Configuration Reference

The config.yml file is used to configure the ConversionFlow model pipeline. This reference document provides detailed explanations of each parameter and its options.

General Configuration

Parameter	Type	Description	Default
project_name	str	A descriptive name for identification and organization purposes	None

Preprocessing Configuration (preprocess)

Scaling (preprocess.scaling)

Parameter	Type	Description	Default
enabled	bool	Flag to enable or disable feature scaling	true
method	str	Scaling method to be used if scaling is enabled	standard
exclude_from_scaling	list	Feature names to exclude from the scaling process	[]

Available scaling methods:

• standard: Standardization (z-score normalization)

• minmax: Min-Max scaling

• robust: Robust scaling using quartiles

Missing Values (preprocess.missing_values)

Parameter	Type	Description	Default
strategy	str	Strategy for handling missing values	fill
fill_value	any	Value to use for filling missing values	0

Available strategies:

• fill: Fill missing values with the specified fill_value

• drop: Drop rows containing missing values

Model Configuration (model)

Parameter	Type	Description	Default
name	str	Name for the model, used for identification	None
description	str	Brief description of the model	None
nodes	dict	Nodes (variables) in the Bayesian Network	None
edges	list	Directed edges (relationships) in the Bayesian Network	None

Example node definition:

nodes:
  stage1:
    - car_configuration
    - finance_calculator
  stage2:
    - brochure_request
    - contact_request
  final:
    - test_drive

Example edge definition:

edges:
  - [car_configuration, brochure_request]
  - [car_configuration, contact_request]
  - [finance_calculator, brochure_request]
  - [brochure_request, test_drive]
  - [contact_request, test_drive]

Priors Configuration (priors)

Parameter	Type	Description	Default
sigma_node	float	Standard deviation for the prior distribution of node intercepts	5.0
sigma_beta_global	float	Global standard deviation for prior distributions of regression coefficients	5.0
beta_distributions	dict	Prior distribution for each directed edge	None

Example beta distribution definition:

beta_distributions:
  car_configuration_brochure_request:
    dist: HalfCauchy
    parameters:
      loc: 0
      scale: 5

Inference Configuration (inference)

Parameter	Type	Description	Default
draws	int	Number of MCMC samples to draw after tuning	1000
tune	int	Number of MCMC samples to discard as warm-up	500
chains	int	Number of independent MCMC chains to run	4
random_seed	int	Seed for the random number generator	42

Output Configuration (output)

Parameter	Type	Description	Default
directory	str	Directory where all output files will be saved	output/
figures.extension	str	File extension for output figures	png
figures.dpi	int	DPI for output figures	300

Logging Configuration (logging)

Parameter	Type	Description	Default
level	str	Logging level, determining the verbosity of logs	INFO
file	str	Path to the log file where messages will be written	logs/pipeline.log

Complete Example

project_name: "Toyota Conversion Analysis"

preprocess:
  scaling:
    enabled: true
    method: standard
    exclude_from_scaling: [date, user_id]
  missing_values:
    strategy: fill
    fill_value: 0

model:
  name: "toyota_conversion_model"
  description: "Bayesian Network model for Toyota conversion flow"
  nodes:
    stage1:
      - car_configuration
      - finance_calculator
    stage2:
      - brochure_request
      - contact_request
    final:
      - test_drive
  edges:
    - [car_configuration, brochure_request]
    - [car_configuration, contact_request]
    - [finance_calculator, brochure_request]
    - [brochure_request, test_drive]
    - [contact_request, test_drive]

priors:
  sigma_node: 5.0
  sigma_beta_global: 5.0
  beta_distributions:
    car_configuration_brochure_request:
      dist: HalfCauchy
      parameters:
        loc: 0
        scale: 5

inference:
  draws: 1000
  tune: 500
  chains: 4
  random_seed: 42

output:
  directory: "output/"
  figures:
    extension: "png"
    dpi: 300

logging:
  level: "INFO"
  file: "logs/pipeline.log"