How to use ChatGPT for Time Series in R
Written by Matt Dancho
Writing R code is a slow process especially when you are first learning Time Series Analysis. What if you could speed it up? You can and this is how. In this free R-tip, I share a real case study where I made the working R code for my time series analysis and saved over 3 hours of work using ChatGPT.
Table of Contents
Today I share how to automate 80% of your Time Series Analysis code with ChatGPT
. Here’s what you’re learning today:
- ChatGPT Prompts: The mistakes you’ll make and how to get ChatGPT to write your time series analysis code correctly.
- 2-Minute Case Study: How I made a time series analysis with
ChatGPT
.
- Bonus: Sneak Peek At My NEW Shiny App that extends this ChatGPT tutorial 👇
Bonus: A shiny app that extends this tutorial
SPECIAL ANNOUNCEMENT: AI for Data Scientists Workshop on December 18th
Inside the workshop I’ll share how I built a SQL-Writing Business Intelligence Agent with Generative AI:
What: GenAI for Data Scientists
When: Wednesday December 18th, 2pm EST
How It Will Help You: Whether you are new to data science or are an expert, Generative AI is changing the game. There’s a ton of hype. But how can Generative AI actually help you become a better data scientist and help you stand out in your career? I’ll show you inside my free Generative AI for Data Scientists workshop.
Price: Does Free sound good?
How To Join: 👉 Register Here
R-Tips Weekly
This article is part of R-Tips Weekly, a weekly video tutorial that shows you step-by-step how to do common R coding tasks. Pretty cool, right?
Here are the links to get set up. 👇
My Previous ChatGPT R Tutorials
So listen, time series is a tough problem to solve. And there’s a learning curve with ChatGPT. To help I’ve done 2 R-Tips on ChatGPT. These are good to start with if you’ve never used ChatGPT before (or want starter tutorials that are great for beginners):
-
I showed you how ChatGPT I made a full R Shiny App in under 10-minutes. That was insane!
-
I demonstrated how important ChatGPT prompts are giving you a full ChatGPT prompt guide and walkthrough
But now I want to take it a step up and demonstrate using ChatGPT for a more difficult problem…
Using ChatGPT for Time Series in R
I’m going to be fully transparent here. ChatGPT is NOT 100% with Time Series yet. It got me 80% of the way there. BUT the last mile, I actually had to know time series.
I also tried Bard. Bard was even worse than ChatGPT’s GPT-4 model.
To help, I want to give you some pointers.
ChatGPT Prompt Guide for Time Series
The mistake I made when I first started using ChatGPT was not being specific enough in directing chatgpt
what R code I want it to write for me. I’ve since learned from this.
Here’s a good starter chatgpt prompt for my time series problem:
Looks pretty straightforward, but it’s actually quite complex. Let’s break it down.
The key is specifying:
- Which R Libraries to Use: I specified to forecast using the
modeltime
library. Full Disclosure: I’m the creator. But, it IS awesome for time series.
- What Time Series Goal: For my time series project it’s to forecast. Other time series problems might involve exploratory plotting, anomaly detection, or decomposition.
- What Inputs to the Time Series Problem: For this time series project, I want to forecast initially the next 30 days (the forecast horizon) and it’s a daily time series data set with 2 columns: date and value.
And ChatGPT generates the code for me:
Steal my code (it's legal).
Does the code work?
A big problem that data scientists are facing is that chatgpt
code isn’t working out of the box.
So let’s test the code. I ran it, and it turns out…
ChatGPT has Hallucinations
It didn’t run. What happened is so common people have termed it “ChatGPT Hallucinations”.
Yep, chatgpt makes stuff up. And this is why it’s important to know how to code. Case in point: modeltime.arima
is not an R package. I have no clue where it got this.
The reality: You need to know how to code and do time series
Fortunately, I’ve been doing this a while. And I just had to debug the code. So here’s what my process looks like:
- ChatGPT wrote the code (15 Seconds)
- I debugged the code (10 minutes)
The net time was a little over 10 minutes for me to get this code working.
Get My Working Code Here (It's in the 064_chatgpt_time_series folder).
And the modified ChatGPT code now produces a test forecast using an ARIMA model. Now, let’s set aside that the forecast isn’t actually that good (I’ll improve this with a different model). But I have something usable.
Total Time Savings: 30 minutes.
I don’t have the completed solution yet
I just have a test forecast at this point. I had to go back and request chatpgt to refit the ARIMA model on the full dataset and forecast the next 30 days.
What it gave me again had some hallucinations that I had to fix. No big deal. This took maybe 5 minutes.
Steal my code (it's legal).
And, the modified chatgpt code now produces a 30-day forecast into the future, which is pretty cool.
So at this point I’ve spent 15 minutes making this forecast. Something that would have taken about an hour to write 100 lines of code.
Total Time Savings: 45 minutes.
But here’s where the time saving stacks up.
BONUS: I asked ChatGPT to make a Shiny App to Forecast with Modeltime
I asked ChatGPT to make a Shiny App using my corrected code. Making a shiny app is a task that would have taken me 3+ hours.
I spent 15 minutes correcting the code. And I was able to produce this shiny app that uses a better XGBoost Model to forecast this time series.
Total Time Savings: 3 hours 30 minutes
Want my Bonus Shiny Time Series App?
Want all the code I just showed you PLUS my bonus Shiny Time Series App? It’s my free gift to you: You can have my shiny app library.
Steal my shiny app from this tutorial.
The code for the bonus shiny time series app is inside of R-Tip 064_chatgpt_time_series
. Enjoy.
Oh, I forgot to mention something. There’s a…
Big Problem: You still need to learn how to do Time Series
If you learn one thing from this tutorial, it better be that ChatGPT isn’t the full solution. It’s a productivity enhancement. As you just saw, I completed 4 hours of work in under 30 minutes.
But, what’s going to happen when you try this on your own?
Don’t know? Allow me to answer: If you can’t do data science, time series, OR code… Well, then you aren’t going to get very far.
Would you like help? Here’s how I can help you become a data scientist and land a 6-figure job you love and learn how to use ChatGPT better (all for free).
SPECIAL ANNOUNCEMENT: AI for Data Scientists Workshop on December 18th
Inside the workshop I’ll share how I built a SQL-Writing Business Intelligence Agent with Generative AI:
What: GenAI for Data Scientists
When: Wednesday December 18th, 2pm EST
How It Will Help You: Whether you are new to data science or are an expert, Generative AI is changing the game. There’s a ton of hype. But how can Generative AI actually help you become a better data scientist and help you stand out in your career? I’ll show you inside my free Generative AI for Data Scientists workshop.
Price: Does Free sound good?
How To Join: 👉 Register Here