14 Clicks to Build a Beginner AI Project
Our Lunch Break Webinars are getting cooler every week. Recently, we turned viewers into participants during an Artificial Intelligence lunch with Peltarion. We’d hate to see you missing out, so here’s a tutorial on leveraging AI to tell genuine disaster reports from personal outbursts on Twitter.
AI Project Made Simple
Our host for the AI webinar was Anna Gross. Although certainly knowledgeable, she is a Business Development Lead at Peltarion, not a dedicated Artificial Intelligence engineer. It would be cliche to say “if Anna can execute this project, so can you”—Anna may have gone down this road a few times by now—but you get the idea. It wouldn’t take a technical background or prior knowledge to stick with us.
As this is an introductory project, we offer a few shortcuts to make life easier. You won’t have to collect the data set (i.e. a fancy term for tweets) or set up a Machine Learning environment yourself. This would take unproportionally vaster expertise in Data Science (and more clicks!).
Setting Up AI Project (2 Steps)
First, let’s get the data. We will be using a dataset of 10,000 tweets published by the machine learning community Kaggle. Just follow this link , sign up, and download train.csv. Do not worry about accepting the rules for Kaggle’s related competition: you won’t have to submit anything there.
Now, time to follow-up on the free environment promise. Create a Pel tarion account and log into the platform to proceed.
Exporting Data for AI Project (7 clicks)
1. Hit the big yellow button
2. Hit a green button (preferably the one with solid background)
3-5. Hit the prominent green button with a solid outline, select train.csv, and confirm the choice
6-7. Note that Peltarion has already split the data into “Training” and “Validation” sets. You can play around with the proportion but such clicks will be on you. The essential click here is changing the sequence length for text to 150 so that the model grabs the entire tweet. Once this is settled, follow the green ’n’ solid pattern to confirm the change.
Training Data for AI Project (7 clicks)
1. Save version button will now be replaced by Use in new experiment. This is the easiest click of the tutorial.
2. Actually, the follolwing click presents a decent case for being the easiest, too. You don’t need to change anything on this screen before choosing Next.
3-5. Change input to text (tweets are indeed texts) and switch target to the target column. This defines the binary behavior of our algorithm: it predicts tweets to be genuine disaster reports (1) or rubbish (0). We are happy with default values for the rest of the tab, so the solid green area is your cursor’s next destination.
6. Return-of-the-Jedi-level plot twist time: You click Create, not Next
7. Just like with the Star Wars franchise, I believe 7 was enough. Hit Run and wait a couple of hours to see the results. No spoilers from us in the article; you can check out entries for Kaggle’s competition to see what others got.
In the meantime, check out the recording of the webinar as Anna Gross talks AI and provides more context to the project:
AI Projects For Real
Working with data and utilizing it via Machine Learning (and its subset, Artificial Intelligence), requires Python knowledge. You can learn this language from our entry-level Python course.