RocketML: Moseley’s fast experimentation tool for Data Scientists
Linus Pauling, the two time Nobel prize winner once said, “if you want to have good ideas, you must have many ideas”. What is the trick to having many ideas? There are many tricks. That is for another blog post. In this post let’s understand what it means for Data Scientists.
Before I jump in the “Moseley’s Fast Experimentation Tool” part, let me just say, it’s been a while since I’ve posted. As with many of you, this has been a busy time for us trying to adjust to new normal. Several sentiments sunk in deeper than before.
First, we are here to build. Widely distributed article by Marc Andreessen titled “It is time to Build ‘’, asks the world to focus on building. I couldn’t agree more with the benefits of the mental shift that a slogan like that creates. Building anything requires forethought and commitment. It reduces emotions. In a world overloaded with emotion, it is a helpful shift today. This is not to minimize the importance of fighting for social justice. It is just another way to help us all grow. As a startup company building is our second nature. It is our survival instinct. It is the ONLY advantage we have over established companies. We focus on building something that matters, building something that is remarkable. That is truly the only sustaining mission of RocketML. We double down on this mission during these troubled times.
We are in the purest sense the “ultimate builders”. We build tools that help build machine learning models. Data Scientists are the users of the the tool we build. It is a great pleasure and responsibility to build our platform in a way that makes is delightfully easy for our users to in turn build great things. To do this, we borrowed our inspiration from two of the greatest Scientists: Henry Moseley (Physicist) and Linus Pauling (Chemist).
Henry Moseley is the creator of the “modern periodic table” we all used in our middle schools. He died at a young age of 27, but not before making a massive contribution by definitively establishing a systematic mathematical relationship between the wavelengths of the X-rays produced and the atomic numbers of the metals that were used as the targets in X-ray tubes. This has become known as Moseley’s law. What is less known about Henry Moseley is that he spent most of his time building a “contraption” for running his experiments rapidly. This contraption later on became X-ray Spectrography. It took him more than a year to build this tool (Moseley’s X-ray spectrometer) that enabled rapid iteration, which in turn helped complete the puzzle of what is now called “periodic table”. Once Moseley built his contraption, it took him merely weeks to solve the periodic table puzzle. How is this related to what we do at RocketML?
We build ‘moseley’s x-ray spectrography for data scientists’ to help them solve their own ‘periodic table’ puzzle.
Moseley had two principles. The first was that when one starts to set up an experiment one must not stop for anything until it is set up. The second was that when one starts the experiment itself one must not stop till it is finished.
RocketML was built to enable these principles for Data Scientists.
It is a tool for Data Scientists to experiment fast without having to spend time in setting up and waiting for ML training to complete for each run of their experiments. Following Moseley’s principles is very hard for current day Data Scientists. Most data scientists struggle with poorly built tools. Tinkering with poorly built tools and related unproductive tasks take a toll on them and the end result is — they fail in following Moseley’s principles.
RocketML on the other hand let’s data scientists focus on what is important to them, ie., discover models that work well when exposed to the real world. It comes with over 200+ latest and most powerful machine learning, data science tools ready to be used on large amounts of compute for rapid iteration if data scientists choose to do so.
DataScientists task may seem different than discovering the structure of THE “periodic table”. However, most Data Scientist teams across enterprises are tasked to find multiple models. It is a data scientist’s job to find these critical models and stack them up like a Periodic table!
Data Scientist’s success also comes from experimenting more than they normally do.
“If you want to build great ML models; You must build many”
This is an adaptation of Linus Pauling’s principle.
RocketML, comes with 200+ libraries for rapid iterations on either GPU or Cluster of CPUs to increase the speed of experimentation on large datasets in the hopes of discovering models. It is our mission to help Data Scientists adhere to the principles laid down by great scientists before us like Henry Moseley and Linus Pauling.
We are about ready to release RocketML for broader segments. If you are interested to try out RocketML for your team or company, please fill out a request. We will be in touch very quickly!
Originally published at https://www.linkedin.com.
