Configuration Reference

This reference describes key AMMM V2 configuration fields and run-time CLI overrides.

Core YAML Keys

Data and split

Key	Type	Notes
raw_data_granularity	daily | weekly | monthly	controls frequency assumptions
date_col	string	default typically date
target_col	string	required
target_type	revenue | conversion	affects naming of efficiency outputs
train_test_ratio	float	1.0 means train-only split
data_rows	object	optional date/window slicing controls

Features and media

Key	Type	Notes
media	list[object]	each object contains display_name, impressions_col, spend_col
extra_features_cols	list[string]	control regressors
extra_features_impact	dict	supports sign inversion (for example negative)
ignore_cols	list[string]	excluded columns

Prophet-assisted baseline

Key	Type	Notes
prophet.include_holidays	bool	include holiday effects
prophet.holiday_country	string	country holiday calendar
prophet.yearly_seasonality	bool	yearly component
prophet.weekly_seasonality	bool	weekly component
prophet.daily_seasonality	bool	daily component
prophet.trend	bool	target-assisted trend control

Sampling controls

Key	Type	Notes
tune	int	MCMC tuning steps
draws	int	posterior draws
chains	int	chain count
target_accept	float	NUTS target acceptance
ad_stock_max_lag	int	adstock lag depth
seed	int	reproducibility

Diagnostics gating

Key	Type	Allowed values	Behaviour
diagnostics_gating	string	strict | warn | off	strict halts on failed convergence gate

Custom Priors

custom_priors supports overriding defaults for parameter groups such as:

• intercept
• beta_channel
• alpha
• lam (or aliases)
• likelihood
• gamma_control

Each prior block follows:

custom_priors:
  beta_channel:
    dist: HalfNormal
    kwargs:
      sigma: 1.0

CLI Flags (runme.py)

Inputs and outputs

• --data
• --config
• --holidays
• --results-dir

Run mode

• --no-scenarios
• --scenarios
• --multiperiod
• --multiperiod-weeks
• --no-seasonality
• --use-adstock

Performance/sampler

• --fast
• --jax
• --gpu
• --chain-method (vectorized|parallel|sequential)
• --draws
• --tune
• --chains
• --target-accept

Ramp and planning controls

• --ramp-abs
• --ramp-pct
• --ramp-config
• --strict-ramp-config
• --ramp-eps
• --round-increment
• --seasonality-clip-min
• --seasonality-clip-max

Output Contract Touchpoints

Configuration choices affect stage outputs, especially:

• 10_pre_diagnostics/ (pre-checks and prior predictive)
• 50_diagnostics/ (convergence, calibration, Pareto-k)
• 70_optimisation/ (scenario/single-period/multi-period outputs)

Caveat

Passing diagnostics confirms computational and predictive adequacy; it does not confirm causal validity.