Upload trained machine learning model to Thingspeak

Hello,

I trained a machine learning model on MatLab Desktop and the App Classification Learner. Now I want to upload the trained model to the ThingSpeak Cloud, so that I can make predictions with new data in the cloud. How can I do this ?

Thank you in advance!

Answer 1

[Hans Scharler]

There is a "Developing an IoT Analytics System with MATLAB, Machine Learning, and ThingSpeak" paper linked from the ThingSpeak website that may help you.

https://www.mathworks.com/company/newsletters/articles/developing-an-iot-analytics-system-with-matlab-machine-learning-and-thingspeak.html

In this example, Robert takes the model and generates MATLAB code and includes it as a function.

[Vinod]

The article demonstrates the preferred approach to train the model on a desktop and embed the trained model parameters in the MATLAB code that is operationalized on ThingSpeak.

Another, less preferred, approach is to upload the trained model to DropBox and have the code operationalized on ThingSpeak pull the model down parameters from DropBox on demand and apply it to the ingested data.

[Marcel Zehner]

Okay, thanks.

But is there a possibility to upload the trained model like it is exported from the Classification Learner App? So that I can only use the expression in the cloud:

yfit = trainedModel.predictFcn(testdata)

This is how the Classification Learner App suggest the prediction after the export of the model.

When I export the model it is a 1x1 struct with the model parameters.

[Vinod]

Depending on how complex it is to recreate the struct you can do that in your script, or, you can export the MAT file to DropBox and pull it in at runtime using the functions I linked above.

[Marcel Zehner]

Thank you. I am trying the version with the Dropbox. In the code to download the .mat file is a downloadpath. Do you know which download path I have to use so that the ThingSpeak know the variables? Right now the download is successful, but the variables did not get recognized.

Thank you in advance.

[Abhishek Pokhrel]

@Marcel , did you managed to find a solution? I am looking for the same solution as yours. I have trained a classification model and i want to export that model to ThingSpeak so that it can predict based on the real time data.

@Vinod, I would like to know if there's some tutorial related to this.

[Marcel Zehner]

Hey, yes I managed to find a solution. I used the solution with DropBox. I uploaded the trained model (the .mat file) to DropBox. After that I wrote MatLAB Code that ran in ThingSpeak. This code downloads the model at runtime using pull the model down parameters from DropBox from @Vinod. Next you can use the trained model in your code to make a prediction with your real time data in ThingSpeak. I am getting the new data from a ThingSpeak field. For this I used the function

thingSpeakRead

Then you can apply the read data to your model.

Regarding to the problem with the downloadpath I wrote before. You can write :

downloadPath = pwd;

[Sirapadsorn Chaisaen]

@Marcel Zehner I try to use function https://www.mathworks.com/matlabcentral/fileexchange/67833-download-files-from-your-dropbox-api-folder-using-matlab but I got some errors. What variable to be specified in the function? for example, dropboxAccessToken,fileNames. I don't know how to use it.

[Sirapadsorn Chaisaen]

@Vinod In pull the model down parameters from DropBox what the fileNames variable look like? Is a path or just the name of file (i.e. net.mat) ? if it file' path ,what code that I supposed to write?

I write:

fileNames = ["Apps/<IoT with ML in everyday life>/fddfd/net206010.mat"];

but it doesn't work. Can you help me ?

Sorry if my english is hard to understand.

[Vinod]

@Sirapadsorn Chaisaen: Just use the filename, you do not need the full path.

fileNames = 'net206010.mat';

[Sirapadsorn Chaisaen]

@Vinod Thank you so much!