_dehb.yaml

@@ -20,6 +20,6 @@ hydra:
           output_path: ${hydra.sweep.dir}
     search_space: ${search_space}
   run:
-    dir: ./tmp/${now:%Y-%m-%d}/${now:%H-%M-%S}
+    dir: ./tmp/mlp_dehb/
   sweep:
-    dir: ./tmp/${now:%Y-%m-%d}/${now:%H-%M-%S}
+    dir: ./tmp/mlp_dehb/

examples/configs/mlp_hebo.yaml

@@ -11,6 +11,6 @@ hydra:
     n_trials: 10
     search_space: ${search_space}
   run:
-    dir: ./tmp/${now:%Y-%m-%d}/${now:%H-%M-%S}
+    dir: ./tmp/mlp_hebo/
   sweep:
-    dir: ./tmp/${now:%Y-%m-%d}/${now:%H-%M-%S}
+    dir: ./tmp/mlp_hebo/

examples/configs/mlp_smac.yaml

@@ -28,6 +28,6 @@ hydra:
           output_directory: ${hydra.sweep.dir}
     search_space: ${search_space}
   run:
-    dir: ./tmp/${now:%Y-%m-%d}/${now:%H-%M-%S}
+    dir: ./tmp/mlp_smac/
   sweep:
-    dir: ./tmp/${now:%Y-%m-%d}/${now:%H-%M-%S}
+    dir: ./tmp/mlp_smac/

examples/configs/sac_pb2.yaml

@@ -20,6 +20,6 @@ hydra:
       load_tf: true
     search_space: ${search_space}
   run:
-    dir: ./tmp/${now:%Y-%m-%d}/${now:%H-%M-%S}
+    dir: ./tmp/sac_pb2/
   sweep:
-    dir: ./tmp/${now:%Y-%m-%d}/${now:%H-%M-%S}
+    dir: ./tmp/sac_pb2/

examples/configs/sac_pbt.yaml

@@ -19,6 +19,6 @@ hydra:
       load_tf: true
     search_space: ${search_space}
   run:
-    dir: ./tmp/${now:%Y-%m-%d}/${now:%H-%M-%S}
+    dir: ./tmp/sac_pbt/
   sweep:
-    dir: ./tmp/${now:%Y-%m-%d}/${now:%H-%M-%S}
+    dir: ./tmp/sac_pbt/

hydra_plugins/hyper_pbt/bg_pbt_utils.py

@@ -1,4 +1,5 @@
 """All of this is copied/lightly adapted from the original BG-PBT code: https://github.com/xingchenwan/bgpbt."""
+
 from __future__ import annotations
 
 import logging
@@ -122,8 +123,8 @@ def get_start_point(cs: CS.ConfigurationSpace, x_center, frozen_dims: list[int]
         # print(param_name)
         if np.isnan(new_config_array[i]) or (frozen_dims is not None and i in frozen_dims):
             continue
-        param_name = cs.get_hyperparameter_by_idx(i)
-        if type(cs[param_name]) == CSH.CategoricalHyperparameter:
+        param_name = cs.at(i)
+        if isinstance(cs[param_name], CSH.CategoricalHyperparameter):
             new_config_array[i] = rng.choice(range(len(cs[param_name].choices)))
         elif (
             type(cs[param_name]) in [CSH.UniformIntegerHyperparameter, CSH.NormalIntegerHyperparameter]
@@ -139,9 +140,9 @@ def get_start_point(cs: CS.ConfigurationSpace, x_center, frozen_dims: list[int]
     config = deactivate_inactive_hyperparameters(config, cs)
 
     try:
-        cs.check_configuration(config)
+        config.check_valid_configuration()
     except ValueError:
-        config = CS.Configuration(cs, config.get_dictionary())
+        config = CS.Configuration(cs, dict(config))
     new_config_array = config.get_array()
     if return_config:
         return new_config_array, config
@@ -167,23 +168,23 @@ def construct_bounding_box(
     weights = weights / np.prod(np.power(weights, 1.0 / len(weights)))
     lb, ub = np.zeros_like(x), np.ones_like(x)
     for i, _dim in enumerate(x):
-        if np.isnan(x[i]) or i >= len(cs):
+        if np.isnan(_dim) or i >= len(cs):
             lb[i], ub[i] = 0.0, 1.0
         else:
-            hp = cs[cs.get_hyperparameter_by_idx(i)]
-            if type(hp) == CSH.CategoricalHyperparameter:
+            hp = cs[cs.at(i)]
+            if isinstance(hp, CSH.CategoricalHyperparameter):
                 lb[i], ub[i] = 0, len(hp.choices)
             else:
-                lb[i] = np.clip(x[i] - weights[i] * tr_length / 2.0, 0.0, 1.0)
-                ub[i] = np.clip(x[i] + weights[i] * tr_length / 2.0, 0.0, 1.0)
-                if type(hp) in [
-                    CSH.UniformIntegerHyperparameter,
-                    CSH.NormalIntegerHyperparameter,
-                    CSH.NormalFloatHyperparameter,
-                    CSH.UniformFloatHyperparameter,
+                lb[i] = np.clip(_dim - weights[i] * tr_length / 2.0, 0.0, 1.0)
+                ub[i] = np.clip(_dim + weights[i] * tr_length / 2.0, 0.0, 1.0)
+                if any[
+                    isinstance(hp, CSH.UniformIntegerHyperparameter),
+                    isinstance(hp, CSH.NormalIntegerHyperparameter),
+                    isinstance(hp, CSH.NormalFloatHyperparameter),
+                    isinstance(hp, CSH.UniformFloatHyperparameter),
                 ]:
-                    lb[i] = max(hp._inverse_transform(hp.lower), lb[i])
-                    ub[i] = min(hp._inverse_transform(hp.upper), ub[i])
+                    lb[i] = max(hp.to_vector(hp.lower), lb[i])
+                    ub[i] = min(hp.to_vector(hp.upper), ub[i])
     return lb, ub
 
 
@@ -199,20 +200,20 @@ def get_dim_info(cs: CS.ConfigurationSpace, x, return_indices=False):
         # do not sample an inactivated hyperparameter (such a hyperparameter has nan value imputed)
         if x[variable] != x[variable]:
             continue
-        if type(cs[cs.get_hyperparameter_by_idx(variable)]) == CSH.CategoricalHyperparameter:
-            cat_dims.append(cs.get_hyperparameter_by_idx(variable))
+        if isinstance(cs[cs.at(variable)], CSH.CategoricalHyperparameter):
+            cat_dims.append(cs.at(variable))
             cat_dims_idx.append(i)
-        elif type(cs[cs.get_hyperparameter_by_idx(variable)]) in [
+        elif type(cs[cs.at(variable)]) in [
             CSH.UniformIntegerHyperparameter,
             CSH.NormalIntegerHyperparameter,
         ]:
-            int_dims.append(cs.get_hyperparameter_by_idx(variable))
+            int_dims.append(cs.at(variable))
             int_dims_idx.append(i)
-        elif type(cs[cs.get_hyperparameter_by_idx(variable)]) in [
+        elif type(cs[cs.at(variable)]) in [
             CSH.UniformFloatHyperparameter,
             CSH.NormalFloatHyperparameter,
         ]:
-            cont_dims.append(cs.get_hyperparameter_by_idx(variable))
+            cont_dims.append(cs.at(variable))
             cont_dims_idx.append(i)
     if return_indices:
         return cat_dims_idx, cont_dims_idx, int_dims_idx
@@ -238,21 +239,21 @@ def sample_discrete_neighbour(cs: CS.ConfigurationSpace, x, frozen_dims: list[in
     config = CS.Configuration(cs, vector=x.detach().numpy() if isinstance(x, torch.Tensor) else x)
 
     try:
-        cs.check_configuration(config)
+        config.check_valid_configuration()
     except ValueError:
         # there seems to be a bug with ConfigSpace that raises error even when a config is valid
         # Issue #196: https://github.com/automl/ConfigSpace/issues/196
         # print(config)
-        config = CS.Configuration(cs, config.get_dictionary())
+        config = CS.Configuration(cs, dict(config))
 
     # print(config)
     config_pert = deepcopy(config)
     rng = np.random.default_rng()
     selected_dim = str(rng.choice(cat_dims + int_dims, 1)[0])
-    index_in_array = cs.get_idx_by_hyperparameter_name(selected_dim)
+    index_in_array = cs.index_of(selected_dim)
     while config_pert[selected_dim] is None or (frozen_dims is not None and index_in_array in frozen_dims):
         selected_dim = str(rng.choice(cat_dims + int_dims, 1)[0])
-        index_in_array = cs.get_idx_by_hyperparameter_name(selected_dim)
+        index_in_array = cs.index_of(selected_dim)
 
     # if the selected dimension is categorical, change the value to another variable
     if selected_dim in cat_dims:
@@ -629,12 +630,10 @@ def __init__(
         # extract the dim indices of the continuous dimensions (incl. integers)
         self.cont_dims = [
             i
-            for i, dim in enumerate(self.cs.get_hyperparameters())
+            for i, dim in enumerate(self.cs.values())
             if type(dim) in [CSH.UniformIntegerHyperparameter, CSH.UniformFloatHyperparameter]
         ]
-        self.cat_dims = [
-            i for i, dim in enumerate(self.cs.get_hyperparameters()) if type(dim) == CSH.CategoricalHyperparameter
-        ]
+        self.cat_dims = [i for i, dim in enumerate(self.cs.values()) if isinstance(dim, CSH.CategoricalHyperparameter)]
 
         # initialise the kernels
         self.continuous_kern = ConditionalMatern(
@@ -794,7 +793,7 @@ def rbf(d, ard):
             if exp == "rbf":
                 k_cat = rbf(diff1, self.ard_num_dims is not None and self.ard_num_dims > 1)
             else:
-                raise ValueError("Exponentiation scheme %s is not recognised!" % exp)
+                raise ValueError(f"Exponentiation scheme {exp} is not recognised!")
             if diag:
                 return torch.diag(k_cat).float()
         return k_cat.float()

hydra_plugins/hyper_pbt/pb2_utils.py

@@ -33,8 +33,7 @@ def __init__(self, input_dim, variance=1.0, lengthscale=1.0, epsilon=0.0, active
     def K(self, X, X2):
         """Compute the kernel."""
         # time must be in the far left column
-        if self.epsilon > 0.5:  # noqa: PLR2004
-            self.epsilon = 0.5
+        self.epsilon = min(self.epsilon, 0.5)
         if X2 is None:
             X2 = np.copy(X)
         T1 = X[:, 0].reshape(-1, 1)
@@ -188,11 +187,9 @@ def K2(self, x, x2):
     def K(self, x, x2):
         """Compute the kernel."""
         # clip epsilons
-        if self.epsilon_1 > 0.5:  # noqa: PLR2004
-            self.epsilon_1 = 0.5
+        self.epsilon_1 = min(self.epsilon_1, 0.5)
 
-        if self.epsilon_2 > 0.5:  # noqa: PLR2004
-            self.epsilon_2 = 0.5
+        self.epsilon_2 = min(self.epsilon_2, 0.5)
 
         # format data
         if x2 is None:
@@ -538,7 +535,7 @@ def exp3_get_cat(row, data, num_rounds, index):
         sum_w = np.sum(weights)
         weights = [w / sum_w for w in weights]
 
-        count += 1
+        count += 1  # noqa: SIM113
 
     # now we select our arm!
 

hydra_plugins/hypersweeper/hypersweeper_sweeper.py

@@ -277,8 +277,6 @@ def run_configs(self, infos):
             for j in range(len(overrides)):
                 performances.append(res[j].return_value)
                 self.trials_run += 1
-        if self.maximize:
-            performances = [-p for p in performances]
         return performances, costs
 
     def get_save_path(self, config_id, seed=None):
@@ -471,7 +469,6 @@ def run(self, verbose=False):
                     performance = float(np.mean(performance))
 
                 logged_performance = -performance if self.maximize else performance
-
                 value = Result(performance=logged_performance, cost=cost)
                 self.optimizer.tell(info=info, value=value)