generate_transaction_datasets.qmd

---
title: "Generate Synthetic Transaction Datasets"
author: "Mick Cooney <mickcooney@gmail.com>"
editor: source
execute:
  message: false
  warning: false
  error: false
format:
  html:
    light: superhero
    dark: darkly
    anchor-sections: true
    embed-resources: true
    number-sections: true
    smooth-scroll: true
    toc: true
    toc-depth: 3
    toc-location: left
    code-fold: true
    code-summary: "Show code"
---

```{r knit_opts}
#| include: false


library(conflicted)
library(tidyverse)
library(scales)
library(magrittr)
library(rlang)
library(purrr)
library(glue)
library(stringi)
library(tidyquant)


source("lib_utils.R")
source("lib_btyd.R")


conflict_lst <- resolve_conflicts(
  c("xml2", "magrittr", "rlang", "dplyr", "readr", "purrr", "ggplot2")
  )


options(
  width = 80L,
  warn  = 1,
  mc.cores = parallelly::availableCores()
  )

set.seed(42)

```


In this workbook we use some initial inputs to generate synthetic data to help
us explore various customer CLV  models.

In each instance we will create 50,000 customers, as that should be more than
enough for any application.

Should we need a smaller customer count, we can take a subsample from this
dataset.

```{r set_n_customers}
#| echo: true

n_customers <- 5000

first_tnx_date <- as.Date("2020-01-01")
final_tnx_date <- as.Date("2023-01-01")
```


# Generate Short Timeframe 50K Synthetic Cohort Data

We now repeat this exercise, but rather than using the new customer data from
the transaction dataset, we also create a synthetic set of 'new customers' and
generate a transaction dataset based on that.

We use a short time frame from the data, spanning two years, and construct all
customers and transactions within that timeframe.

```{r generate_shortframe_synthetic_cohort}
#| echo: true

synthdata_shortframe_cohort_tbl <- generate_customer_cohort_data(
    n_customers = n_customers,
    first_date  = first_tnx_date,
    last_date   = final_tnx_date,
    id_prefix   = "SFC"
    )

synthdata_shortframe_cohort_tbl |> glimpse()
```

Now that we have generated our cohort data, we move on to generating our
transaction data based on the PNBD model.


```{r calculate_shortframe_synth_customer_data}
#| echo: true

pnbd_params_lst <- list(
  mu_mn       =   0.10,
  mu_cv       =   1.00,

  lambda_mn   =   0.25,
  lambda_cv   =   1.00,

  amt_hiermn  = 100.00,
  amt_hiercv  =   1.00,
  amt_custcv  =   1.00
  )

synthdata_shortframe_simparams_tbl <- synthdata_shortframe_cohort_tbl |>
  generate_pnbd_customer_simulation_params(
    params_lst     = pnbd_params_lst
    )

synthdata_shortframe_transactions_tbl <- synthdata_shortframe_simparams_tbl |>
  generate_pnbd_customer_transaction_data(final_tnx_date = final_tnx_date) |>
  generate_transaction_metadata()

synthdata_shortframe_transactions_tbl |> glimpse()
```


## Write Data to Disk

We now write this data to disk.

```{r write_synth_shortframe_data_disk}
#| echo: true

synthdata_shortframe_cohort_tbl       |> write_rds("data/synthdata_shortframe_cohort_tbl.rds")
synthdata_shortframe_simparams_tbl    |> write_rds("data/synthdata_shortframe_simparams_tbl.rds")
synthdata_shortframe_transactions_tbl |> write_rds("data/synthdata_shortframe_transactions_tbl.rds")
```



# Generate Long Time-frame 50K Synthetic Cohort Data

We now repeat the data synthesis but for a much longer period of time, so
that the censoring effects of the observation time are less apparent in the
dataset.

```{r generate_longframe_synth_cohort}
#| echo: true

first_tnx_date <- as.Date("2010-01-01")
final_tnx_date <- as.Date("2023-01-01")

synthdata_longframe_cohort_tbl <- generate_customer_cohort_data(
    n_customers = n_customers,
    first_date  = first_tnx_date,
    last_date   = final_tnx_date,
    id_prefix   = "LFC"
    )

synthdata_longframe_cohort_tbl |> glimpse()
```



```{r calculate_longframe_synth_customer_data}
#| echo: true

pnbd_params_lst <- list(
  mu_mn       =   0.10,
  mu_cv       =   1.00,

  lambda_mn   =   0.25,
  lambda_cv   =   1.00,

  amt_hiermn  = 100.00,
  amt_hiercv  =   1.00,
  amt_custcv  =   1.00
  )


synthdata_longframe_simparams_tbl <- synthdata_longframe_cohort_tbl |>
  generate_pnbd_customer_simulation_params(
    params_lst     = pnbd_params_lst
    )

synthdata_longframe_transactions_tbl <- synthdata_longframe_simparams_tbl |>
  generate_pnbd_customer_transaction_data(final_tnx_date = final_tnx_date) |>
  generate_transaction_metadata()

synthdata_longframe_transactions_tbl |> glimpse()
```


## Write Data to Disk

We now write this data to disk.

```{r write_longframe_synth_data_disk}
#| echo: true

synthdata_longframe_cohort_tbl       |> write_rds("data/synthdata_longframe_cohort_tbl.rds")
synthdata_longframe_simparams_tbl    |> write_rds("data/synthdata_longframe_simparams_tbl.rds")
synthdata_longframe_transactions_tbl |> write_rds("data/synthdata_longframe_transactions_tbl.rds")
```





# R Environment {.unnumbered}

```{r show_session_info, echo=TRUE, message=TRUE}
options(width = 120L)
sessioninfo::session_info()
options(width = 80L)
```