Overview
- Salesforce has open sourced TransmogrifAI – their own automated machine learning library that powers the Einstein AI platform
- It’s an impressive solution to building enterprise-level machine learning systems in a scalable and production-ready manner
- The library is available on GitHub for you to source and start working with, today!
Introduction
Salesforce is one of the leaders in the analytics/data science space. Automated machine learning (Auto ML) is easily the hottest tool in that industry. When you combine them both, you get a game-changing experience that has the potential to disrupt the way businesses approach machine learning projects!
Einstein AI is Salesforce’s flagship ML platform that is powered by their own TransmogrifAI library. It is an automated machine learning library, written in Scala and running on top of Spark, built for dealing with structured data. It is used by Salesforce to build machine learning solutions for production-ready scenarios, in a scalable and time efficient manner. If you’re wondering how it’s pronounced, Salesforce says it’s trans-mog-ri-phi.
In a blog post by Shubha Nabar, Senior Director, Data Science, she shared the general workflow of TransmogrifAI, it’s advantages, how YOU can use it for your individual use case (or your business), and also elaborated on the way TransmogrifAI has been designed. It really is an impressive solution to building enterprise-level ML systems, a process way more difficult than what other blog posts would have you believe!
The below image illustrates the workflow of TransmogrifAI:
Their GitHub page (link below) has links to documentations and Wikis to help you get started with the library. It even includes a small code example run on the popular Titanic dataset. In just a few lines of code, the model gave an accuracy of 85%, an impressive start. And don’t worry, you don’t lose interpretability at any point – TransmogrifAI lets you manually specify which features you want, the algorithm you prefer to run, etc.
To install it on your machine, follow the below steps, as mentioned on TransmogrifAI home page:
- Install Java 1.8
- Get Spark 2.2.x
- Set an environment variable
export SPARK_HOME=<SPARK_FOLDER> - Add the TransmogrifAI libs to your Gradle or sbt project
Excited yet? You can get started with TransmogrifAI NOW by heading over to the library’s GitHub page and downloading the code.
Our take on this
A game changer in ML? Quite possibly. With Auto-Keras being launched last week, and now this, open-source Auto ML is having its day in the sun right now. Unlike Auto-Keras (which is incredibly useful for deep learning by the way), TransmogrifAI has a bunch of examples, proper documentations and a detailed wiki of terms (what else did you expect from an industry leader like Salesforce?).
It’s inspiring to see market leaders take the initiative and enable the entire community with powerful libraries. I will definitely be using this for my next hackathon, and encourage you to do the same!
Subscribe to AVBytes here to get regular data science, machine learning and AI updates in your inbox!
Senior Editor at Analytics Vidhya.Data visualization practitioner who loves reading and delving deeper into the data science and machine learning arts. Always looking for new ways to improve processes using ML and AI.
