Using ChatGPT to generate MATLAB Visualizations
Latest activity Reply by Hans Scharler
on 28 Aug 2024
It's been over six years since I've written any serious MATLAB code, so I thought it would be fun to see how easily ChatGPT could help me out. While others have probably already used ChatGPT to generate MATLAB code, I didn't find any evidence of it when I searched through the ThingSpeak forum. That inspired me to post an example to get people thinking about it.
This example reads four temperature fields from the same channel and plots them on a single graph.
My ChatGPT prompt:
The prompt is pretty straightforward and essentially walks through all the elements of the chart that I wanted. It's also important to consider any filtering or "data cleansing" that should be done. Since this was my first time doing this, I decided to use an existing plot I was already familiar with as my "target state".
The prompt: "I would like you to generate some MATLAB code to create what is called a MATLAB Visualization. Its purpose will be to generate a chart of 4 fields found in a ThingSpeak channel. The ThingSpeak channel name is "Nest Nanny 2 (NN-02)" and its channel id (needed for the code) is xxxxxxx. The read api key is XXXXXXXXXXXXXXXX. The chart title should be "Nest Nanny 02 (NN-02) Todays Temperature Readings" It should plot data from 4 fields (field 3 - 6) who's names are "Surface Temp", "1/4 Temp", "3/4 Temp", and "Deep Temp". I would like the plot lines for these 4 fields to be different colors (red, orange, cyan, and blue) but on the same chart. I would like Temperature in Fahrenheit to be on the Y axis and Time on the x axis. I would like it to display the last 100 data points. I'd like to use this colororder. colororder(["#FF0000","#FFBB00","#00BBFF", "#0000FF"]) I would like the legend to be to the right of the chart, not contained within the chart. No temperature data should have a value of 50 or less and we can assume it's a sensor issue when this is observed. If a data value is 50 or less, use the value of it's nearest (in the time series) non-zero, non-negative value for that field. The data is stored in the channel as Fahrenheit and does not need to be converted from Celsius. "
Because I did this exercise using an existing plot and had a clear objective in mind, it was relatively quick to iterate four or five times with ChatGPT to refine my prompt. In just a few minutes, the generated plot almost perfectly matched my existing plot. (I xxxx'ed out the channel ID and API key before posting.) For someone who now only occasionally uses MATLAB, this approach saves time and will hopefully allow me to tap into elements of MATLAB that are now beyond my skill level and memory. ;)
ChatGPT generated MATLAB code:
% MATLAB code to create a ThingSpeak Visualization for temperature readings
% Define channel ID and Read API Key
channelID = xxxxxxx;
readAPIKey = 'XXXXXXXXXXXXXXXX';
% Number of data points to retrieve
numPoints = 100;
% Read data from the ThingSpeak channel for fields 3 to 6
[data, time] = thingSpeakRead(channelID, ...
'Fields', [3 4 5 6], ...
'NumPoints', numPoints, ...
'ReadKey', readAPIKey);
% Replace any value 50 or less with the nearest non-zero, non-negative value
for i = 1:size(data, 2)
% Identify indices where the data is 50 or less
badIndices = find(data(:,i) <= 50);
% Identify indices where the data is valid (greater than 50)
validIndices = find(data(:,i) > 50);
% Interpolate to replace bad values
if ~isempty(validIndices)
data(badIndices, i) = interp1(validIndices, data(validIndices, i), badIndices, 'linear', 'extrap');
end
end
% Plot the data
figure;
colororder(["#FF0000","#FFBB00","#00BBFF", "#0000FF"]); % Set the color order
plot(time, data(:,1), 'LineWidth', 2); hold on;
plot(time, data(:,2), 'LineWidth', 2);
plot(time, data(:,3), 'LineWidth', 2);
plot(time, data(:,4), 'LineWidth', 2);
hold off;
% Add labels and title
xlabel('Time');
ylabel('Temperature (°F)');
title('Nest Nanny 02 (NN-02) Todays Temperature Readings');
% Add legend and move it to the right of the chart
legend({'Surface Temp', '1/4 Temp', '3/4 Temp', 'Deep Temp'}, ...
'Location', 'eastoutside');
% Format the time on the x-axis
datetick('x', 'HH:MM:SS', 'keepticks');
grid on;
% Adjust the figure for better readability
set(gca, 'FontSize', 12);
Plot Comparison:
3 Comments
Time Descending
his is terrific work, thanks for adding the suggestion. You can use the MATLAB playground to help you debug the code and get AI suggestions at the same time. Here is a similar bog post.
Just to be silly, i think its great that chat gpt knew to hid your api key in the code above!
Keep it coming, share any more great or interesting code you get from chatgpt.
Wow, thanks for the link Christopher. MATLAB Playground is awesome. I just dropped my prompt into it "as is" and got the correct graph. It also generated better code. For example, both ChatGPT and I used a for loop when cleansing the data of bad values but of course, the Playground used matrix math and saved several lines of code.
Not that the numbers of lines of code matter too much but they were as follows: me - 70, ChatGPT 53, MATLAB Playground - 37.
I'm going to switch over to usung the Playground as the code appears to be better than ChatGPT's and it provides a nice sandbox to iterate quickly in. Props to whichever team created it!
(Also, ChatGPT didn't remove my channel id and API key. I did that before posting)
Thanks for the feedback. We are always working on the system to improve responses.
Sign in to participate
