This is the sixth in a series of interviews with members of the Machine Commons supplier Collective. Subscribe to the site to be alerted about future posts, or become a partner today!
Javed Inamdar runs Data Science Lab, an analytics consultancy over in Pune, India.
How’s life in the Pandemic?
“Well, we’re located in India.”
He takes the slightest of breaths.
“It’s the 3rd most effected country in the world after the US and Brazil, in one of the highly effected cities and unable to leave the city.”
The so named ‘third world’ has had a really
tough go of the last year. Inequalities have been made more apparent.
Countries that went through the SARS scare a few years ago have fared well, so I’m personally optimistic that COVID-19 will
transpire to help the world prepare for a worse pandemic in the future.
And, business through the pandemic?
“Business wise, there’s been very strict lockdown. Everyone is facing a lot of issues.”
“Products just aren’t on sale, so businesses are just focused on surviving. Analytics isn’t exactly a necessity!”
Tell me what you do?
“By education I’m a mechanical engineer. I joined a company called SAS (number one analytics company in world by market share) and worked in the advanced analytics division. Eventually, headed the advanced analytics division.”
“Last year, I joined a friend who started Data Science Lab.”
“It’s an analytics consulting company. We can handle any analytics challenge, though mostly focused on predictive modelling.”
Any interesting projects you’ve been working on?
“We finished a project just prior to lockdown – wheat crop prediction. This was for the India Government’s Departure for Agriculture, using remote sensor data to predict crop yield at the village level.”
“The device is known as a UFO - it captures electricity consumption, data is then pushed from the cloud to analytics engines.”
Cool.
“Another was for an airport software development company, involving the rostering of various airport resources (human, such as janitors and luggage handlers, and machine)."
"Some resources are outsourced so forecasting and optimization is needed. We built the optimization engine in this project.”
“The objective was to minimise cost whilst meeting constraints.”
ML used to optimise borrowing requirements through predictive analysis. Nice.
“Another nice example is a time series project for a fertiliser manufacturing company, again providing demand forecasting.”
“In this kind of forecasting there are multiple products, so a time series hierarchy can be very complex. The client had accuracy ranging between 15-30% for all time series."
"Through the use of sophisticated forecasting techniques, we were able to raise the accuracy to 75-87%.”
Why is forecasting the core focus for you?
“Demand forecasting is a problem that every company faces. They say ‘an improvement of 1% in forecast accuracy leads to an improvement of 3% on the bottom line.’”
Right, it’s easy to understand. Applying ML to save costs has become a no brainer.
“But right now it’s very slow. Saying so, we are starting some new projects in areas of IoT Analytics (machine monitoring and image processing) in addition to our forecasting assignments.”
I've heard it can be highly competitive in India, how do you find it under normal conditions?
“Very good actually! The BFI (banking, finance, insurance) sector is highly mature, so they have great analytics practices. Manufacturing, e-commerce, retail, analytics… not so high. So, lots of opportunity in those sectors. It’s a challenge to get into these organizations. If companies are setting up their own analytics practice, then it’s difficult.”
Do you think India could become a data analytics powerhouse?
“All my life I’ve worked with American clients, so I understand the way they use data and the type of human resources they have. India is still very far behind.”
“But, yes, I think it’s going to be. Data maturity is not there yet and so many people are stuck on excel spreadsheets. They haven’t moved to tableau or power BI.”
“I think in a few years’ time, we’ll see a lot of companies adopting AI.”
Let’s hope so. Every major report I’ve read on global AI adoption has cited the widening gap between adopters and later adopters of AI as a major cause for concern. Things could get less equitable before they get better.
How did you get into this?
“Right place, at the right time! In year 2005 I joined SAS. It wasn’t pre-planned – being in analytics, but I joined and learned through the best analytics minds in the world!”
What do you find interesting about machine learning?
“It’s all about data. That’s what I’ve learned in all these years. You need to learn to read the data. Machine learning is just one of the tools.”
A lot of wise engineers share this sentiment.
“In the kitchen, we have the knife. You can perform various operations with it, but you still need to know what operations to perform. You need to learn to read the data. What type of tools to use, what type of insights you can get.”
“Also, you need to closely work with the business users. I worked with Amazon or Nestle data analysts, and you needed to understand their view in order to know the type of tools to use.”
What are the most important aspects of ML?
“I’m seeing a focus on deep learning, reinforcement learning. Everyone is looking at convolution neural networks, reinforcement learning, etc.”
“But people are missing out the basic statistics, on simple stuff: what are the basic distributions falling behind these?”
I’ve heard this one before. Data Scientists leaning towards self-structuring data techniques to uniform distributions!
“If someone comes to a doctor with back pain, maybe just exercise will solve it - they don’t need surgery! People in data science seem to be jumping to surgery.”
“There’s a preference to use unnecessary deep learning algorithms to solve even forecasting problems. The actual approach should be to use the most simple algorithm. If complex models aren’t able to beat the naïve models then what’s the use?”
Any new developments in ML you’re watching?
“Cloud ML is definitely coming up. Services like AWS ML are making data scientist’s lives very, very easy.”
“It’s not just about building the models in data science, it’s about cleaning data. They say data prep is 70%. ML ops (creating docker containers, scaling) this is really challenging and these cloud stacks are excelling in this. Takes a lot of the headache away.”
What about auto ml?
“It still has a long way to go. Each dataset is different. You need to understand each and every feature, how they interact, so you can come up with common features and know if some aren’t important. I just don’t think auto ml is there yet – a long way to go!”
What are you watching out for next?
“Managing data is becoming a big challenge. There’s so much data from so many sources, so creating a good data strategy, with strong data cataloguing is very important.”
“Less focus on data engineering and more on the integrity aspect – garbage in garbage out!”
“People build really advanced models then realize the data they fed those models was garbage.”