Interview with Abhishek Thakur | World's First Triple Grandmaster | Kaggle

hello this is Walter Reed from Kaggle aka inversion and today I am talking with epic Sakura who is kegels first-ever triple grande master that means achieving grand master status in competitions achieving grand master status in Colonel's and grand master status in discussions so first off congratulations that is one heck of an achievement and thank you very much yeah it was yeah it took away it took a while so we're going to get to that so why don't we start with just giving a little bit about your background introduction how you got into Kegel I'm sure a lot of people would love to know your story yeah so it's a it's a funny story so I joined cackled my profile says eight years ago but I started with competitions in 2013 six years ago so I was doing I was doing an internship when I was studying in my bachelors and there I heard about random for us and I told once my colleagues and he was like why don't you look at this site kaggle everyone is using random forest and you know random forest so probably you can win some computations so I joined Kaggle and I didn't do anything so I looked at the competitions and everything seemed like too weird to me you have to download files and upload files my internet is not that fast so in 2013 so I was at at that time I was doing my electronics engineering and 2013 when I was doing my Master's I saw everyone is talking about natural language processing people are talking about neural networks and machine learning so I should also start learning this and where do I start from so I remember two years ago this guy told me about this website called Kaggle so I even packed it I tried to search what are the machine learning completion platforms and then I found Kerala game and it couldn't register my account because I already had one so then I started with then I saw like okay I I was able to log in a password and all and then I saw this competition on recognizing facial expressions emotion recognition so that's where I started from started with very basic image processing approaches and with MATLAB so I was trying to determine the angle between I and nose and lips to find what emotion it is yes so that's interesting so you learned about random forest and you were interested in NLP and your first competition was image recognition so which know you haven't always been to image processing so when when I was doing my bachelor's the time I was working on fingerprint recognition and I was doing OCR recognition implementing OCR algorithms on microcontrollers microprocessors at that time and I heard about a little piece because my friends were working in NLP and they were using neural networks and RN and they just started at that time and they were talking about that that's how I got interested in machine learning so so you were using MATLAB word did you have a GPU when you started no I didn't have when I had a computer with 4gb RAM ok 4 gigabytes marquest so I'm guessing you weren't one of these competitors so there's a very very very very very small percentage of competitors almost actually you can count them on your hand that fine Kaggle join Kegel and do really well on their first competition did you knock it out of the park on your first competition or I wasn't 50% and in the first compeition as far as I remember there were around 60 or 70 competitors in that one because the image competition and yeah it's not like these days when you have a lot of people working in image competitions and I think yeah that was also won by University of Toronto guys yes so what's your educational background that can you into machine learning my educational background so I'm a bachelor's in electronics engineering my coding skills were limited to MATLAB and VHDL programming and I learned some C C++ on my own I was always interested in computer science but I couldn't get computer science for my bachelor's so I decided to do master's in computer science that's when I moved to Germany and there I was working my master's thesis was on computer vision and from there I got introduced to K nearest neighbors again and then I started with machine learning okay in in let's see so you were started competing with MATLAB how long did it take you to switch over to Python I assume you use Python mostly now yeah I used by the MATLAB would I start started the competition with MATLAB in 2013 I could Matt 11 2013 and yeah with Python it was always easy and I learned all the Python I know from Carol and from looking at what other people have shared their functions and all the code and it's quite similar to my nuts app so it's only I think the indexes are different very painful and hard to go back to when indexed yeah and it's open source and so I don't have to buy the license and at the time I was working for the University so I had licensed for MATLAB but I don't need that anymore right so I looked and it was three years ago this month that kaggle redesigned their progression system right it used to be the master was the highest you could get and three years ago they changed it so of the grand master status but also you could have the three tiers competitions kernels and discussion yeah so it took three years before somebody was able to get Grand Master and all three so why don't you talk us through what was your first Grand Master or was second and then what was the hardest for you to finally get when cattle when Kendall switched this they had the new progression system I I was a bit afraid I was like okay I'm going to lose my status and I'll be the I'll be a master and I then I have to work again to get the Grand Master but I got lucky because I had I think there was a computation StumbleUpon and there I got my first solo gold so I had that solo so I didn't have to fight for gold medals and competitions after that discussion so I'm very good at discussing things on cattle and so that got some wood wood ups and I I think it was last year when I became compeition description grandmaster end of last year or beginning of this year the thing it took me a while because I was not very active on Kaggle for in 2018 and half of 2017 and that's why I think it took a while to get description grandmaster and colonels was I think it was the most difficult one so but I got I got a lot of gold medals in the last few weeks and that's all because of the new competitions that were launched and how I was sharing my kernels like how I was designing them yeah yeah we launched five competitions in the last week of last month so that gave a great opportunity for somebody who is interested in jumping in quick and doing some starter kernels and getting up votes so that's fantastic and you know I remember back so I've been competing on keto for about seven years and I remember back even from the earlier days even before kernels before scripts URLs one that would post starter kernels in the forum and back then right you'd have to actually upload a raw Python file but you know it was at the beep that was it to beat the benchmark or eating beating the benchmark so you know so even back then so what was your motivation for you know for shirring even before we had this progression system so it goes way back in so I learnt everything I know about machine learning from goggles so I saw so people are using these different algorithms different approaches I would look at them and I was in the university so I had a lot of time to read stuff and I was anyways reading papers so I started reading papers and implementing them and back in I don't remember the year probably was thirteen or fourteen there was a competition called Amazon mmm access challenge and people share a lot of cool stuff in that challenge and then I saw like okay people are sharing this that's why I'm able to learn and because of that sharing I was able to get a good rank in that competition that was like one of my very first competitions and that's why I started sharing and when when you're sharing code people are also commenting on that and some people are people are being thankful some people get mad but that's that's a different story but people are also sharing how you can improve on what you have shared so they are also giving different ideas so some some people it's good for some people to learn new things how to approach problems like getting a quick start to the problem and it's also good for me because I learn from those comments ya know that that's fantastic it's interesting I join cattle in 2012 and that's the time that like Coursera started so now all of a sudden not only do you have software with Python that's free and scikit-learn but but also you know free education and with cattle for people to be able to share scripts and share kernels and share you know what they're doing it's this kind of amazing thing where we're you know anyone that's really motivated to learn can can learn these things and find these things but it's also a very intimidating right I mean it was intimidating I'm sure for you and you started it was from so I'm curious like so if somebody was brand new to Kaggle and a little worried about starting a competition what advice would you give them to getting started be persistent don't get disappointed so yeah I think it's true for you too I I know that you got a very low rank in the first competition you were doing I got a very low ranking the first few competitions that I did and the the basic idea is never give up and read the solutions after the competition has ended so there you can learn the most and just carry on with the next one so what if what if you fall down from top 10 mm of ranks you you will learn why it happened or what if you're not even able to make to the top 10 you will learn from the solutions of the other people and yeah that's that's what I do I'm still doing that so like even now if I'm unable to get a good rank or I don't know how to handle so many images right now it's just about the images so then I will look at the solutions of how others have tackled this problem and from there I learned and I can use it not just in cattle in other competitions but also in industry those and if I do yeah yeah you know there's there's a lot of people who say well there's a you know disconnect but when I started doing data science from not on cattle but in industry any every new problem that I saw I could relate it to a competition that I worked for their signal processing or image or whatnot so it's a huge advantage so would you consider yourself like a competitive person what are you driving towards this to be first the triple grandmaster or did it just happen in the last few weeks I was so I I wanted to get this title before somebody else did and since one more competition was anything the jigsaw one which got postponed so I thought like okay I should share more kernels before this competition ends but I also went for quality kernels so where you have something to learn from yeah one of one of my kernels that got a gold was just a joke kernel I remember now but that was long ago yeah mmm I think I think that's it about sharing stuff okay oh well fantastic so now right now you know you've achieved triple grandmaster do we need to add anything to keep you motivated or does the point decay and the ranking system is that enough to keep you yeah okay that's it do you have anything else that you'd like to add or ask before we and interview not so much I mean I think I've already said this before but I'm really thankful to goggle and everyone who is working there and it has helped me a lot in learning and also in industries so even I relate industrial problems to gaggle and the only difference that I see between industries and kaggle is okay now you nowadays you can use ensembles and stack models because you have so many resources available but the only difference right now is how you gather the data yeah yeah Cagle is very very nice in giving that nice rarely in the real world yeah well congratulations again on your achievement and you know thanks for all of the contributions you've given over the years I know I personally learned a lot of what you've posted and look forward to having you share for years to come so thanks again you much I'll continue doing it

Channel: Kaggle

Views: 76,462

Rating: 4.9575033 out of 5

Keywords: Kaggle, Kaggel, coffee chat, live-coding, live, learn, api, cli, python, data, data science, interview, questions, transfer learning, coding, networks, programming, technology, tech, machine learning, AI, artificial intelligence, coders, programmers, help, tutorial, projects, 101, rstats, stats, statistics, what is kaggle, how to, github, developer, kernels, datasets, data visualization, deep learning, sql, challenge, competition, whitehat, code, lesson, CS, Interview with Abhishek, Triple Grandmaster, Abhishek Thakur

Id: 8lniZVqRLA0

Length: 15min 24sec (924 seconds)

Published: Wed Jul 10 2019