Parameters

Complete reference for CatBoost search spaces, lag features, and Optuna optimization in parameters.yml

The system’s behavior is controlled by a centralized configuration file located at conf/base/parameters.yml. This ensures that experiment parameters — including CatBoost hyperparameters and Optuna search spaces — are versioned alongside the code.

​

Feature Engineering

Configuration for lag generation used by the create_feature_interactions node.

parameters.yml

feature_engineering:
  demand_lags: [1, 2, 22, 23]   # Target (cnt) lag values
  feature_lags:
    hour: [1, 2, 3]              # Hour of day lags
    day_of_week: [1, 2, 3]       # Day of week lags
    weather_condition: [1, 2, 3]  # Weather condition lags
    temp: [1, 2, 3]              # Temperature lags
    humidity: [1, 2, 3]          # Humidity lags

​

Optimization Search Spaces

Defines the boundaries for the Optuna Bayesian optimization of CatBoost.

parameters.yml

optuna:
  n_trials: 150
  primary_model: catboost
  catboost_search_space:
    learning_rate:
      min: 0.01
      max: 0.3
    depth:
      min: 4
      max: 12
    l2_leaf_reg:
      min: 1.0
      max: 10.0
    n_estimators:
      min: 100
      max: 500
    min_data_in_leaf:
      min: 5
      max: 50

​

Advanced Features

The DAG supports optional multi-objective and ensemble training, controlled by flags in parameters.yml.

parameters.yml

advanced_features:
  enable_multi_objective: true
  enable_ensemble: true
  multi_objective_trials: 50
  objectives:
    - rmse
    - mae
  preference_weights:
    rmse: 0.6
    mae: 0.4
  ensemble_type: stacking

Set enable_multi_objective or enable_ensemble to false to skip advanced training tasks and reduce pipeline runtime.

​

Monitoring & Automation

Parameters used by the automated_retraining_dag.py to decide when retraining is triggered.

parameters.yml

monitoring:
  drift_threshold: 0.1

automation:
  time_interval_days: 7
  performance_threshold: 5.0
  data_volume_threshold: 10000

​

General Settings

Parameter	Value	Description
train_fraction	0.8	80/20 chronological train/test split.
data_processing.random_state	42	Random seed for reproducibility.
optuna.n_trials	150	Number of optimization trials per run.
optuna.primary_model	catboost	Primary model algorithm.
dataset_name	preprocessed_bike_data	Name of preprocessed dataset.
enable_data_docs	false	Toggle data documentation generation.

---

​

Related Pages

Training & Optimization

How these parameters drive the CatBoost + Optuna training pipeline.

Feature Engineering

How lag and feature interaction parameters are applied to raw data.