One of the barriers to making greater use of machine learning (ML) is that it takes a high level of expertise to design ML models and implement them through code. Data scientists and data engineers with the necessary skills are in short supply and high demand.
But those barriers are being broken down with the launch of no-code ML tools like Google Cloud AutoML. According to Google, Cloud AutoML enables teams with limited ML expertise to create and train high-quality models through an easy-to-use graphical interface — without needing to write a single line of code. Cloud AutoML does this by using machine learning itself to figure out the right ML model for the job, trained on your specific training data.
But will a no-code model really work as well — in terms of both speed of execution and the accuracy of predictions — as a “traditional” handcrafted one?
Putting no-code ML to the test
We wouldn’t recommend a tool to our customers without putting it to the test ourselves — and a project for one of our customers provided the perfect opportunity.
As a leading provider of Governance, Risk and Compliance (GCR) services, this company wanted to be able to identify drug information leaflets in a corpus of free-text documents written in more than 10 languages. They asked us to run a workshop to help their in-house team understand how advanced natural language processing solutions on Google Cloud might be applied to this problem.
The in-house team had already built a Native Bayes model using Python, and wanted us to evaluate their model against the options in Google Cloud. The Ancoris team created two ML solutions, along with their associated data ingestion, training and prediction pipelines.
The first solution took a traditional approach, with one of our data scientists developing and training a Support Vector Machine (SVM) model using the SKLearn Python library.
The second solution was created in Cloud AutoML, using its graphical user interface to prepare and train the model.
What did we learn?
- The in-house hand coded model took around 3 days to build and around 70 seconds to train, and delivered predictions to an accuracy of around 95%
- The Ancoris hand coded model took around a day to build and around 75 seconds to train, and was slightly more accurate at just under 99%.
- The Cloud AutoML model took just 10 minutes to set up and 6 hours to train, and delivered an accuracy of 97.5%.
As you can see, it was a close call between all three models. But it’s worth noting that the cost of those 6 hours spent training the model was just $18 — which is rather less than we paid our data scientist for a day’s work on his model.
Reducing the need for data science experts
Our data scientist doesn’t think he’ll be out of a job just yet, but he points out that “even three years ago, some of these problems would have been very difficult to solve and would have involved a lot of technology and expert skill sets. Google’s ML solutions and especially Cloud AutoML are taking all of that away.”
As for the GCR provider who asked us to show us what was possible, they told us, “The workshop provided great insight into our data with all the supporting details — from the machine learning fundamentals to a productionised workflow — allowing us to immediately put the improvements into action.”
Working with our data analytics and AI team
Our Data, Analytics and AI practice brings together a highly committed team of experienced data scientists, mathematicians and engineers. We pride ourselves in collaborating with and empowering client teams to deliver leading-edge data analytics and machine learning solutions on the Google Cloud Platform.
We operate at the edge of modern data warehousing, machine learning and AI, regularly participating in Google Cloud alpha programs to trial new products and features and to future-proof our client solutions.
We have support from an in-house, award winning application development practice to deliver embedded analytics incorporating beautifully designed UIs. We are leaders in geospatial data and one of the first companies globally to achieve the Google Cloud Location-based Services specialisation.
If you'd like to find out more about how we can help you build your own modern data and analytics platform, why not take a look at some of our customer stories or browse our resources. Needless to say, please get in touch with our team if you'd like more practical support and guidance.