update quantile scripts

Author: SvenKlaassen

scripts/irm/cvar.py

@@ -0,0 +1,13 @@
+from montecover.irm import CVARCoverageSimulation
+
+# Create and run simulation with config file
+sim = CVARCoverageSimulation(
+    config_file="scripts/irm/cvar_config.yml",
+    log_level="INFO",
+    log_file="logs/irm/cvar_sim.log",
+)
+sim.run_simulation()
+sim.save_results(output_path="results/irm/", file_prefix="cvar")
+
+# Save config file for reproducibility
+sim.save_config("results/irm/cvar_config.yml")

scripts/irm/cvar_config.yml

@@ -0,0 +1,63 @@
+# Simulation parameters for CVAR Coverage
+
+simulation_parameters:
+  repetitions: 200
+  max_runtime: 19800 # 5.5 hours in seconds
+  random_seed: 42
+  n_jobs: -2
+
+dgp_parameters:
+  n_obs: [5000] # Sample size
+  dim_x: [5] # Number of covariates
+
+# Define reusable learner configurations
+learner_definitions:
+  linear: &linear
+    name: "Linear"
+
+  logit: &logit
+    name: "Logistic"
+
+  lgbmr: &lgbmr
+    name: "LGBM Regr."
+    params:
+      n_estimators: 200           # Fewer trees — faster
+      learning_rate: 0.05         # Balanced speed and stability
+      num_leaves: 15              # Modest complexity for smaller data
+      max_depth: 5                # Limit tree depth to avoid overfitting
+      min_child_samples: 10       # Minimum samples per leaf — conservative
+      subsample: 0.9              # Slightly randomized rows
+      colsample_bytree: 0.9       # Slightly randomized features
+      reg_alpha: 0.0              # No L1 regularization (faster)
+      reg_lambda: 0.1             # Light L2 regularization
+      random_state: 42            # Reproducible
+
+  lgbmc: &lgbmc
+    name: "LGBM Clas."
+    params:
+      n_estimators: 200           # Fewer trees — faster
+      learning_rate: 0.05         # Balanced speed and stability
+      num_leaves: 15              # Modest complexity for smaller data
+      max_depth: 5                # Limit tree depth to avoid overfitting
+      min_child_samples: 10       # Minimum samples per leaf — conservative
+      subsample: 0.9              # Slightly randomized rows
+      colsample_bytree: 0.9       # Slightly randomized features
+      reg_alpha: 0.0              # No L1 regularization (faster)
+      reg_lambda: 0.1             # Light L2 regularization
+      random_state: 42            # Reproducible
+
+dml_parameters:
+  tau_vec: [[0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8]] # Quantiles
+  trimming_threshold: [0.01]
+  learners:
+    - ml_g: *linear
+      ml_m: *logit
+    - ml_g: *lgbmr
+      ml_m: *lgbmc
+    - ml_g: *lgbmr
+      ml_m: *logit
+    - ml_g: *linear
+      ml_m: *lgbmc
+
+confidence_parameters:
+  level: [0.95, 0.90] # Confidence levels

scripts/irm/lpq.py

@@ -0,0 +1,13 @@
+from montecover.irm import LPQCoverageSimulation
+
+# Create and run simulation with config file
+sim = LPQCoverageSimulation(
+    config_file="scripts/irm/lpq_config.yml",
+    log_level="INFO",
+    log_file="logs/irm/lpq_sim.log",
+)
+sim.run_simulation()
+sim.save_results(output_path="results/irm/", file_prefix="lpq")
+
+# Save config file for reproducibility
+sim.save_config("results/irm/lpq_config.yml")

scripts/irm/lpq_config.yml

@@ -0,0 +1,46 @@
+# Simulation parameters for LPQ Coverage
+
+simulation_parameters:
+  repetitions: 200
+  max_runtime: 19800 # 5.5 hours in seconds
+  random_seed: 42
+  n_jobs: -2
+
+dgp_parameters:
+  n_obs: [5000] # Sample size
+  dim_x: [5] # Number of covariates
+
+# Define reusable learner configurations
+learner_definitions:
+  logit: &logit
+    name: "Logistic"
+
+  lgbmc: &lgbmc
+    name: "LGBM Clas."
+    params:
+      n_estimators: 200           # Fewer trees — faster
+      learning_rate: 0.05         # Balanced speed and stability
+      num_leaves: 15              # Modest complexity for smaller data
+      max_depth: 5                # Limit tree depth to avoid overfitting
+      min_child_samples: 10       # Minimum samples per leaf — conservative
+      subsample: 0.9              # Slightly randomized rows
+      colsample_bytree: 0.9       # Slightly randomized features
+      reg_alpha: 0.0              # No L1 regularization (faster)
+      reg_lambda: 0.1             # Light L2 regularization
+      random_state: 42            # Reproducible
+
+dml_parameters:
+  tau_vec: [[0.3, 0.4, 0.5, 0.6, 0.7]] # Quantiles
+  trimming_threshold: [0.01]
+  learners:
+    - ml_g: *logit
+      ml_m: *logit
+    - ml_g: *lgbmc
+      ml_m: *lgbmc
+    - ml_g: *lgbmc
+      ml_m: *logit
+    - ml_g: *logit
+      ml_m: *lgbmc
+
+confidence_parameters:
+  level: [0.95, 0.90] # Confidence levels

scripts/irm/pq.py

@@ -0,0 +1,13 @@
+from montecover.irm import PQCoverageSimulation
+
+# Create and run simulation with config file
+sim = PQCoverageSimulation(
+    config_file="scripts/irm/pq_config.yml",
+    log_level="INFO",
+    log_file="logs/irm/pq_sim.log",
+)
+sim.run_simulation()
+sim.save_results(output_path="results/irm/", file_prefix="pq")
+
+# Save config file for reproducibility
+sim.save_config("results/irm/pq_config.yml")

scripts/irm/pq_config.yml

@@ -0,0 +1,46 @@
+# Simulation parameters for PQ Coverage
+
+simulation_parameters:
+  repetitions: 200
+  max_runtime: 19800 # 5.5 hours in seconds
+  random_seed: 42
+  n_jobs: -2
+
+dgp_parameters:
+  n_obs: [5000] # Sample size
+  dim_x: [5] # Number of covariates
+
+# Define reusable learner configurations
+learner_definitions:
+  logit: &logit
+    name: "Logistic"
+
+  lgbmc: &lgbmc
+    name: "LGBM Clas."
+    params:
+      n_estimators: 200           # Fewer trees — faster
+      learning_rate: 0.05         # Balanced speed and stability
+      num_leaves: 15              # Modest complexity for smaller data
+      max_depth: 5                # Limit tree depth to avoid overfitting
+      min_child_samples: 10       # Minimum samples per leaf — conservative
+      subsample: 0.9              # Slightly randomized rows
+      colsample_bytree: 0.9       # Slightly randomized features
+      reg_alpha: 0.0              # No L1 regularization (faster)
+      reg_lambda: 0.1             # Light L2 regularization
+      random_state: 42            # Reproducible
+
+dml_parameters:
+  tau_vec: [[0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8]] # Quantiles
+  trimming_threshold: [0.01]
+  learners:
+    - ml_g: *logit
+      ml_m: *logit
+    - ml_g: *lgbmc
+      ml_m: *lgbmc
+    - ml_g: *lgbmc
+      ml_m: *logit
+    - ml_g: *logit
+      ml_m: *lgbmc
+
+confidence_parameters:
+  level: [0.95, 0.90] # Confidence levels