Use ChatGPT to document your code today

2 Likes

Let’s build off the example in the last post on Use ChatGPT to help understand complex code. Suppose you wrote that complex code. And you’ve worked all day to figure it out and you are… admittedly… proud and spent. You might think “this code took me forever – there’s no way I’m going to forget what I did”. But the experienced hacker in me says:

We forget. And it happens.

The good news? You’re just one ChatGPT conversation away from solid documentation.

(Lawyerly note: that’s assuming, of course, that you can safely share your code with a search engine. Since I work in teaching and learning, my code is far less proprietary than, say, a bank’s.)

So before you close your coding IDE for the day, try asking ChatGPT something like this:

The response from ChatGPT:

/***************************************************************
  Purpose:
    Collapse person-level CPS microdata to one row per State x Year
    and compute weighted labor market rates overall and by subgroups.

  Key columns used:
    - state_fip, state_name   : state identifiers
    - yearquarter             : calendar quarter (we extract calendar year)
    - unemp (0/1)             : unemployed indicator
    - in_LF (0/1)             : in the labor force indicator
    - educ_ltd                : education group text (e.g., "Some College")
    - child_status            : child status text (e.g., "Child < 5")
    - WTFINL                  : person weight (used for weighted rates)

  Notes:
    - (condition=1) returns 1 if true, 0 if false. Multiplying by WTFINL
      gives a weighted count for that condition.
    - sum( ... ) / sum( ... ) creates a weighted proportion (rate).
****************************************************************/

/********************************************************  By State */
proc sql;
  create table hhs.covid_labor_supply as 
  select
    distinct                    /* (Redundant with GROUP BY below, but harmless) */
    state_fip,                  /* State FIPS code */
    state_name,                 /* State full name */
    year(yearquarter) as Year format=9.,   /* Collapse quarters into calendar year */

    /****************************************************************
      Labor Force Status | All Women
      - Unemployment rate: weighted unemployed / weighted labor force
      - LFP rate:         weighted labor force / weighted population
    ****************************************************************/
    sum( (unemp=1) * WTFINL )
      / sum( (in_LF=1) * WTFINL )                 as UE_Women
        label="Unemployment Rate" format=percent9.1,

    sum( (in_LF=1) * WTFINL )
      / sum(              WTFINL )                as LFP_Women
        label="LFP Rate"           format=percent9.1,

    /****************************************************************
      Labor Force Status | By Education
      Same formulas as above, but within each education subgroup.
      (Multiplying by (educ_ltd="...") zeros out everyone else.)
    ****************************************************************/

    /* ------------------------- Unemployment (by education) ------------------------- */
    sum( (educ_ltd="High School Diploma") * (unemp=1) * WTFINL )
      / sum( (educ_ltd="High School Diploma") * (in_LF=1) * WTFINL )  as UE_Women_HS
        label="EDUC <= HS" format=percent9.1,

    sum( (educ_ltd="Some College") * (unemp=1) * WTFINL )
      / sum( (educ_ltd="Some College") * (in_LF=1) * WTFINL )         as UE_Women_SCollege
        label="Some College" format=percent9.1,

    sum( (educ_ltd="College +") * (unemp=1) * WTFINL )
      / sum( (educ_ltd="College +") * (in_LF=1) * WTFINL )            as UE_Women_CollegeP
        label="College +" format=percent9.1,

    /* --------------------------- LFP (by education) ------------------------------- */
    sum( (educ_ltd="High School Diploma") * (in_LF=1) * WTFINL )
      / sum( (educ_ltd="High School Diploma") * WTFINL )              as LFP_Women_HS
        label="EDUC <= HS" format=percent9.1,

    sum( (educ_ltd="Some College") * (in_LF=1) * WTFINL )
      / sum( (educ_ltd="Some College") * WTFINL )                     as LFP_Women_SCollege
        label="Some College" format=percent9.1,

    sum( (educ_ltd="College +") * (in_LF=1) * WTFINL )
      / sum( (educ_ltd="College +") * WTFINL )                        as LFP_Women_CollegeP
        label="College +" format=percent9.1,

    /****************************************************************
      Labor Force Status | By Child Status
      Same idea, but grouped by child_status categories.
    ****************************************************************/

    /* ------------------------- Unemployment (by child status) --------------------- */
    sum( (child_status="No Children")   * (unemp=1) * WTFINL )
      / sum( (child_status="No Children")   * (in_LF=1) * WTFINL )    as UE_Women_NoKids
        label="No Children" format=percent9.1,

    sum( (child_status="Older Children") * (unemp=1) * WTFINL )
      / sum( (child_status="Older Children") * (in_LF=1) * WTFINL )   as UE_Women_OlderKids
        label="Older Children" format=percent9.1,

    sum( (child_status="Child < 5")      * (unemp=1) * WTFINL )
      / sum( (child_status="Child < 5")      * (in_LF=1) * WTFINL )   as UE_Women_YoungKids
        label="Young Children" format=percent9.1,

    /* ----------------------------- LFP (by child status) -------------------------- */
    sum( (child_status="No Children")   * (in_LF=1) * WTFINL )
      / sum( (child_status="No Children")   * WTFINL )                as LFP_Women_NoKids
        label="No Children" format=percent9.1,

    sum( (child_status="Older Children") * (in_LF=1) * WTFINL )
      / sum( (child_status="Older Children") * WTFINL )               as LFP_Women_OlderKids
        label="Older Children" format=percent9.1,

    sum( (child_status="Child < 5")      * (in_LF=1) * WTFINL )
      / sum( (child_status="Child < 5")      * WTFINL )               as LFP_Women_YoungKids
        label="Young Children" format=percent9.1

  from hhs.hhs_otj_raw

  /* GROUP BY 1,2,3 means “group by the 1st, 2nd, and 3rd selected columns”
     i.e., state_fip, state_name, Year. This collapses to one row per State-Year. */
  group by 1,2,3

  /* Sort the output the same way (by state_fip, state_name, Year). */
  order by 1,2,3
  ;
quit;

And a few more helpful hints from ChatGPT:

Wow. Just wow.

A solid prompt to ChatGPT makes my code so much easier to understand — for future me and for my collaborators today.

Even better, it got me thinking about a few things I probably wasn’t considering yet. I love that.

And with that, my brain dump on what I’m loving about using ChatGPT as my coding partner comes to a close.

This list reflects where I’m at right now — and I’m sure it’ll grow as I keep exploring new ways to integrate LLMs into my workflow.

In the meantime, I have just one thing to say to ChatGPT:

I’d also love to hear from you — how are you using LLMs to make your coding life easier?

Bonus points if you share thoughts on how LLMs can support those new to SAS, since I’m especially interested in early-career examples for my work in Academic Programs.

Use ChatGPT to document your code today

The 2025 SAS Hackathon has begun!

SAS AI and Machine Learning Courses