I am looking for a way to create a couple/few step code to do following:   I have two variables, ID and DECISION (screenshot attached), and I have another variable in a different dataset (variable called Var1) that can be empty or any number from 0 to infinite (with decimals), for example first row in the Var1 is 5, and second row is Var1 is 15.75   What do I need is to create a new variable that would assign ID number based on where it falls in a combination of two variables (ID and DECISION) - meaning for value 5, it should return 1 in a newly created variable, and for 15.75, return 2, etc for other variations.   Ideally, i would prefer to break down (ID and DECISION variables first) and maybe put the in rows so i can merge to my main dataset with variable that i want to run against ID and DECISION.   screenshot of variables in my first dataset  (my input variable) attached below.       ... View more

Hi everyone - hope you can help with a quick solution. I looked through existing topics, and found relevant, but when trying to apply it in Miner, still have an issue to produce a decision tree.   here is what i try to do: I have a dataset with binary target variable (default indicator: 1 or 0) and an interval (numeric) variable 'time on books' that goes from 0, 1, 2 ... up to ~300 and represents months on books.   In Miner, I want to build a simple decision tree that have a maximum depth: 1, and maximum branch: 3 - so just splitting 'time on books' into 3 nodes.   the issue is: Once i set up constraints, such as branch, depth, node leaf size, etc..., and once "Run" decision tree i do not have any split for the "Input" variable, but only keeping a "root node" with information on how many are in 0 and 1 groups. I tried "Interactive Binning" and "Transform Variables" to have this input variable either binned or transformed, but not luck.   If anyone had same issue, i must have doing something wrong in the setup, because i can create simple decision trees for other variable i have, but struggling with "time on books" and another character nominal variable.   Any help highly appreciated! ... View more

Hi All, I posted similar couple weeks ago, however these days I need to perform slightly different exercise. I attached 'have' & 'want' in the excel, please have a look. This is just a snapshot of what I have - my data is over 10 years with over a hundred million records     What I have: several years of historical data (month) of accounts with an indicator of a late payment for a respective month and a rate (how many accounts had late payment in a given month). In each month there are unique accounts, such as Aug-10 has 1,270,000 unique accounts, while entire dataset is 33M non-unique accounts. Another point to mention, some account get closed overtime (once it happens, they won't appear on a following month), however new account added. This is why MoM total accounts go up, but not all accounts from previous month would appear on the following month. This is why in table 'want' I would have 1,647,283 unique accounts (combination of all unique accounts that appear at least 1 time during observation period).   What I need is following: I need to reduce data to unique account level, while not loosing insights of monthly late payment rate and overall payment rate.   Until now, I tried two approaches: 1. SurveySelect with STRATA by account and SAMPSIZE=1 2. rand('Uniform') with following sorting and NODUPKEY by account   In both cases, I could achieve reducing accounts to unique account, however when running frequency for each month and total, rates where not close to rates in 'have' - such as, oldest months came back with rate ~6.5% and earliest with ~3.5%, while total rate ~ 5.2.   Can anyone suggest procedure/steps to be taken to handle this through random selection/exclusion to reduce data and keep all insightful information?   Thanks in advance!   ... View more