AutoMLData ScienceMachine Learning

Introducing ObviouslyAI — No-Code Machine Learning Solution

This article was originally published on Towards Data Science on August 13th, 2020.

A couple of folks from contacted me a few days back to introduce their service — a completely no-code machine learning automation tool. I was a bit skeptical at first, as I always am with supposedly fully-automated solutions, but I decided to give it a try. I’ll share my thoughts in this article, and discuss if service is worth the try.

I find it somewhat difficult to watch tools like this one automate machine learning, and decrease the need for machine learning engineers in small and medium-sized companies. The reasons are many, but the biggest is that the purpose of machine learning was to automate other professions, but we’ve managed to automate machine learning with machine learning. Good job.

It’s not entirely a bad thing, as we can now focus on more important things, instead of fitting algorithm after algorithm, with the aim of squeezing that additional 0.15% accuracy.

So, what is Obviously AI?

Hero section on their home page explains it pretty well:

The total process of building ML algorithms, explaining results, and predicting outcomes in one single click.

After playing around with it for a bit, I must say that it delivers. So, what’s the catch? Good question. We are accustomed to quite pricy solutions in this day and age, and ObviouslyAI is not an exception. It has a more than decent free plan, limited to CSV files only with no more than 50,000 rows. That’s more than enough for basic exploration.

I’m on the Free plan currently, and it’s more than enough for my needs. We’ll now go through a concrete example of training a machine learning model with this service, and you’ll see how stupidly easy the entire thing is.

Before we start, I want to make a quick disclaimer. Even though folks at ObviouslyAI asked me to review their service, I am in no way affiliated with them, nor will I try to convince you to switch to a paid account. Everything I say is based purely on the Free version.

Registration and setup

It was at this step that the first strange thing happened. I’ve gone ahead and opened the Signup page, and was prompted to enter an email. What was strange is that my personal Gmail account wasn’t eligible for registration.

A business email account is a must.

I have a business email account from my company, so that wasn’t an issue, but might be a dealbreaker for some of you. I can’t verify if the same thing happens for other email providers, but Gmail doesn’t work at this point in time. Strange.

Nevertheless, I’ve completed the registration process and verified the email address, and then I was presented with a nice-looking dashboard:

Image for post

There are some sample datasets built-in, but I guess those work flawlessly. We won’t be using those for our machine learning tasks, and will instead be using a well-known Wine dataset. Let’s build a model in the next section.

Building a model

This step is stupidly simple, as stated earlier. The first to do is to upload the dataset. We’ll use the Add Dataset button on the sidebar to do so:

Image for post

Once clicked, a modal should appear on which we can drag and drop (or click to upload) our dataset. Keep in mind these constraints (free version):

  • File size must be less than 25MB
  • There must be at least 1000 rows
  • There must be at least 5 columns

Our wine dataset passes all of those conditions, so we can upload it:

Image for post

Once the upload is finished, we’ll get to this well-presented exploration modal:

Image for post

From here, we just need to follow the instructions. Let’s click on the Use for Prediction button. We’re almost finished with the preparation. In the next modal window we just need to choose the target variable:

Image for post

And that’s it! The service complains that we should reduce the number of unique values in the target variables, but we can ignore that. To finish, just click on the Start Predicting button. That’s all you have to do.

The model is trained. Done. It’s that easy.

That doesn’t mean that model is any good, so we’ll explore how it performed in the next section.

Model evaluation

Once the model is trained, we’re presented with the report dashboard. It consists of a few areas:

  • Drivers
  • Personas
  • Export Predictions
  • Advanced Analytics
  • Tech Specs

We’ll explore a couple of those here, the first being the Drivers area.

Drivers area

Put simply, this area tells us which variables are most important for forecasting, ergo which variables have the greatest prediction power. In our case, variables densityalcohol, and free_sulfur_dioxide are the top 3:

Image for post

Nicely formatted and easy to understand. Let’s proceed.

Export Predictions area

There’s no point in machine learning without making predictions on new, previously unseen data. That’s where the free version falls short, unfortunately. We can make predictions by uploading a CSV of previously unseen data — only attributes without the target variable.

Image for post

That’s all the free version offers. It might be enough for you, but I was expecting to see more.

What paid version gets you is deployed version of your model in the form of a REST API, which makes predictions that much easier to make from any programming language:

Image for post

This option isn’t supported in the free version, unfortunately, but can you blame them?

Tech Specs area

This area displays some basic information about the model, such as which algorithm was used, what was the accuracy on the train, test, and validation subsets, etc:

Image for post

It’s a nice section to get a basic understanding of your model, but that’s it.

And that’s pretty much it for this introductory article to ObviouslyAI. Let’s wrap things up in the next section.

Before you go

In a nutshell, ObviouslyAI is obviously awesome, and such an easy service to recommend. For small to medium-sized businesses I can even see it as the only data science solution, maintained by one or more software developers that are training models with a couple of clicks and making predictions with API calls.

Data science teams could deliver a better solution, sure, but that team would potentially cost tens of thousands USD per month, where this solution is somewhere below $200 per month for the most expensive option. You do the math.

It was obvious right from the start that data science will become just another flavor of software engineering, but it is services like this one that change the minds of even the most stubborn individuals.

What are your thoughts? Have you tried ObviouslyAI? Feel free to drop your thoughts in the comment section.

Join my private email list for more helpful insights.

Dario Radečić
Data scientist, blogger, and enthusiast. Passionate about deep learning, computer vision, and data-driven decision making.

You may also like

1 Comment

  1. Hi Dario,

    Thanks for the review. As a fellow Data Scientist, these automation tools amaze me. Machine Learning has become more and more available for people who are not ML experts. I actually learned first ML techniques prior to scikit-learn era. Imagine that!

    I think these tools go in the right direction and do not threaten as Data Scientist. We should still have jobs working on bigger projects and working with people who would like to have better models than the automatic ones. However, this look like a great solution for smaller companies that one create quickly their first model prototypes.

    Thanks for an objective review, I may play with the free version to see how good results I can get on a different data set.

Comments are closed.

More in AutoML