Kaggle Grandmaster Panel at H2O World Sydney 2022
Kaggle Grandmasters discuss what it is like in Kaggle competitions and how to become a Kaggler.
Talking Points:
Speakers:
Dmitry Larko, Data Scientist, H2O.ai
Sri Satish Ambati, CEO, H2O.ai
Sanyam Bhutani, Senior Data Scientist, H2O.ai
Parul Pandey, Principal Data Scientist, H2O.ai
Chun Ming Lee, Data Scientist, H2O.ai
Shivam Bansal, Director, H2O.ai
Rohan Rao, Data Scientist, H2O.ai
Qishen Ha, Principal Data Scientist, H2O.ai
Mark Landry, Director Data Science & Product, H2O.ai
Read the Full Transcript
Sri Satish Ambati:
Hey, good evening. This is the penultimate session, I believe. This is the KGM panel out here and we have quite a good spread out here. We have Parul Pandey who actually comes in online. And I think she's right in the middle and then from left to right is, we have Sanyam, we have Mark Landry, we have sorry, Qishen right? Yes. And we have Rohan. We have Shivam, Dmitri, and Chun Ming Lee. So this is the KGM panel here. Welcome folks. Thank you.
Sanyam Bhutani:
Hey everyone. I think this is more rare than Haley's comet. Even though Kaggle's only been around for 20 years, this is the biggest gathering ever of Kaggle Grandmasters in Australia. Thank you. So I think Kagglers are an entrusting species, if I may, because if you sit next to a Kaggler. If you talk to them. They almost have superpowers. They're interesting people. They're like almost not humans. They're super humans in some ways. So the question I have for the panel is what is your superpower? Maybe you can start with Parul. What is your Kaggle superpower?
What is Your Kaggle Superpower?
Parul Pandey:
Be part of the session in person? Am I audible? I am?
Sanyam Bhutani:
Yes, now we can hear you.
Parul Pandey:
Okay. But thanks to the production team, for making me appear magically in between the panel. So a very warm welcome from India. As far as my superpower, I don't think I have any superpower in Kaggle. It's just constantly learning things. But one of the very important things that I have learned over the course of my Kaggle journey is writing good code. Code that is understandable by others. Because you know what? It is said that code is more often read than written and normally we have a habit of quickly iterating, especially in Kaggle in competition, when we take part. Quickly iterating over and finding the right solution. But what Kaggle has taught me personally is to write code, efficient code. If I go back and look at it after years, I can still recognize and understand why I put this specific line there. For me that would be it.
Sanyam Bhutani:
Thanks. Pavel asked me to bring her cut out. I had a very interesting conversation at customs, but they finally allowed it. Maybe you can go around the panel and start now.
Chun Ming Lee:
Sure. I guess in terms of my background. So I would say my superpower would be, in terms of language, maybe more technically it's called natural language processing. So I tend to focus on NLP competitions. It's quite interesting because I tend to focus on a certain domain. Whereas I have a lot of colleagues that are good at everything. So I guess this is my superpower, but it's also a limitation in the sense that, for example, when it comes to computer vision, tabular data, I think some of my colleagues are better. But NLP is my superpower.
Dmitry Larko:
Nice. I think my superpower is being persistent, actually. But to be honest, there is a fine line between being persistent and being stubborn, actually. I'm done not trying to cross it, actually. That's, but it's hard. It's really hard.
Shivam Bansal:
Well thanks for the question. Superpower. It always almost feels like an avenger, I would say. In my case my superpower, I would say is structured thinking, end-to-end thinking for open-ended business problems. I always try to focus on how I can break down a problem into a series of structured tasks, small tasks, and then I creatively try to solve them one at a time and that's how I have practiced this in some of the competitions on Kaggle as well.
Rohan Rao:
My superpower, I would say, is not marrying any toolkit or a library, right? So the field of machine learning is progressing and advancing so fast. I think it really helps and it's helped me quite a lot in constantly evolving myself. Just learning new things after every competition, whether it's new libraries, new techniques, or even new languages. To have that open mind and keep adopting as this space moves.
Qishen Ha:
I don't know whether it's a superpower or not. My background is in computer vision. I study and research computer vision during my college, but actually my first Kaggle gold medal is under NLP, a nature language possessing competition, and before I did Kaggle. I know nothing about NLP and I learn and compete at the same time. So finally I can get that gold medal. So I think the ability to learn new things is very important for me. Yeah.
Mark Landry:
I think mine would be trying to deal with difficult data. So when I see a competition that people are struggling to even understand what's the point or what's the metric? Those are the ones that attract me. To just go fight with the data and try to sort something out that people are either going to underfit because they don't understand or overfit because they don't understand. So those are the ones that I gravitate towards and do the best at.
Sanyam Bhutani:
I think there's also a lot of humbleness around this panel. There are rumors that Chun Ming can just read in predictions from the models, so raw predictions, and he just can stare at them and talk to the model. He's a model whisperer of sorts. AutoML and AutoD of H2O were named after Mark Landry and Dmitri Larko. Rohan is a Grandmaster in all categories. Shivam is on his way to become a quadruple Grandmaster. Quishen, whenever he joins a competition, no matter where he's on the public leaderboard, he always jumps up to number one eventually. So if you see on a competition the number one spot is gone for sure.
Becoming Kaggle Grandmasters With H2O Team
My next question, I'll start with Chun Ming. Many of us became Grandmasters after joining H2O. So what was that feeling like? And was H2O pushed for that? Was the team responsible for that?
Chun Ming Lee:
Okay. So I guess for the folks that are new to Kaggle, there are tiers. I guess Grandmasters is the top. The tier below is Master. So before I joined the company, I was a master. The head of the GM team, a very funny guy by the name of Olivier based out of France, was saying that. I was asking him, "Can I join the Grandmaster team and I'm not a Grandmaster yet." And he said, "No problem. Don't worry about it." So I think in that sense, there was no external pressure, but I did feel a lot of social pressure. Cause I'm dealing with the Grandmasters on a day-to-day basis and I'm calling myself part of the Grandmaster team. So I think there was some pressure, but it worked out well because I joined. I've been at the company for about a year and at the start of the year, I competed in my final competition to hit Grandmaster. So I think that was a big achievement for me and I'm very happy about that.
Sanyam Bhutani:
Maybe I can ask Parul as well, because Parul joined us. She's one of the best writers in machine learning, period. And she joined us and very casually decided to pick up notebooks as a category. I think she was one of the fastest ever to become a Grandmaster in that category.
Parul Pandey:
So yeah, I think when I joined H2O I didn't even use Kaggle that much. I used to use it to download data sets, and that's all because I thought Kaggle was all about competition. When I came to H2O, there was this group of kgs and I would always wonder what it takes to bring that group. And then another thing was that there were very few females in the Kaggle arena at that time. Definitely Kim, who was also a Grandmaster at H2O. She's the competition's Grandmaster. There was Rachel that time who was in Kaggle. But other than that, I would go and I would see the leaderboards, I would see the discussion forum. Even that was all. It was an all boys game. Why is that? There are no females even though everything is open.
And I thought probably. First I tried Kaggle just to understand more about it and thanks to Sanyam and Rohan and SRK, who used to be a part of H2O. These people encouraged me. You can pick it up and slowly bring work on it. And then I also thought, using this as a means to educate the community about Kaggle, is more than competitions. Kaggle is a complete world. There are people who are interested in curating dataset or working with them. There are categories for those people who are interested in writing good quality notebooks. I mean, look at Shivam's notebooks, they're just superpowered. Look at those for competing in competitions again and just networking with the best minds on this planet. I have known so many people through Kaggle. We are living in different parts of the world. Nobody knows each other outside the realm of Kaggle and you learn from them. You learn from distributions, you can Kaggle them, you create a team, you work with them. So that's the beauty of Kaggle and that's what I wanted to present to the world. Also encourage people to not be intimidated by Kaggle, but try and experiment because this is that platform which would give you so much without charging a single penny. The relationships that you create here, the networking you create here, they last for a lifetime.
Sanyam Bhutani:
Thanks Parul. Would anyone else like to chime in? Maybe I'll pull in Rohan, because he became a quadruple Grandmaster out of pure boredom. As he says, "Out of pure boredom in Covid." We would watch Netflix series and he decided, "Okay, let me just conquer every category on Kaggle," just very casually.
Rohan Rao:
Yeah, I think that was an interesting period. I mean, COVID has been quite a hard period for a lot of people. And when it began, I think a lot of people suddenly found that they were working from home, me included. And yeah, I mean, a lot of people spent time on Netflix and I was not very active on Kaggle at that point in time. But then I just decided, "Okay, let me just take this as an opportunity since I suddenly have a little bit more time." And I decided to pursue the other three categories where I was not a Kaggle Grandmaster. And fortunately in a span of, I think, eight to nine months I was able to complete the remaining three titles. So, I mean, it was unplanned and I made use of a little opportunity that came my way.
Mark Landry:
Maybe I should take it too cause I'm probably one of the more well-known ones, that have come from one to the other. So I was on the first, it wasn't a Grandmaster panel, because there was no Grandmaster title when we did the first H2O Kaggle group. But yes, there was a lot of pressure. It felt like people didn't put that pressure on me, but it was like a Grandmaster to us and things like that. You had to say at some point. So there, at that time, I had the second or third most silver medals, but I didn't have the gold medal I needed. But you have to do what you do. And so eventually it happened. I never stopped trying. I wouldn't say I tried any extra hard. You just keep at it. I mean, I think Dmitri said, "Persistence." I imagine if you go back to a lot of these panels, we put a lot of time into these things. Yeah, I would say there was a lot of external pressure and it was a relief to get it. But then I've continued and gotten a couple of their gold medals after that too. So it's just, it's fun too. And you still learn stuff even after a while.
Sanyam Bhutani:
There's a difference between Grandmasters' level of casualness. I'm only a discussion Grandmaster and I've seen these guys show up on the leaderboard and they just be on the top with like, very initially. Shivam, Parul, all of them write incredible notebooks, just in half a day. And it's mind blowing to see how fast they can iterate.
Applying Kaggle Experience
Okay, I'll move on to the next question. There's this debate, which I'm sure all of us don't agree with. Is Kaggle relevant at all? I'll ask the question. Where and how do you apply your Kaggle experience actually?
Shivam Bansal:
Well, I can start with this one. So one of my core responsibilities in the company is to interact with various customers. With customers and with Kaggle I do feel some similarities. The similarities are that every customer problem is different and every solution can be different as well, but in general, the approach, the workflow, the mindset can be essentially the same. The same practice I have experienced in Kaggle competition as well on Kaggle, there are different data sets, different types of problems to be solved, but again the approach remains the same. The idea of experimentation, the idea of trying out various things, trying to combine various things in the end and then try to see what works, what gives the best solution. So I try to not take the exact, let's say technique, that I learned in Kaggle, try to solve customer problems with that. But the mindset of persistence, the mindset of experimentation, the mindset of end-to-end thinking that I see in both the places in the customer problems as well as in Kaggle as such.
Dmitry Larko:
So on one bright and sunny day at my early days in H2O Sri approached me and asked me, "Hey, Dmitry, can you come up with a journal script? You can just put the data set in and produce a more or less good solution, basically, right?" I was like, "Sure I can do that." When I sat and realized I cannot, that's how DriverlessAI actually started. So basically I used all my experience, I got on Kaggle, and some experience I didn't get from anywhere. I just used all of my knowledge basically and knowledge I got in the process of building the tool into the DriverlessAI. And yeah.
Working in Teams
Chun Ming Lee:
I guess maybe if I can hop in just to switch tracks a bit. I think there's one non-technical aspect of Kaggle that maybe we don't even think about. In Kaggle competitions, you can actually team up with people, right? So it's not. We are not necessarily just competing by ourselves. We can form teams. These teams are in many cases, you're working with people that you've never met before, that you'll never meet in person, and your only interactions are probably going to be through Slack or through emails. So I think one skill set that we are forced to pick up is: how do you communicate with your teammates? It's actually harder to work with the teammates on Kaggle because they're not being paid for their time. They're doing it on a volunteer basis. So there's this notion that you have to really work hard to communicate well with your teammates, probably even more so than with actual colleagues. So I think that's something that we don't even think about, but that's a very important skill that we can apply to real life as well.
Sanyam Bhutani:
Does anyone else want to jump in? Okay. I have an interesting follow up as well. We have the largest Grandmaster team as you all know. Who's your best Kaggle companion or competitor inside the team at H2O?
Mark Landry:
I'll go first. So I have to echo what he said, just waiting for a little bit. The teamwork for me is what's fun and sharing it with a team and thinking, and I go about it a different way than others do. I just want to talk about the problem and share stuff. So for me, Brandon Murray, who's been with me for about seven years almost now and he's on the Documented AI team and SRK, who's been with us before. So I've teamed up with each of them multiple times. They both go about problems in a similar way that's really complimentary and the way I do. I see what they do, I can't emulate it. I know what they're up to, but I just can't do it the way they do. So, they're both good for just talking about problems and having someone to work through it with a common understanding. So those would definitely be the top two for me.
Shivam Bansal:
Yeah, I can share maybe. For me it has always been Khun Hao who was based in Taiwan. He used to be part of our team, but he decided to start something of his own. So together we competed in more than six competitions in a span of eight or nine months. It was last year and we used to team up and split various types of tasks that can be taken up, like trying different models, trying to combine them, see what works. And we won four silver medals, one gold medal and won one solar competition with all the sharings from different mindset among the team assets. So it has to be Khun Hao for me.
Dmitry Larko:
To me it's going to be Marios Michailidis. I teamed up with him inside H2O, but I also have to acknowledge, because I'm a good son, actually, right? And I have to acknowledge my father, who is a Kaggle Grandmaster as well and with whom I mostly team up on every single competition.
Sanyam Bhutani:
That's some serious peer pressure, right? If your Dad is a Kaggle Grandmaster.
Dmitry Larko:
That is serious peer pressure on my son, who is actually 16. Yeah. I'm good at extrapolating things, right? So you, I mean, it's invertible, right? So if your grandfather is a Grandmaster, your dad is a Grandmaster, well what else can you do in life? Basically.
Sanyam Bhutani:
It's this to me, no one wants to name competitors, maybe Parul?
Parul Pandey:
No, nobody's competitors. I learned from everybody. I would like to keep up with everybody, but mostly I team up with myself. Because I'm mostly part of the competition. But we did participate. So the females at H2O, we participated in a competition, which was women in Data Science competition. I think that was 300. Real fun and how we all got together. We were six on the leaderboard that time. The next time other H2O females got together, they were number one. And so yeah, so every year that happens. We look forward to that. And we really look forward to at least, going on, staying in the top level leaderboard, so that at least. Because this is a competition which is targeted towards women in data science. So we make sure that the top two positions are actually held by women. So we have a lot of pressure to maintain that.
Sanyam Bhutani:
Thanks. As has been established, very loud and clear, no one on this panel likes me and considers me their companion. So I'll sadly move on to the next question.
Parul Pandey:
But Sanyam, you've been asking questions to everybody. I would like to ask you a question. Who all the questions that you ask everybody. Who is your motivation for becoming a Grandmaster after joining H2O? Who is your favorite competitor and things like that?
Sanyam Bhutani:
So I'll share the real story. You get a jacket when you become a Grandmaster. Someone on this panel offered me the jacket and they gave me 45 days to do it. So just as a challenge. I did it in 22 days. And it's only in discussions, but that's my proudest achievement. So that's the honest story behind it. Before that, there was no motivation, but the jacket was motivating enough.
Kaggle Battle Stories
So this aligns with an audience question I see bubbling up. What is your favorite Kaggle battle story? I consider the competitions are really like a battle. You're always looking at the leaderboard. This, you're always staring at the differences in scores. You're seeing who's on top, who's moving up. You sleep, you're down by 50 positions. It's, I'm also almost romanticizing it, but it's quite romantic in that way. So what's your favorite battle story from Kaggle?
Dmitry Larko:
So there was a competition, they asked on this competition, you'd be asked to segment ships out of satellite images. What was interesting about this competition as a battle story, basically, right. So I picked up the model, it was a fairly good model, basically, right? Fairly good architecture. I design a pipeline and I keep this model to train, let's say, for 12 hours, right? It produces some results. They may let this model continue training for another 12 hours. In another 12 hours, I think at the end I had trained for, I don't know, 64 hours. Basically at the same time on a different machine. I did something else. I tried different experiments, I did different submissions, but they went well. While this, where all these additional submissions actually went, were pretty terrible actually. But this model, I keep training and training and training actually ends up in 12th place. A single model out of the whole competition. It was quite an interesting story. And a lesson learned to me. So you adjust it. With a good pipeline and the right architecture, you can go very far.
Rohan Rao:
So I can share an interesting story. So this is quite a few years back. It was a weekend hackathon on analytics with them. So it was one of those competitions where generally a few team up. It helps because the competition duration is just three days and I really wanted to team up with Mark on that. I'm not sure Mark, if you remember this story. So I typed out a message, an email to Mark, this was a little before I joined Edge Tool. And I was just going to send the email and I saw that Mark had already teamed up with SRK. I really felt so bad about it that I decided I'm going to spend the entire 72 hours trying to beat them. And I ultimately did manage to just about beat them. So it was a very interesting weekend for me.
Mark Landry:
Those are fun and we did later do one.
Rohan Rao:
We won together.
Mark Landry:
So yeah.
Sanyam Bhutani:
I think both of you were number one and two on analytics with us. Ranking as well at some point. So yeah, they're not just good on Kaggle, they're good on any problem you throw at them to solve. Parul, do you have any interesting vital stories you'd like to share?
Parul Pandey:
I remember, I think initially I was attempting a competition on my own and it was like a few hours left. And I messed up with the best model because long ago I used to train models in notebooks and I had those entitled IP&Y. So many of them that I forgot which one was the one holding the best model. And unfortunately I missed out on a bronze by just, I think, one position or something. Then after that I really invested in creating a nice pipeline and on code. Basically the way I started my conversation was that I realized, you also have to invest in making a good pipeline and reproducibility actually it is something that I realized. How important was that? Because then you can get the medals.
Mark Landry:
I think for one of mine, like the. It's the battle of teamwork. And so there was one of them. One of the gold medals with SRK and we used to communicate by GitHub issues back then. This is fairly primitive, but I had stumbled on something like a really big piece. He was doing most of the pipelining at that point. But there was a big gap between where we were and the top few. And when you spot that, more for tabular problems, if you know you've missed something and you just want to keep trying until you've found it. And there's just like this 3:00 AM like, "Aha, I found it!" Posting to him. And it was more fun to almost send him that message than it was to really find it. Cause now I've done my part and now we can go chase it. And we did. So there's a lot of those little stories and they almost always, for me, just involve whoever's on the other end of the line, essentially the teammates that you do that with.
Sanyam Bhutani:
For me, it's always the opposite that I find something interesting and it doesn't turn out to be. Constant disappointment.
Mark Landry:
There's a lot of those too.
How to Become Kaggle Grandmaster
Sanyam Bhutani:
Okay. We can. I think we can move on with audience questions. How does someone who's new to ML become a Kaggle master? So in short, what's your advice to someone just starting on?
Dmitry Larko:
I mean the major advice is going to be, you definitely will need, if you want to become a competition Grandmaster, you'll need a lot of free time. Actually, the 5,000 tower rule still applies here. So you have to be ready to spend 5,000 hours competing and learning about machine learning in general, right? 5,000 hours, actually that's two years and three months. A full-time job. So at least that was how it was for me, actually, right? So one moment of time I learned about Kaggle and I basically for two years, I have two jobs, my daily job and my job at night: which was Kaggle. So that wasn't easy.
Chun Ming Lee:
Yeah, I guess maybe if I can hop in it's interesting in my case. Because before joining the company and before getting into data science my most recent education was actually getting an mba. I was a management consultant for a while, but due to burnout, I quit my job and I actually spent a lot of free time catching up on machine learning. So I think it's really hard to overstate free time. In my case I quit my job. But I think the question being paused is especially if you already have a job, right? I think maybe start small. I think one mistake a lot of beginners make is they try to dabble in all domains. So maybe one way of approaching it is to pick the subject that you're interested in. So it could be languages, in my case NLP. It could be computer vision, right? So pick a subfield that you're interested in and the focus, I think, would give you an advantage over someone that's just looking at an entire field as one. So maybe that's one way of thinking about it.
Mark Landry:
I think one thing I would see with people that don't, like if they put in the time but don't become, don't achieve the success they want, is you just have to be honest with yourself about what you're learning and you're making. Look around. There's a lot of information to pick up, but if you just wind up like just going for the best result, and you're going to give you an excuse for yourself why you didn't get it. So early on, especially seeing what other people did that place better than you. And sometimes that can be really simple. It's like, "Oh, I should have thought of that." You can dismiss that if you want or you can learn from that. And I think that so long as you, it takes time, like they say, and I don't know if the question was Grandmaster, but master's is a little easier to get. But I think paying attention while doing it and just persistence as well.
Shivam Bansal:
I can say that data science seems to be a big, wide field. It feels intimidating to start with and sometimes in the learning journey there may be a lot of distractions. Many beginners make this mistake that they tend to deviate from what they started. I have made this mistake many times, but over the years I have tried to correct it. Like, whatever we have started, it could be a skill, it could be a field, try to finish it. Try to complete what we have started and that really has helped both in Kaggle as well as in H2O and in other places as well.
Rohan Rao:
Yeah. Maybe just a little bit to add. So even if you look at many of the top hundred Kagglers or many of the fairly successful Kaggle Grandmasters. I think a lot of them come from very varied backgrounds, right? So, and that's a sign that there is no set prerequisite or requirement for someone to grow or become better in it. Is just an investment of time, effort, learning and just self improving as a lot of the points that have been mentioned.
Careers Outside of Data Science
Sanyam Bhutani:
I think we are running out of time, so I'll try to squeeze in two questions. Have you had it, this is similar to what you were just talking about. It says, "have you had a career outside of data science and what was it?" Maybe you can start with Parul, because I know she has an interesting background.
Parul Pandey:
Yeah. Definitely. I come from an electrical background. I used to work for the power industry in India and in working for an industry, the power industry in India's pretty challenging because at that time they were shifting from government to private. A lot of data was involved and automating, a lot of it was involved. It wasn't called machine learning then, but we were actually crunching such amounts of data and we had our own names. So I come from that background and we did make sense of data. We tried to understand what it is and coming to machine learning, it was pretty natural to see that there are things that are available which can help us automate things faster and create solutions for the public. Create solutions out there, just going to help people in the longer run. So yes, I come from an absolutely, so I was from electrical engineering. I just studied computer science in my first year of engineering, but then I had to learn and trigger everything myself. I was on maternity break, so yes.
Sanyam Bhutani:
Thanks Parul. Anyone else wants to jump in?
Avoiding Burnout
Okay, I guess we can take one more question then. How do you manage your time and avoid burnout while also doing Kaggle and having a full-time life outside of Kaggle?
Shivam Bansal:
Yeah, I can say that if it is feeling like a burnout, then probably it's not something interesting that you're doing. So personally, I try to choose those areas where I can enjoy that along with other activities as such. So if I feel burnout then I try to avoid that type of task immediately. Essentially.
Dmitry Larko:
Yeah, I mean to me, I think I've been burnt out twice from a Kaggle. Exercises, a lot of sleep, taking breaks from the Kaggle, that's usual stuff.
Sanyam Bhutani:
Qishen, do you have anything to add?
Qishen Ha:
Yeah, so to Kaggling is not only to compete, but I think that the main idea of Kaggling is to learn new things. So the question is actually, so how do you deal with your work and learning? So I think people should first focus on your own work and when you feel tired or that. It's time to learn some new things, you can go to Kaggle.
Mark Landry:
Yeah, I would second what he said on competing. There's different ways of competing too. So, I've done about a hundred competitions. So we've all competed to try to get better, but you can find a little pocket. So if you don't have enough time on something, there can be some little thing that I'm trying to optimize. Kaggle to me, their main purpose, there's a lot of great things they do, but the main purpose is to hold a competition with a clean leaderboard that you can trust that nobody else knows the answers to. So you're all on equal playing foot. You don't know what those predictions are doing. And that's important because you can easily cheat on your own. You can see something and exploit that without thinking you are, but you can't do that with Kaggle generally. So you can find little pockets that you can do even for a few days or something like that.
Like, I'm going to find the coolest way to do this and see if I can rise up a certain number of leaderboards. Even if I'm nowhere close to the top. So there's little ways that you can find some way to be interesting, I suppose. Even if you're not going through the grind of trying to be at the top and that can be a burnout for me. But sometimes especially like letting your teammates down. That's not cool. So you've agreed with somebody and you come to terms as I'm going to put this much effort, but you're never explicit about it. But that can be the grind for me. It's like a month in and we're doing well and I'm letting them down because I'm having to do whatever it is. Home, life, work, things like that, communicating that. But that's tough at that point. Like, "Okay, I just gotta sit down at two in the morning and try to get some work out." That happens.
Chun Ming Lee:
I guess maybe some practical advice I could give is, I mean. Mark just mentioned participating in a hundred competitions. I guess in my case, I've been on Kaggle for five years. I participated in a total of 12, a dozen competitions. So I guess one very practical answer is budget. Maybe one competition or two competitions a year and there's a concrete start and end date versus, I mean, a lot of our colleagues are able to do that just participate. But if you just set a time budget, this time of the year, I'm going to work on one competition and I'll do it again next year. I think that makes it very easy for you to budget your time and also tie it to work.
Sanyam Bhutani:
As Mark said, it's a bad thing to let your teammates down and I don't want to let our team down since we are already over time. I'll try to wrap up, but anything you all want to add before we wrap up? Parul, Qishen, anything you want to say before we wrap up?
Parul Pandey:
I will just say enjoy Kaggle and just make the best use of it. It can be very addictive, especially because ranks are involved and you're in this constant cycle of retaining your rank. But just enjoy it and then you'll get the best of it.
Sanyam Bhutani:
I think that's a great mic drop moment to end. People, these are really the smartest brains in data science. You can find them on Kaggle, you can find them on, I think everyone is on LinkedIn. You can also find them on Twitter. They're very active on social media. Just look their name up on Google. You'll find all of their feeds. Connect with them, find us after the panel as well. I think all of them are open to talking. And thank you so much to the virtual audience and to you all as well for attending this panel. And thanks to Parul for joining us virtually. I still have the cut out with me.
Parul Pandey:
Put it out there in the lobby. Okay.
Sanyam Bhutani:
Thanks everyone. Thank you.