In the developer series, we reach out to developers, practitioners and experts from the machine learning community to gain insights on their journey in data science, and the tools and skills essential for their day-to-day operation.
For this week’s column, Analytics India Magazine got in touch with Konstantin Yakovlev, a Kaggle master who is a seasoned product manager with more than 14 years of experience and a not so straightforward journey to the top of ML world.
How It All Began
Typically we would start an interview with an expert by probing into their educational background; however, when we have posed the same question to Konstantin Yakovlev, we were left wondering. Because he casually said that he has no proper education qualification and neither a college degree.
Subscribe to our Newsletter
Join our editors every weekday evening as they steer you through the most significant news of the day, introduce you to fresh perspectives, and provide unexpected moments of joy
“The simple answer is that I do not have complete academic education.”
After enrolling in Moscow Institute Of Physics And Technology (MIPT) — one of the premier Russian institutes, Yakovlev dropped outpost a year and then moved to Portugal at the age of 18. After residing in Moscow for most of his teenage years, he then went on to join the University of Lisbon, which too, met with the same fate as that of its Russian counterpart.
Eighteen years later, today, Yakovlev is not only a seasoned product manager but also a self-taught machine learning practitioner who has topped global level competitions like Kaggle. Currently, he is ranked number 70 in the competitions, 14 in Kernels, and 10 in discussion forums. This is quite impressive for someone like Yakovlev, who had no structured academics in his entire life and still can compete with the participants from the top universities and companies around the world.
A lot of excellent specialists work in the IT field without a specialised education. Well, is it a good option? I do not know.
Yakovlev believes that the modern-day pedagogical approaches are overrated, and most of the universities offer the outdated curriculum. However, he stresses that non-academic career success is not for everyone. He also insists that university education does assist many with the process of learning oneself and shouldn’t be sidelined entirely.
“However, university education gives the foundations of how to properly build the learning process, how to structure the material, how to write, and how to present as well as convey your thoughts.”Konstantin Y
Though Yakovlev regrets the fact of not getting a proper degree once in a while, he admits that this whole ordeal didn’t affect his career or his salary. Today, he is heading the organisational structuring of an undisclosed company in Dubai and is being paid handsomely.
In his last company, he has worked as a product manager in the financial sector, overseeing online banking, SVF, e-wallets. During one of his stints as a product manager, Yakovlev had to deal with problems concerning predictive analysis. He was then introduced to the fascinating world of machine learning algorithms and their insane predictive power.
His fascination for the field has accumulated overtime where he even had to quit his job because of no visible scope for ML. In his pursuit to master machine learning, he also failed to find a good certification program.
According to him, university education often provides a lot of emphasis on theory in isolation from state-of-the-art practices.
Though ML practice doesn’t get as much coverage as the theory does, Yakovlev was determined to test his knowledge by putting it to test on the best available platform—Kaggle.
“One should be able to apply knowledge and tools, rather than just ‘bone up’ the definitions.”Konstantin Y
Being A Part Of The Cream
Yakovlev’s consistency has fetched him gold in one of the most popular Kaggle competition on IEEE CIS Fraud Detection.
This competition required to benchmark machine learning models on a challenging large-scale dataset which consisted of a wide range of features from device type to product features.
A successful model would improve the efficacy of fraudulent transaction alerts for millions of people around the world, helping thousands of businesses in reducing their fraud loss and increase their revenue. Out of 6000 teams, Yakovlev came out on top.
When asked about — what it takes to be at the top, Yakovlev, names three things:
- Motivation or desire
This reminds us of how fortunate we are to have access to a vast amount of online resources for self-education – read, listen, code. He urges the amateurs and aspirants to focus on the goal and dedicate free time for Kaggle.
“There is no silver bullet. One needs to grant themselves free time for Kaggle, at least 2-3 hours a day. More like doing a gym, this should be a regular event, because one-time bursts of enthusiasm will not help you make a quantum leap.”
According to Yakovlev, there is a routine to tackle any ML problem, which is as follows:
- Data analysis (visualization, cleaning)
- Metric understanding (problem analysis)
- Baseline model
- Cross-validation scheme
- Feature engineering
- Feature elimination
- Different models
- Low-cost function optimization
- Hyperparameters tuning
Though the routine is more or less the same for many practitioners around the world, Yakovlev hints at not having any favourite tool or techniques.
But for each item, many details or options may work on some tasks, and may not work on others.
For instance, Yakovlev points out that data cleaning itself can have many angles to it from values imputation, checking outliers drop to handling missing data and segmentation.
What Separates Good From Great
No matter what brilliant specialist you are, stresses Yakovlev, the result is always made by the team. He honours one’s ability to be a team player but also warns that accepting weaknesses and improving is slowly becoming obscure.
It is often difficult for us to overcome our Ego and admit to ourselves that we require help.
On a closing note, Yakovlev advises newcomers to read vociferously and advises the aspirants to be a lifelong student of the craft that they are mastering. And needless to say, in fields like AI, this trait is almost mandatory.
Be it one’s “pet project” or preparation for participating in a competition, “pause when you stumble on an unknown area and try to figure it out by reading and coding obscure things. Read all that comes to your hand — papers, blogs, books, codes.
For beginners, he recommends working on code writing ability with Python, R, etc. and developing a knack for drawing insights from results through visualizations.
Hype Around ML
Talking about the hype around ML and the overwhelming feeling of getting carried away, Yakovlev warns about the two common ML myths:
- ML is magic
- With ML, one can take a ready solution
To clarify, he said, ML is not magic but an augmentation of human capabilities. He further explained the process of seeing connections and patterns that an ordinary person isn’t capable of seeing, working 24/7 with almost instant response time, the ability to work with incredible amounts of data, and how success is usually a collective effort of analysts, engineers and mathematicians.
Yes, we already have many out-of-box tools, and several managers, however, that isn’t enough to make a model to get an excellent result. Firstly, each task has its tools like models and algorithms, and secondly, what has been working for fraud detection with online payments will not be working in the offline sector.
He also adds that an expert is someone aware of things working under the hood and then uses that information to make the right choice. A specialist has to weigh-in thousands of details to converge them to one working solution.
Here is a list of resources, according to Yakovlev that would come in handy for enthusiasts and experts alike.
In his latest Kaggle competition, Yakovlev earned silver for developing a solution that would give insights into how companies can make significant investments to improve building efficiencies, and to reduce costs and emissions.
Now since we know that Yakovlev is a self-taught master, we, at AIM, brought to his attention the much talked about the stereotype of Russians being good at mathematics and if that has helped him win competitions. He debunked it by saying NO.
At least in his case, he finds himself skimming through high school mathematics sometimes and suggests everyone do the same, i.e. going back to basics because there is no silver bullet for mastery.