sample_config.yaml

################################
# Lightweight MMM options
################################

# Library to use: at present only lightweight_mmm is supported
library: lightweight_mmm

# "model name" is the lightweight_mmm option for how to compute adstock
# (i.e. how much effect an ad has over time, as someone's memory fades)
# Values: carryover, adstock, hill_adstock
model_name: carryover

# These lightweight_mmm options determine how many samples to take from the
# Markov chain Monte Carlo process; higher values mean slower runtime.
number_warmup: 1000
number_samples: 1000
number_chains: 2

# These lightweight_mmm options specify how to handle seasonality: the expected
# number of degrees of freedom, and whether to separately model intra-week variation.
degrees_seasonality: 2
weekday_seasonality: null

# Fixed seed for MCMC process: set this if subsequent runs need to
# have exactly the same results. If not set, this defaults to a
# seed generated with get_time_seed() in lightweight_mmm.
# seed: 1

# Custom priors: use this to guide the model based on what you already know
# about the range of possible values for a parameter (typically after running
# many MMMs and/or many experiments).
# This example sets a prior for the baseline trend coefficient to a normal
# distribution with mean 0 and stddev 1 (same as lightweight_mmm's default).
# custom_priors:
#   coef_trend:
#     type: normal
#     loc: 0
#     scale: 1

################################
# Data handling options
################################

# Experimental feature.  Leave as false for now.
log_scale_target: false

# Frequency of the input data (required)
# Values: daily, weekly
raw_data_granularity: daily

# Specify the expected total number of rows in the CSV input (optional),
# and the range of dates to use from the input data (required)
data_rows:
  total: 117
  start_date: 2023-01-01
  end_date: 2023-04-27

# Proportion of the dataset to use for training (in-sample), with the remainder
# used for an out-of-sample test. Defaults to 90% train, 10% test.
train_test_ratio: 0.9

################################
# Column definitions
################################

# Column names from the input to be ignored (optional).
# All columns in the input data should be listed somewhere in this file,
# so use this field when testing out which columns to include.
# ignore_cols:
#   - "Unwanted Column"

# Column name for the date index of each row (optional,
# case-sensitive; defaults to lowercase "date").
# Values in this column must be ISO 8601 format (YYYY-MM-DD).
# date_col: date

# Column name for the target or output metric (e.g. number of leads, sales volume)
# that we want to increase with marketing spend (required).
target_col: kpi

# Column names for "extra features", i.e. factors external to
# marketing that influence the target metric (optional).
extra_features_cols:
  - extra_feature_1

# Required: each block under "media" represents an advertising channel,
# defined by a display name (used in charts), a column name for
# impressions data, and a column name for cost data.
media:
  - display_name: Ad Platform 1
    impressions_col: media_1_impressions
    spend_col: media_1_cost

  - display_name: Ad Platform 2
    impressions_col: media_2_impressions
    spend_col: media_2_cost

  - display_name: Ad Platform 3
    impressions_col: media_3_impressions
    spend_col: media_3_cost