pivot_forecast/readme.md

worked on so far

setup

the basic assumption is a single sales table is available to work with that has a lot of related data that came from master data tables originally. the goal then is to break that back apart to whatever degree is necessary.

• run schema.sql and perd.sql to setup basic tables
• create a table fc.live as copied from target (will need to have columns version and iter added if not existing)
• run target_info.sql to populate the fc.target_meta table that holds all the columns and their roles
• fill in flags on table fc.target_meta to show how the data is related
• run build_master_tables.sql to generate foreign key based master data

routes

• all routes would be tied to an underlying sql that builds the incremental rows
• that piece of sql will have to be build based on the particular sales layout
  • columns: a function to build the columns for each route
  • where a function to build the where clause will be required for each route
  • the result of above will get piped into a master function that build the final sql
  • the master function will need to be called to build the sql statements into files of the project

route baseline

• forecast = baseline (copied verbatim from actuals and increment the dates) + diffs. if orders are canceled this will show up as differ to baseline
• regular updates to baseline may be required to keep up with canceled/altered orders
• copy some period of actual sales and increment all the dates to serve as a baseline forecast

TO-DO:

• join to period tables to populate season; requires variance number oof table joins, based on howmany date functions there are 🙄
• some of the app parameters can be consolidated, the baseline period could be one large range potentially, instead of 2 stacked periods
• setup something to fill in sql parameters to do testing on the function
• update node to handle forecast name parameter
• clean up SQL generation to prevent injection
• calc status is hard-coded right now in the json request -> probably needs to be manuall supplied up front

running problem list

• baseline route
  • problem: how will the incremented order season get updated, adding an interval won't work
    • a table fc.odate, has been built, but it is incomplete, a setup function filling in these date-keyed tables could be setup
    • if a table is date-keyed, fc.perd could be targeted to fill in the gaps by mapping the associated column names
  • problem: the target sales data has to map have concepts like order_date, and the application needs to know which col is order date
    • add column called application hook
  • there is not currently any initial grouping to limit excess data from all the document# scenarios