model_conversion_rpp: Bayesian model that combines conversion estimation and value...
In eliobartos/bayeselio: Conjugate Bayesians Models

Description Usage Arguments Details Value Author(s)

View source: R/model_conversion_payer.R

Model uses total number of cases and successful cases to estimate conversion based on bernoulli-beta model and for successful cases estimates continuous variable using exponential-gamma model

model_conversion_rpp(
  alpha,
  beta,
  success,
  total_sample_size,
  shape,
  rate,
  sum_sample,
  n_post = 1e+05
)

`alpha`	Parameter for prior in bernoulli-beta model representing number of success - 1
`beta`	Parameter for prior in bernoulli-beta model representing number of fails - 1
`success`	Actual number of successful cases (payers) in your data.
`total_sample_size`	Total number of cases in your data.
`shape`	Parameter for prior in exponential-gamma model representing the number of samples from before - 1
`rate`	Parameter for prior in exponential-gamma model representing the sum of samples from before
`sum_sample`	Sum of variable (money) for each successful case.
`n_post`	Size of sample from posterior distribution

Final result is estimated value for each case (not just converted cases). Motivation is when you have some percent of users that spend money and others that spend no money. First we estimate conversion to payers rate, and for payers estimate revenue per payer (rpp). Final posterior distribution is multiple of these two posterior distributions and is representing revenue per user (rpu).

Check model_bernoulli_beta and model_exponential_gamma for details regarding each model.

List of 3 posterior distributions: post_conv representing conversion rate (success rate) from bernoulli-beta model, post_rpp representing value of variable (money) per successful case from exponential-gamma model, post_rpu representing value of variable per each case (money per all cases), post_rpu = post_conv * post_rpp

Elio Bartoš

eliobartos/bayeselio documentation built on Feb. 5, 2021, 2:16 p.m.