The logic of the flow is like this.
The raw dataset contains a time series of money transactions between Originated Parties ('nameOrig") and the Destinated Parties ("nameDesc"). "Step" is the time step from 1 onward. Each step represents one hour. "Type" is the transaction type. In the workflow, I only want to look at "CASH_OUT", that's why I used a filter in the workflow to extract all "CASH_OUT" transactions. "Amount" is the transaction amount.
The aim of the workflow including the Macro is to identify a sliding window of 100 hours so that we can do a rolling sum of the transaction amount, and rolling sum of the transaction counts for each nameDesc (Destination Party) afterwards. Please note that the workflow provided are only for identifying, for each nameDesc, the immediate previous time step of sliding window starting point up to the current row. To give you an example (see the attached png)
for nameDesc "C970", it has 15 transactions from step 9 to 236. "Sum_figure" is the number of transaction happened during the same hour (or same time step). Note that transaction did not happen each hour. Now, we want to find out the rolling sum of transactions for a sliding window of 100 hours. So, for the transactions of C970 done at timestep 131, its sliding window would start from 131-100 = 31, i.e. any transaction happened during the time frame from 31 to 131 (including both 31 and 131 will be counted). Therefore, the "sliding window edge" is defined to be the previous time step that C970 has a transaction, which is timestep 21. The Batch Macro is tasked to sort out the "sliding window edge" for each transaction of each nameDesc.
I hope this explain my logic. Thanks.
