add did cs multi simulation scripts

Author: SvenKlaassen

scripts/did/did_cs_multi.py

@@ -0,0 +1,13 @@
+from montecover.did import DIDCSMultiCoverageSimulation
+
+# Create and run simulation with config file
+sim = DIDCSMultiCoverageSimulation(
+    config_file="scripts/did/did_cs_multi_config.yml",
+    log_level="DEBUG",
+    log_file="logs/did/did_cs_multi_sim.log",
+)
+sim.run_simulation()
+sim.save_results(output_path="results/did/", file_prefix="did_cs_multi")
+
+# Save config file for reproducibility
+sim.save_config("results/did/did_cs_multi_config.yml")

scripts/did/did_cs_multi_config.yml

@@ -0,0 +1,65 @@
+# Simulation parameters for DID Multi Coverage
+
+simulation_parameters:
+  repetitions: 10
+  max_runtime: 19800 # 5.5 hours in seconds
+  random_seed: 42
+  n_jobs: -2
+
+dgp_parameters:
+  DGP: [1, 4, 6] # Different DGP specifications
+  n_obs: [2000] # Sample size for each simulation (has to be a list)
+  lambda_t: [0.5]
+
+# Define reusable learner configurations
+learner_definitions:
+  linear: &linear
+    name: "Linear"
+
+  logistic: &logistic
+    name: "Logistic"
+
+  lgbmr: &lgbmr
+    name: "LGBM Regr."
+    params:
+      n_estimators: 300            # More trees to learn slowly and steadily
+      learning_rate: 0.03          # Lower learning rate to improve generalization
+      num_leaves: 7                # Fewer leaves — simpler trees
+      max_depth: 3                 # Shallow trees reduce overfitting
+      min_child_samples: 20        # Require more samples per leaf
+      subsample: 0.8               # More row sampling to add randomness
+      colsample_bytree: 0.8        # More feature sampling
+      reg_alpha: 0.1               # Add L1 regularization
+      reg_lambda: 1.0              # Increase L2 regularization
+      random_state: 42             # Reproducible
+
+  lgbmc: &lgbmc
+    name: "LGBM Clas."
+    params:
+      n_estimators: 300            # More trees to learn slowly and steadily
+      learning_rate: 0.03          # Lower learning rate to improve generalization
+      num_leaves: 7                # Fewer leaves — simpler trees
+      max_depth: 3                 # Shallow trees reduce overfitting
+      min_child_samples: 20        # Require more samples per leaf
+      subsample: 0.8               # More row sampling to add randomness
+      colsample_bytree: 0.8        # More feature sampling
+      reg_alpha: 0.1               # Add L1 regularization
+      reg_lambda: 1.0              # Increase L2 regularization
+      random_state: 42             # Reproducible
+
+dml_parameters:
+  # ML methods for ml_g and ml_m
+  learners:
+    - ml_g: *linear
+      ml_m: *logistic
+    - ml_g: *lgbmr
+      ml_m: *lgbmc
+
+  score:
+    - observational # Standard DML score
+    - experimental # Experimental score (no propensity estimation)
+
+  in_sample_normalization: [true, false]
+
+confidence_parameters:
+  level: [0.95, 0.90] # Confidence levels