Update ema-workbench dependency to v3.0.0

.github/workflows/pytest.yml

@@ -18,7 +18,7 @@ jobs:
       - name: Set up Python
         uses: actions/setup-python@v5
         with:
-          python-version: "3.10"
+          python-version: "3.12"
 
       - name: Install Poetry
         run: |

justice/util/EMA_model_wrapper.py

@@ -63,11 +63,11 @@ def model_wrapper_emodps(**kwargs):
     weights = np.zeros(weights_shape)
 
     for i in range(centers_shape[0]):
-        centers[i] = kwargs.pop(f"center {i}")
-        radii[i] = kwargs.pop(f"radii {i}")
+        centers[i] = kwargs.pop(f"center_{i}")
+        radii[i] = kwargs.pop(f"radii_{i}")
 
     for i in range(weights_shape[0]):
-        weights[i] = kwargs.pop(f"weights {i}")
+        weights[i] = kwargs.pop(f"weights_{i}")
 
     # Populating the decision variables
     centers_flat = centers.flatten()
@@ -183,10 +183,10 @@ def _compute_inverse_range(min_value: float, max_value: float) -> float:
 
 
 def _extract_vector(kwargs_dict, prefix, size, macro_idx):
-    """Pull a flat vector of length `size` from kwargs named `{prefix} {macro_idx} {i}`."""
+    """Pull a flat vector of length `size` from kwargs named `{prefix}_{macro_idx}_{i}`."""
     vector = np.empty(size, dtype=float)
     for i in range(size):
-        vector[i] = kwargs_dict.pop(f"{prefix} {macro_idx} {i}")
+        vector[i] = kwargs_dict.pop(f"{prefix}_{macro_idx}_{i}")
     return vector
 
 

justice/util/pareto_nash_run.py

@@ -61,13 +61,13 @@ def build_policy_bank_from_5row_csv(
         row = df.iloc[a]
         for i in range(n_agents):
             centers = np.array(
-                [row[f"center {i} {j}"] for j in range(centers_len)], dtype=np.float64
+                [row[f"center_{i}_{j}"] for j in range(centers_len)], dtype=np.float64
             )
             radii = np.array(
-                [row[f"radii {i} {j}"] for j in range(radii_len)], dtype=np.float64
+                [row[f"radii_{i}_{j}"] for j in range(radii_len)], dtype=np.float64
             )
             weights = np.array(
-                [row[f"weights {i} {j}"] for j in range(weights_len)], dtype=np.float64
+                [row[f"weights_{i}_{j}"] for j in range(weights_len)], dtype=np.float64
             )
             policy_bank[i, a, :] = np.concatenate([centers, radii, weights])
 

pyproject.toml

@@ -5,12 +5,12 @@ description = "JUSTICE is an open-source Integrated Assessment Modeling Framewor
 authors = [{ name = "PBiswas", email = "p.biswas@tudelft.nl" }]
 license = { text = "BSD-3-Clause" }
 readme = "README.md"
-requires-python = ">=3.10, <4"
+requires-python = ">=3.12, <4"
 dependencies = [
     "fair (==2.1.3)",
     "h5py (==3.13.0)",
     "platypus-opt (==1.4.1)",
-    "ema-workbench (==2.5.3)",
+    "ema-workbench (>=3.0.0)",
     "scipy (>=1.10.0)",
     "deap (>=1.4.0)",
     "ipyparallel (>=8.6.1)",
@@ -19,16 +19,21 @@ dependencies = [
     "seaborn (>=0.13.0)",
     "tyro (>=0.9.24)",
     "wandb (==0.20.1)",
-    "numpy (==1.24.3)",
+    "numpy (>=1.26.0, <2.0)",
+    "pandas (>=2.1.0)",
+    "python-cdd (==0.0.98)",
+    "pytest (>=8.0)",
+]
+
+[project.optional-dependencies]
+rl = [
     "torch (==2.4.1)",
     "SuperSuit (==3.9.3)",
     "pettingzoo (==1.24.3)",
-    "python-cdd (==0.0.98)",
     "momaland (==0.1.1)",
     "gymnasium (==0.28.1)",
     "morl-baselines (==1.0.0)",
     "tensorboard (==2.19.0)",
-    "pytest (>=8.0)",
 ]
 
 [build-system]

run_optimization.py

@@ -19,7 +19,6 @@
 import numpy as np
 from ema_workbench import (
     Model,
-    Policy,
     RealParameter,
     ScalarOutcome,
     CategoricalParameter,
@@ -28,14 +27,9 @@
     SequentialEvaluator,
     MPIEvaluator,
     Constant,
-    Scenario,
-    perform_experiments,
-)
-from ema_workbench.em_framework.optimization import (
-    ArchiveLogger,
-    EpsilonProgress,
-    EpsNSGAII,
+    Sample,
 )
+from platypus import EpsNSGAII
 
 from justice.util.EMA_model_wrapper import (
     model_wrapper_emodps,
@@ -298,12 +292,12 @@ def run_optimization_adaptive(
     centers_shape = n_rbfs * n_inputs
     weights_shape = len(data_loader.REGION_LIST) * n_rbfs
 
-    centers = [RealParameter(f"center {i}", -1.0, 1.0) for i in range(centers_shape)]
+    centers = [RealParameter(f"center_{i}", -1.0, 1.0) for i in range(centers_shape)]
     radii = [
-        RealParameter(f"radii {i}", SMALL_NUMBER, 1.0) for i in range(centers_shape)
+        RealParameter(f"radii_{i}", SMALL_NUMBER, 1.0) for i in range(centers_shape)
     ]
     weights = [
-        RealParameter(f"weights {i}", SMALL_NUMBER, 1.0) for i in range(weights_shape)
+        RealParameter(f"weights_{i}", SMALL_NUMBER, 1.0) for i in range(weights_shape)
     ]
     model.levers = centers + radii + weights
 
@@ -316,7 +310,7 @@ def run_optimization_adaptive(
         ),
     ]
 
-    reference_scenario = Scenario("reference", ssp_rcp_scenario=reference_index)
+    reference_scenario = Sample("reference", ssp_rcp_scenario=reference_index)
 
     filename = f"{social_welfare_function.value[1]}_{nfe}_{seed}.tar.gz"
     timestamp = datetime.datetime.now().strftime("%Y_%m")  # _%d_%H_%M_%S
@@ -332,18 +326,6 @@ def run_optimization_adaptive(
     os.makedirs(directory_name, exist_ok=True)
 
     rank = _mpi_rank()
-    if rank == 0:
-        convergence = [
-            ArchiveLogger(
-                directory_name,
-                [lever.name for lever in model.levers],
-                [outcome.name for outcome in model.outcomes],
-                base_filename=filename,
-            ),
-            EpsilonProgress(),
-        ]
-    else:
-        convergence = []
 
     if optimizer == Optimizer.EpsNSGAII:
         algorithm_class = EpsNSGAII
@@ -368,7 +350,8 @@ def run_optimization_adaptive(
                 nfe=nfe,
                 epsilons=epsilons,
                 reference=reference_scenario,
-                convergence=convergence,
+                filename=filename,
+                directory=directory_name,
                 population_size=population_size,
                 algorithm=algorithm_class,
             )
@@ -379,7 +362,8 @@ def run_optimization_adaptive(
                 nfe=nfe,
                 epsilons=epsilons,
                 reference=reference_scenario,
-                convergence=convergence,
+                filename=filename,
+                directory=directory_name,
                 population_size=population_size,
                 algorithm=algorithm_class,
             )
@@ -390,7 +374,8 @@ def run_optimization_adaptive(
                 nfe=nfe,
                 epsilons=epsilons,
                 reference=reference_scenario,
-                convergence=convergence,
+                filename=filename,
+                directory=directory_name,
                 population_size=population_size,
                 algorithm=algorithm_class,
             )
@@ -524,15 +509,15 @@ def run_optimization_momadps(
     levers = []
     for macro_idx in range(n_macro_regions):
         levers.extend(
-            RealParameter(f"center {macro_idx} {i}", -1.0, 1.0)
+            RealParameter(f"center_{macro_idx}_{i}", -1.0, 1.0)
             for i in range(centers_len)
         )
         levers.extend(
-            RealParameter(f"radii {macro_idx} {i}", SMALL_NUMBER, 1.0)
+            RealParameter(f"radii_{macro_idx}_{i}", SMALL_NUMBER, 1.0)
             for i in range(radii_len)
         )
         levers.extend(
-            RealParameter(f"weights {macro_idx} {i}", SMALL_NUMBER, 1.0)
+            RealParameter(f"weights_{macro_idx}_{i}", SMALL_NUMBER, 1.0)
             for i in range(weights_len)
         )
     model.levers = levers
@@ -570,7 +555,7 @@ def run_optimization_momadps(
         ),
     ]
 
-    reference_scenario = Scenario(
+    reference_scenario = Sample(
         "reference", ssp_rcp_scenario=reference_ssp_rcp_scenario_index
     )
 
@@ -589,15 +574,6 @@ def run_optimization_momadps(
     rank = _mpi_rank()
     lever_names = [lever.name for lever in model.levers]
     outcome_names = [outcome.name for outcome in model.outcomes]
-    if rank == 0:
-        convergence = [
-            ArchiveLogger(
-                directory_name, lever_names, outcome_names, base_filename=filename
-            ),
-            EpsilonProgress(),
-        ]
-    else:
-        convergence = []
 
     optimizer_map = {
         Optimizer.EpsNSGAII: EpsNSGAII,
@@ -628,7 +604,8 @@ def run_optimization_momadps(
             nfe=nfe,
             epsilons=epsilons,
             reference=reference_scenario,
-            convergence=convergence,
+            filename=filename,
+            directory=directory_name,
             population_size=population_size,
             algorithm=algorithm_class,
         )
@@ -768,10 +745,10 @@ def run_single_agent_momadps(
     for macro_idx in range(n_macro_regions):
         action_row = policy_bank_df.iloc[actions[macro_idx]]
         for i in range(centers_len):
-            fixed_centers[macro_idx, i] = action_row[f"center {macro_idx} {i}"]
-            fixed_radii[macro_idx, i] = action_row[f"radii {macro_idx} {i}"]
+            fixed_centers[macro_idx, i] = action_row[f"center_{macro_idx}_{i}"]
+            fixed_radii[macro_idx, i] = action_row[f"radii_{macro_idx}_{i}"]
         for i in range(weights_len):
-            fixed_weights[macro_idx, i] = action_row[f"weights {macro_idx} {i}"]
+            fixed_weights[macro_idx, i] = action_row[f"weights_{macro_idx}_{i}"]
 
     model.constants = [
         Constant("n_regions", n_regions),
@@ -814,14 +791,14 @@ def run_single_agent_momadps(
 
     levers = []
     for i in range(centers_len):
-        levers.append(RealParameter(f"center {variable_macro_index} {i}", -1.0, 1.0))
+        levers.append(RealParameter(f"center_{variable_macro_index}_{i}", -1.0, 1.0))
     for i in range(radii_len):
         levers.append(
-            RealParameter(f"radii {variable_macro_index} {i}", SMALL_NUMBER, 1.0)
+            RealParameter(f"radii_{variable_macro_index}_{i}", SMALL_NUMBER, 1.0)
         )
     for i in range(weights_len):
         levers.append(
-            RealParameter(f"weights {variable_macro_index} {i}", SMALL_NUMBER, 1.0)
+            RealParameter(f"weights_{variable_macro_index}_{i}", SMALL_NUMBER, 1.0)
         )
     model.levers = levers
 
@@ -834,7 +811,7 @@ def run_single_agent_momadps(
         ScalarOutcome("fraction_above_threshold", kind=ScalarOutcome.MINIMIZE),
     ]
 
-    reference_scenario = Scenario(
+    reference_scenario = Sample(
         "reference", ssp_rcp_scenario=reference_ssp_rcp_scenario_index
     )
 
@@ -852,17 +829,9 @@ def run_single_agent_momadps(
     os.makedirs(directory_name, exist_ok=True)
 
     rank = _mpi_rank()
+
     lever_names = [lever.name for lever in model.levers]
     outcome_names = [outcome.name for outcome in model.outcomes]
-    if rank == 0:
-        convergence = [
-            ArchiveLogger(
-                directory_name, lever_names, outcome_names, base_filename=filename
-            ),
-            EpsilonProgress(),
-        ]
-    else:
-        convergence = []
 
     set_ema_context(
         model=model,
@@ -884,7 +853,8 @@ def run_single_agent_momadps(
             nfe=nfe,
             epsilons=epsilons,
             reference=reference_scenario,
-            convergence=convergence,
+            filename=filename,
+            directory=directory_name,
             population_size=population_size,
             algorithm=algorithm_class,
         )
@@ -918,7 +888,7 @@ def build_random_policy(model, seed=1234):
     for lever in model.levers:
         low, high = lever.lower_bound, lever.upper_bound
         policy_data[lever.name] = rng.uniform(low, high)
-    return Policy("random_policy", **policy_data)
+    return Sample("random_policy", **policy_data)
 
 
 def setup_model_from_config(config_path, mapping_base_path="data/input"):
@@ -1015,15 +985,15 @@ def setup_model_from_config(config_path, mapping_base_path="data/input"):
     levers = []
     for macro_idx in range(n_macro_regions):
         levers.extend(
-            RealParameter(f"center {macro_idx} {i}", -1.0, 1.0)
+            RealParameter(f"center_{macro_idx}_{i}", -1.0, 1.0)
             for i in range(centers_len)
         )
         levers.extend(
-            RealParameter(f"radii {macro_idx} {i}", SMALL_NUMBER, 1.0)
+            RealParameter(f"radii_{macro_idx}_{i}", SMALL_NUMBER, 1.0)
             for i in range(radii_len)
         )
         levers.extend(
-            RealParameter(f"weights {macro_idx} {i}", SMALL_NUMBER, 1.0)
+            RealParameter(f"weights_{macro_idx}_{i}", SMALL_NUMBER, 1.0)
             for i in range(weights_len)
         )
     model.levers = levers