The no code space is growing, and we are thrilled to be at the forefront of it when it comes to data science. However, with so many products using “no-code” in their marketing, one would wonder how no-code truly manifests when it comes to a user’s journey.
We recently came across an article talking about Google’s Cloud AutoML and how it can use no-code to run predictions. Clearly, this was an interesting and useful tool for those looking to get started with ML. However, the tool came with long technical documentation and mentions such as "neural architecture” which felt far from being “no-code.” We wanted to take the opportunity and dissect how “no-code” is truly approached from a product perspective than just as a marketing buzzword.
As the in-house non-technical person, I had to enlist the help of our ML engineers to dissect Google’s ML capabilities. With words like “data fusion” and “neural architecture,” it was a lot for a mere Creative Director to follow. That’s when we came up for the idea of this post.
We believe no-code machine learning should aim for democratization by oversimplifying the process.
Say you’re a Product Designer. You want to build the most powerful product you can in the simplest way. You want to cut out extra steps, complications, or words and images into the interface for a seamless experience. We take the same approach to no-code ML. We take out anything unnecessary to give the user direct insight while providing the most accurate model possible.
True no-code products:
- Democratize tech in their respective space
- Allow effortless tasks to be done by all—regardless of their technical background
- Help SMBs, freelancers, and entrepreneurs jump start their business
- Provide agility to users without breaking the bank
We’ve seen it with WordPress, Squarespace, and Webflow where businesses can create a complex website without knowing code. This revolutionized how websites are made and truly democratized site building for all.
In this post, we will go through building a machine learning model with Google ML and add our approach to the process.
Adding Data to Build the Algorithm and Creating a Data Pipeline
- Upload to cloud storage bucket
- Create instance on data fusion
- Follow a long documentation
- With BigQuery, create a new database
Obviously AI’s Process:
- Drag and drop CSV file
- No long documentation
- if BigQuery, just click 1 button.
The first step to our machine learning flow starts with the data. While we can connect to your favorite databases, we believe simplifying it to dragging and dropping a CSV (comma separated values) file into our platform and hitting “Go.” This is effortless because CSV files are very common and don’t require a third-party tool to get started. You can store your data locally on your laptop of G-Drive and upload to our no-code platform.
With Google, you have to work with four different products and read technical documentation to enable an API. We believe, enabling an API shouldn't be this complex. That's why, on Obviously AI you don't need to know complex documentations, you don't need XYZ. All you do is just click a the API button.
To build a data pipeline with Google you have to create a new BigQuery table and transform it (add filters, etc.)
We believe the data pipeline should be automatically built where you have the option to go to advanced view and drag and drop filters.
Creating a Machine Learning Model
- Define multiple parameters (e.g. id column, features, target column)
- Define model parameters, e.g. loss function, algorithm to use
- Results built in 30 mins
Obviously AI’s Process
- Define only one parameter (e.g. column you want to predict.)
- Result in < 1 min
No-code is built on being agile. If we can cut down the time from uploading data to insight, why wouldn’t we? With Google, you have to train the model before predicting. While it is vague how long this will take. Once the model is trained, the post says it will take 262.26 seconds to make a prediction.
We believe this should take less than a minute.
Taking Action from the Model and Understanding the Tech
We believe that just getting to the model isn't enough for businesses to get value. It is about taking action on that model and enabling anyone to quickly and easily understand, interpret and explain that model which truly builds business value. Most tools end their journey after the model is built.
At Obviously AI, your journey really just starts when the model is built. Here are some details:
Actions You Can Take After Making A Prediction With Google:
- Export data to other tools to take action
Actions You Can Take After Making A Prediction With Obviously AI:
- Explore interactive feature importance graph
- Deep dive into top drivers of prediction result
- Create customer personas and predict their actions
- View decision trees, understand how the ML algorithm really makes decisions
- Export via API
- Export to Google Sheets to share with team/add to company files
- Export to Zapier (coming soon)
Obviously AI also provides tech specs to understand the ML model better inside the report.
If you wish to learn more about Google’s ML process just read this article on neural architecture search (not sure what that is).
There’s a Difference Between No-Code ML and Technical ML
We believe true no-code machine learning should be as non-technical as possible while still allowing the user to understand the process behind the data prediction.
While our approach to machine learning is to democratize machine learning for everyone, we don’t take the term “no-code” lightly. We want to be leaders in the no-code machine learning space and to do this we need to simplify the process, make ML effortless, and offer agility to those who need to jump start their data science initiatives.
Read some case studies on how you can apply ML to business.