MLOps Zoomcamp: Free MLOps course. Register here!

DataTalks.Club

Home Articles Slack Events Podcast Books

From Physics to Machine Learning

Season 3, episode 6 of the DataTalks.Club podcast with Tatiana Gabruseva

Links:

Did you like this episode? Check other episodes of the podcast, and register for new events.

Transcript

Alexey: This week we will talk about switching careers from physics to doing machine learning. We have special guests today — Tatiana. Tatiana is a computer vision deep learning engineer. She is a Kaggle competition master and she currently works as a senior machine learning engineer at Cork University Hospital. Welcome Tatiana. (1:57)

Tatiana: Hello! Thanks for having me. (2:18)

Tatiana’s background

Alexey: Before we go into our main topic of switching careers from physics, let us start with your background. Can you tell us a bit about your career journey so far? (2:20)

Tatiana: I started my career with a PhD in physics. Then I worked in science and different institutes, universities developing different optical systems and lasers. Then I started to have children. I had two maternity leaves in a row. During maternity leave, I decided that it’s nice to stay active. Nano photonic is not an option because you need a lab for that. A lot of my friends got interested in machine learning and they started talking about that. I decided to check courses on artificial intelligence and machine learning and I liked that. I started with courses, doing them one by one. Eventually, on maternity leave, I realized it’s what I want to do. It resulted in a career change at the end. (2:32)

Tatiana: I also got a small internship that brought me to that job in the hospital. I developed necessary links. In that internship, I was doing some machine learning projects for analyzing signals for medical needs. In the hospital, they had a similar project. My name eventually came up and that’s how I got in a position. Now I am developing a decision support tool to help doctors to identify anomalies during the labor process and act promptly when a C-section is necessary to prevent some bad severe outcome during labor for the fetus.

12 career hacks and changing career

Alexey: You mentioned quite a few interesting things in your story. I think we will cover some of them later, like how you approached your learning. A few months ago, I was checking my Twitter and I saw your thread there. After reading this thread, I decided to contact you. It got my attention, it was really interesting. The threat was about 12 career hacks. Can you tell us about this threat? So what are these career hacks? (4:20)

Tatiana: On Twitter I was talking about career change. It’s a popular topic. Everybody changes career, at least a couple of times in their lifetime. The first time is always tricky. People worry about career changes. They ask questions on “how to do it?” and “how to approach that challenge?” I learned a few useful hacks during my career change. I learned it the hard way. I decided I should share my experience, so people can speed up the process a bit, using my advice, if they find it useful. (5:05)

Hack #1: Change your social circle

Tatiana: The first career hack — one of the most important ones — is to change your social circle. Be careful about your social circle. If you have friends from childhood and parents, you don’t change them. But you have a lot of colleagues and you have a lot of shallow acquaintances. If they are not making a difference in your network, if they are not helping you in your career change, if they keep telling you that “You are going to fail”, “It’s a bad idea, you have a good job, why are you changing your career?”, “Changing career and maternity leave that does not work, nobody does it”. In this case this circle does not help you. It’s important to cut off such social interactions. Then you will have a space for new people to come into your life — you will have more free time. Then you can have people that are more supportive, that already changed their career and that can help you and guide you through that. (5:47)

Tatiana: That is essentially what I did. I cut off about 70 percent of my communications with ex-colleagues and all those pessimistic people who are telling me that I will fail, changing careers is a bad idea. When I stopped communicating with them, I started to look for options. That’s how I found the ODS community. I entered the community and that is where I found my new network in a career. Not only in my career now, I also have friends there. They were very supportive.

Tatiana: There is a saying that you are the average of three to nine people, people who you are friends with. So, it’s important to be careful in choosing your friends and choosing your circles. That is what the DataTalks.club community is also abou — to create that kind of circle, that kind of social interaction that helps you grow.

Hack #2: Forget your fears and stereotypes

Tatiana: The second hack is to forget about your fears and stereotypes. A lot of people are quite afraid of change. They have a lot of stereotypes, “Changing career at 35, oh you are too old”, “Changing career on maternity leave, that is not going to work” and so on. Those are not real. Those are just your fantasies. They have nothing to do with reality. (7:50)

Tatiana: Some people contact me and ask me questions like “I am 26. Am I too old to change my career?” And I am like “Seriously? You are 26. You are not too old!” And you are not old at 36, 46, 56. You are going to change your career. At 60, there is no chance you are going to do the same stuff that you are doing today. That means that between today and between your 60s, there will be quite a lot of changes anyway. Just get used to that thought.

Hack #3: Forget distractions

Tatiana: Career hack #3. Forget about distractions that do not bring value in your life. In my case I realized that such distractions were Instagram, Facebook, watching movies. Even Twitter and LinkedIn — those nets are important for PR — but when you are changing careers, you have nothing to PR yet. You did not yet make that pet project. You are just learning. So, shut down all the social networks. You will see how much time you have. You need to create time. When you are on maternity leave, you do not have time. You only work when you have a chance. If you are going to spend time on Instagram and Facebook, it’s just not practical. (8:53)

Tatiana: You need to cut off all those distractions. Some useful things like LinkedIn, you are going to use them later when you are ready. But don’t open that when you are changing your career and you are learning. Because you open it for a minute and then you absorb for 15 minutes. That’s what usually happens.

Alexey: I think there are some apps that help you do this. You can have an extension for chrome that blocks Facebook, Instagram. You can also install it on your phone. Did you use something like this? (9:55)

Tatiana: Well, I have a good internal will. If it helps, if you need, use an external thing to block you from that. (10:10)

Alexey: That is what I need usually. (10:22)

Tatiana: Yeah. When I was stopping smoking, I got off smoking. And that was it. I used to smoke quite a lot, like a pack of cigarettes. But I just cut it as soon as I got pregnant. That is it. Same here. I just cut it and did not open it. Now I open it again, but that is because I need that now for career options — to network and so on. But not all... For Instagram — that’s it. (10:23)

Hack #4: Don’t overestimate others and don’t underestimate yourself

Tatiana: When you are changing a career you have this impostor syndrome. Suddenly, everybody is so smart. They know everything. But you don’t know anything. You keep learning and learning and learning. But you still have this impostor syndrome — because it is hard to validate yourself. So, the hack is don't overestimate others and don’t underestimate yourself. (10:49)

Tatiana: But how are you going to estimate your skills correctly, if you change in career? You need external validation. For external validation, I started to go to interviews. Interviews show you which skills you lack, which skills you are good at. It helps you to put your place as a specialist with this skill set on the market.

Tatiana: Go to the interviews even if you are not ready to start the job. My mistake was when I was on maternity leave, I was waiting for Alisa to grow up a bit, so I can start going for interviews. That was wrong. I could have done it earlier — just for external validation — to see which skills I need to increase and which skills are actually okay — to get rid of that impostor syndrome.

Tatiana: But the best thing to fight the imposter syndrome is not to go and interview yourself, but to conduct interviews. Be on the other side of the table. There are mock interviews. A mock interview is not a real interview. There are platforms for that like pramp — you interview others who are trying to go to the big tech companies. When you are on the other side, when you are an interviewer, you are going to immediately see all the weak sports for yourself. That helps you: now you can adjust your own behavior that fits better and helps you to pass the interview again.

Alexey: How does it help? When you interview others, how does it help to find weak spots in yourself? (12:48)

Tatiana: When you interview others, and the other person keeps talking on and on about his experience, you realize “Oh my god, I am doing exactly that. I need to shut up. I am talking too much”. When I saw it in others, I was like “That’s not good”. I realized I need to write a concise story in two sentences, maximum three and practice it. Because if I am going like this guy, that is not good. (12:55)

Tatiana: Or you ask a person to code, and the person immediately starts coding without thinking how to solve the problem. You are like “Yeah, I am doing exactly that. I shouldn’t do that.” First you need to think and then to write a code. It’s much better.

Tatiana: It’s hard to spot it in yourself. It’s easier when you do mock interviews with others. We will come back to that when we discuss interview preparation. But yeah, that really helps me because it also was a change in psychology.

Tatiana: I was working in science. Academics have a bit of a different mindset. When you come to industry, the mindset is a bit changed. It’s important to have those mock interviews from people with industry. They are not about your potential and how smart you are, it’s about which projects you have done. Projects that are similar to the project that they need to do. Tailoring an interview helps.

Tatiana: It also helps when you see that others struggle as well. Your imposter syndrome will disappear when you start to interview others. All your impressions that everybody is ingenious but not you — that will disappear. You will see “Aha! He is also having problems with that. It’s okay that I have it.”

Hack #5: Attention genius

Tatiana: I call the next hack “attention genius”. You don’t notice everything in your life. You are not capable of noticing everything. You only notice things that you pay attention to. And you can choose what to pay attention to. You can focus on positive things. Or you can focus on negative things. You did not solve the problem today — you can focus on that. Or you solved that problem — you can focus on that. Some people will say “You will not succeed”, “You will fail”. You are going to have haters. It’s important to cut off attention from those people. Whatever is going to put you down, just cut off attention from that. You got a hater. Great. Do not pay attention to him. Pay attention to those who help you, to your mentor, to more important things that bring value to your life. It’s good advice in general. Choose where to focus your attention. (14:52)

Hack #6: Make a team

Tatiana: The next one is make a team. It’s harder to progress just by yourself or to learn by yourself. It’s more beneficial and faster to learn and to grow when you have a team of like-minded people. You can create a team for nearly everything. You can create a team to play kaggle. You can create a team to write a paper, to do the pet project, even to do the course. You can do some courses in a team together. Recently, I had a course about venture capitals in a team of five. That was fun. We could discuss things together. It’s more fun to learn that way. What’s most important — other people have different expertise, and you can absorb that expertise from them. When you are kaggling with great guys, who know a lot about other stuff, you learn from them much faster. (15:56)

Tatiana: The hack is not “waiting for the team to come”. It’s really “make a team”. You want to make a team. You are going to find the people, convince them to play with you on Kaggle or to do pet projects, do that paper. You need to find win-win strategies. You need to find people who will also benefit from that, who want the same course or who need that paper. When you are doing that together, you learn from them, you help them. It’s a win-win strategy. It’s good because it saves your time as well. It gives you a result — you have a paper or you have Kaggle master, but in a shorter time. It’s also critical in maternity leave.

Alexey: How do you find these people? How do you go about convincing them? (17:29)

Tatiana: Sometimes people find me, quite a lot of times they contact me and… (17:37)

Alexey: Now. Because now people know you. But let’s say some time ago. (17:42)

Tatiana: When I just started, I was finding people on Kaggle chat and writing to them. One guy was from Australia. He seemed nice in chat and he was nervous in Kaggle. I could see that he programs better than me. That’s what I needed. But he doesn’t have a clue how to solve the problem. I had a clue because I have read the papers. So, I just wrote to him that I have a PhD in physics — the competition was about astronomy — maybe we can benefit from each other? He replied “yes”. He was in Australia, I was in Ireland and then we started to do that together and we got to the top. Then I invited one guy from Moscow, who is a software engineer with great experience. He was Kaggling for a while, and that is how we got first silver, almost gold. It was good. (17:46)

Alexey: So the hack here is “just to reach out to people”. (18:37)

Tatiana: Reach out to people, exactly. The same with papers. I wanted to write a paper but I did not have material. So, I reached out to a guy from ODS who won a competition. I asked him “Can I please use your code and write a paper based on that?” I wrote to him and he gave me access to everything, even his server for calculations. He helped me with advice as well. (18:43)

Tatiana: Then I was looking for somebody else to help me with writing because I was short on time. I just wrote “I am looking for a co-author who will help me to craft a paper”. That is how Alexander Kalinin wrote to me. He is great at writing papers. He helped me to craft that, to finish that thing. And it got accepted to the CVPR workshops. That is how you make it. You find win-win strategists. You find people who’re also interested in that.

Alexey: Reach out to people, have something to offer. Have a clear offer in the first message. Let’s imagine a situation when I won a Kaggle competition. That never happened but I like imagining this sometimes. If somebody wrote to me saying “Awesome solution. Can I write a paper about this?” It’s very difficult to say “no”, right? (19:39)

Tatiana: Yeah. I also thought this way and it worked out. Also if you don’t know whom to contact, like in kaggle, you see the person, so you can write the person. Or you can also do that in DataTalks.Club. You also can write an open message, like on LinkedIn, “Hello, I am working on that paper, I am looking for co-authors. I need help with this and that. I do calculations, I have all the code set up, but I need somebody for running experiments. I’m looking for a student who can volunteer for that and have a paper.” I can assure you, within 24 hours, you are going to have a lot of students from all places in the world. They will be happy to help you with doing those tasks for being a co-author. (20:05)

Alexey: Yeah. Thanks! So we are done with six hacks? There are six more. (20:53)

Tatiana: We need to hurry up. (20:57)

Alexey: I think we have enough time. (21:01)

Hack #7: Less is more. Forget about perfectionism

Tatiana: Hack #7 is less is more, forget about perfectionism. There’s the Pareto rule that says that 20% of the effort gives you 80% of the result. That’s true. Focus your efforts where you are going to gain that result. But there are those minor things that you need to finish, they can take an enormous amount of time. Maybe you are not good at that. Maybe you don’t know anything about that. Well, outsource it. Delegate it. Find that win-win strategy, and get a team. The best is to focus your energy on those 20% that bring 80% of results. If you are not a front-end developer and you need to do something on front-end, find a front-ender. Find that win-win strategy. Or outsource it, pay for that, delegate. Forget about being perfect. Nobody is perfect. That perfectionism, it takes a lot of extra time. You simply don’t have it. If you are changing careers on maternity leave, you have no time. You have to be careful on how you spend it. (21:04)

Alexey: How do you find this 20%? How do you even know that this 20% is important and this 80% is not? (22:14)

Tatiana: Things can take you enormously long. If you are not good at that or do not know how to make it, it can take you enormously long. If this is not something that is going to contribute to your future career, if this is not the skill set that you require, then don’t spend time on learning that new framework. You are not going to use it anyway. Outsource it. Of course, there are cases when you want to learn these new skills. It's a different story. If you are not learning front-end or you don’t need Java, don’t spend time on all that. When you are not going to do it at work, you are going to forget about it. (22:25)

Alexey: What if you don’t know if it’s something you need or not? I imagine if you are switching careers, if you are learning machine learning, you don’t know if you need to learn this topic. Do you need to learn neural networks for this project? In this case you can ask somebody who is more experienced? (23:14)

Tatiana: Yes. You need mentors, that’s another hack. (23:39)

Alexey: That’s another hack. Okay. (23:43)

Hack #8: Initial creation

Tatiana: Hack #8 is initial creation. When an artist creates a painting, the first creation always happens in the head. The artist visualizes the painting, he can create it in his mind. And then he paints what he created. The same initial creation applies to everything. You want to do some code for a Kaggle competition or for your pet project or your paper — when you have some free time. But you are busy with a kid, you cannot work right now. But your mind is free. (23:45)

Tatiana: You are not working on the computer, but you can save lots of time if you are thinking about what you are going to write or what you are going to code. For instance, you are cooking dinner and you are thinking how you will approach the architecture for that machine learning solution, or what you are going to write in that paper, or what you are going to write in the proposal and so on.

Tatiana: You create it in your head during the day when you are busy with other stuff. You have a lot of time when you create it in your head. You just cannot get your hands on the computer. Then, during the evening, when you finally have a few hours to put your hands on the computer, you have already created everything in your head. This is going to speed up you. You need to do it and you already know what to do.

Alexey: Do you capture these ideas somehow? Or they stay in your head? For me, if I don’t write it down, then they it’s gone the next minute. (25:14)

Tatiana: No. I remember. When I think through something, I remember that. Maybe it's just a question of habit. That’s how I used to write essays at schools. I would think and then go and write. (25:23)

Alexey: So, it’s useful to train your memory to remember this. But even if you are busy with your daily stuff, it doesn’t hurt to have a piece of paper and a pen, so you could quickly write something. (25:43)

Tatiana: You can also have a dictaphone. Sometimes I turn on the microphone and I record things that come to my mind. Things that are useful, things that I don’t want to forget. (25:59)

Tatiana: This initial creation also includes visualization of your goals and of yourself. You want to change your career, it’s also a project. You are creating yourself professionally. So visualize that, visualize yourself in a new job and in your role. Be confident knowing what you are doing, visualize your goals. It helps a lot because you trick your mind in the state that it’s already achieved. There is a practice that’s called Sankalpa.

Tatiana: It’s from Buddhism, and it’s a meditation technique. You visualize your goal in the present, like if this already happened. Your desire, your wish, whatever you want — visualize it. Repeat it three times in the present tense “This has happened”. And then go away. Do your meditation. It works because you trick your mind into thinking that it’s already happened. And you start to behave in a way that it actually happened. For instance, you want to stop smoking. Just tell yourself “I don’t smoke anymore. I don’t smoke. I stopped smoking. I don’t smoke anymore”. You tricked your mind and in future, it’ll be easier for your brain to not think about cigarettes, because you are convincing your brains that you stopped smoking. (27:03)

Hack #9: Find mentors

Tatiana: Hack #9 is to find mentors. A mentor is somebody who already passed the ways that you want to pass. Somebody who’s already in the role that you want to achieve, or who is a person that you want to become. A mentor can speed up your process a lot. You don’t know where to start learning, how to do it? Your mentor can help you. He or she passed that way. Choosing your mentors and finding your mentors comes from the hack number one of changing your social circles. You need to find the social circles where you can find mentors. For me it was ODS. I found a lot of mentors there who were helping me, literally writing me the plan of what I need to learn, what is important to learn. (28:08)

Tatiana: Sometimes I would use the crowd opinion as a mentor. I’d ask a question to everybody and a lot of people will react. Some people think this way, some people think that way, but there is such a thing as common sense. If many people reply to you with the same thing, then that is valid because everybody is mentioning this.

Alexey: You just reached out to people. I think it also goes back to the other hack about making a team. Is this how you found team members and this is how you found mentors? (29:23)

Tatiana: It’s a bit different. You create a team just for short periods — for some project. But for a mentor, ideally you find him or her for a longer time — for years. For that case, developing relationships is different. For making a team, I just say “I am working on that project. I just need your help with this… Are you interested to join?” And if you are looking for a mentor, you don’t write to a stranger “Hey, do you want to be my mentor?” (29:34)

Alexey: Some people do that. (30:11)

Tatiana: It’s not how it works. You communicate with people, you ask questions. And then you establish those relationships. When you establish relationships, the person can become your mentor. It doesn’t happen at a spot, it usually comes gradually. You keep asking people who know more. Sometimes I was asking questions to everyone and some people replied to me in private. Then I started to directly ask people who were replying me often. One mentor who's really great, he wrote me directly “You can ask me questions about this topic if you want”. That is great! (30:13)

Alexey: You need to be active for that. You need to be asking questions. (31:04)

Tatiana: You need to be active, you need to be visible. It’s always personal. Nobody is going to mentor you. If she doesn't like you and if you don’t like your mentor, it is just not going to work. It should be about chemistry. With some people you get along well and with some people you do not. It really helps to have some chemistry and mutual connection. Having a mentor is like having a friend. There has to be a similarity in a way. (31:07)

Hack #10: Say “no”

Tatiana: It took me 30 years to learn the next one. You have to learn how to say “no”. Say “no” to some people and to the projects you do not want to do. People sometimes bombard you with some offerings. When you are changing careers, you have a lack of time. You have to choose what to say “yes” to, what you say “no” to. You don’t want to be a member of that committee. Not now. You have other things to do. It’s very easy to become overwhelmed on maternity leave. (31:42)

Tatiana: They want you to be a committee in a class or in school as well as somewhere else. Some people ask you for other projects. You just have to say “no”. Choose wisely what to focus on. Learn to say “no” to people. Don’t sign up for things that you do not want to do just because you are polite and you don’t want to make that person unhappy. This is not your job to make everybody happy. You have to take care of yourself. So say “no” when you mean “no”.

Alexey: Do you have some rules like “no more than one project at a time”? How do you decide when to say “no”? Let’s say an interesting opportunity comes up, but you already know that you have some other thing on your plate. (32:56)

Tatiana: Yes. You see if you have enough at the moment, but this is interesting, you can say that “Let’s talk in one month, currently I am busy”. Sometimes things that come up are not interesting. You have to be upfront about that. Imagine you started an interview process, you got 2-3 interviews. But you got some other offers, some other options. You see that there is no point in continuing this interview process. If you didn’t like the team, just say “no”. Some people continue it to the end. But I don’t think it’s wise because you are spending your time. (33:15)

Tatiana: Of course. If you have extra time, you can finish all the interviews. Even if you don’t intend to take a job, you can still continue. But if you really like your time and you want to save it, if you after the 2-3 interviews realize that you didn’t like the team, just say “No, I don’t want to continue the process. I got other options”. It’s fine.

Hack #11: Look for failures

Tatiana: Hack #11. Do not be afraid of failures. Look for them. A lot of people don’t try things because they are afraid to fail. This is wrong. Only through failures you learn. The best lessons you get is when you tried something and it didn’t work out. You made some mistakes, you realize those mistakes and that is how you grow. If you don’t have any failures, you try new things and you do not fail, it means that you don’t try hard enough or you are not ambitious enough. You have to be ambitious, get higher goals where you will fail. This is the growth process. If you are not failing for a while, the projects that you are doing are too easy for you. You need to scale up a bit or make more ambitious goals. Then you will fail. Failures bring a lot of lessons and a lot of valuable information. You can build success on failures. (34:25)

Tatiana: But people are afraid of them. For some people, it’s painful. It can be a problem. If a person has a big ego and if the person fails, he is like “Oh my god! I was wrong here. This is so painful.” I don’t feel pain from failures, to be honest. I observe kids. When my kid tries something and it doesn’t work or it fails, he is never disturbed. He is never like “Oh my god! I made a mistake! This is not six, this is five”. They don’t feel pain about it. That’s how children learn. They don’t have that ego and they don’t have pain when they fail. They fail and they learn. You have to have this childish attitude in learning. Be like a child. Focus on learning and appreciate failure. That is how you grow.

Alexey: A failure is when you don’t pass an interview. When you get rejected. Right? (36:27)

Tatiana: Exactly! Some people feel so much pain about that. It’s a bit funny because that is how you learn. You have to be happy that you have failures. (36:33)

Alexey: It’s difficult to be happy about this. You need to convince yourself. I think one of your other hacks was about that. (36:43)

Tatiana: Exactly. I have children, so I observe them. My child says “This is green”. I was like “This is actually blue”. First of all, he doesn’t feel any pain that he made a mistake. He says “Ah, okay, blue. But I don’t like to call it ‘blue’, let’s call it ‘green’”. You can learn from children a lot. They actually fail all the time. This is how they learn to walk. If the child is afraid of failing, he will never learn to run. You have to have that attitude. That is how you learn (36:54)

Hack #12: Take care of yourself

Tatiana: The number 12 is the most important one. Take care of yourself. You have to sleep, you need support. If you need help with your impostor syndrome, get a psychologist. They are great at that. Do meditations. Take care of yourself. You need to dedicate some time during the day to take care of yourself. Because if you do not take care of yourself, no one else will. (37:30)

Tatiana: You have to love yourself. Get your husband to make dinner if you are tired, go to sleep. Ask your parents to help to look after your child and get to sleep if you need it. I am talking about maternity leave. But besides maternity leave, when you are changing careers, you are going to need some help. It’s to ask for that. It’s okay, it’s not selfish to take care of yourself. If you don’t take care of yourself, you will burn out. That is not a good idea. This is important.

Alexey: I am listening to you, and some things are contradicting each other in my head. You have two kids. You need to look after them. They need a lot of attention. Then you have some projects. You are taking part in Kaggle competitions, you are writing papers. You also set ambitious goals. You are looking for failures, so you want to set ambitious goals. At the same time, you also need to sleep, to meditate, to talk to a psychologist to fight the imposter syndrome. And it’s just 24 hours in a day. How do you find time to do all that? (38:39)

Tatiana: For kids, I hired a nanny. You need help, hire a nanny. For cleaning I hired a maid. She is coming and cleaning. When I am busy, we get Chinese takeaway. It’s fine to get Chinese takeaway. My husband helps me in the morning with kids and in the evening as well, because now during the day time I can work. On maternity leave when I was with kids, then I was working after his job. He finishes at five, and he takes the kids and really helps me, and I go learn. (39:36)

Tatiana: It’s also important to cycle a few days a week. That is why I was talking about that first initial creation. When you are cycling, you can think through what you are going to code in the evening or what you are going to write. That helps with your time.

Tatiana: Also, keep focus. When you start doing something — coding, learning — forget about distractions. Shut off Twitter and Facebook. Focus on learning, on your project. You realize that if you have just four hours a day for concentrated work, you start at evening when everybody goes to sleep until midnight, you are going to have a lot of work done. With focused work, you can accomplish the same as people during their full-time working day, if you just four hours straight into that.

Alexey: So, you need to get into the flow. (41:15)

Tatiana: You need to get into the flow. It also needs practice. I am an artist. For me it’s easy. When I start painting, if a plane falls next to me, I may not notice it. I can be that much into creation. I can do the same in coding and writing and learning. Some people cannot do it. But I believe you can practice it, like artists do. When you are painting, everything disappears, you are just there. It can be hours. Suddenly three hours passed, wow! I made a painting. (41:18)

Tatiana: I had time from five till midnight, about seven hours. Seven hours of really focused work. You get one hour more for cycling, for sports. But, actually you don’t need that much time. Especially if during my day time, when I was cooking and taking care of the baby, going for a walk, I think through what I am going to create, that is enough. I don’t work more hours but work deeper during the hours that I have. I think that’s another hack that I should have included actually.

Kaggle vs internships and pet projects

Alexey: Yeah. For your next twitter thread. We have a couple of questions. Some of these questions overlap with what we wanted to talk about. A question from Alexandra. What does Tanya think about Kaggle versus Omdena. Omdena is a platform similar to data science for social good. They have some initiatives. If you don’t have any experience, you go there, you have some mentors. You work for a couple of months on a certain project, you get help, you really work with somebody on a specific project. Since you don’t know what it is, I am wondering if you tried any similar initiatives like this one. (42:34)

Tatiana: I tried that internship project which actually was for social good. It’s more like a pet project. I cannot compare to that platform. But I can talk about the pros and cons for Kaggle, and she can decide for herself. (43:45)

Tatiana: Kaggle can teach you a couple of things. Kaggle is great for learning machine learning. It’s good for practicing machine learning on relatively small and clean data sets, with quite good distribution. It’s like a simplified environment when you already have the data, the tasks and the metrics are chosen. You just try different machine learning approaches. It can help you greatly in starting off in machine learning. There are notebooks where you can read the code of others and learn from that.

Tatiana: It helps you to learn exploratory data analysis. It helps you to try different machine learning approaches quickly. You can try different state-of-the-art approaches from papers also quickly — some people will publish it in notebooks, some you will adjust and implement. You also learn how to create validation properly. A lot of people in the beginning make mistakes in validation, and then they fall on the private leaderboard. It’s a great community where you can find teammates and potentially mentors.

Tatiana: What Kaggle does not teach you is all the rest. Kaggle doesn’t teach you how to transform business problems into machine learning problems. It doesn’t teach you how to communicate with different stakeholders, realize their needs, choose metrics that reflect business, which data do you need, how much data do you need, where to find the data, how to collect the data, how to label the data.

Tatiana: Kaggle doesn’t teach you how to deploy that model, things like CI/CD, DevOps, Deployment, Monitoring and Maintenance. If your data drifts, what are you going to do? How are you going to monitor that? It doesn’t cover any of these things. In realify, if you take a job of a machine learning / deep learning engineer, Kaggle covers around 10%-15%, when you just play with different models.

Alexey: And your internship experience — did it cover anything outside of just modeling? (46:40)

Tatiana: It’s important to have pet projects — they cover more. You need to collect the data, decide what data you need, how much data you need, label it. If you already have a data set, that’s a different thing. For an internship project, you usually don’t choose the topic that you are going to work on. It’s better if you need to choose it or you can talk to different stakeholders, understand the business needs and propose a project yourself. Pet projects rarely go to production. The difference [of an internship] with pet projects — it covers a bit more things. There is an opportunity to have a mentor or supervisor — that is great. That is what I had and that is a big boost, because he can help you to start, introduce good practices and help you to avoid some mistakes. This is going to save you time. (46:42)

Tatiana: So my advice is to do both: some pet projects and Kaggle. For the pet project, instead of doing something similar to Kaggle, it’s better to do that end-to-end. Find a problem that you want to solve with machine learning, find a person who can benefit from that, convince that person. Then you will deploy it, you find the data, you collect the data, you label it. Do that all end-to-end — when you have zero data, no metrics, nothing, up to production, when it goes to some website or to the server. At least pack it in Docker.

Alexey: I guess it makes sense to start with Kaggle? If you just became interested in machine learning, you are lost. You will not be able to translate the business problem into machine learning terms, you will not be able to select a metric — you have no idea about these things. I guess Kaggle makes sense for a start. (48:49)

Tatiana: I was changing career after science. For me it was quite easy to think about business. after a career in science and after PhD, you think that. But yeah maybe Kaggle is better for starting. (49:14)

Resources for learning machine learning

Alexey: And what resources for learning machine learning can you recommend? In addition to Kaggle. Kaggle is great. I’m not sure if we call it “a resource” but it’s definitely a good website to check. (49:29)

Tatiana: It’s definitely a learning resource. Before going to Kaggle, I recommend familiarize yourself with basics. I was doing courses on Coursera. I really love the one from Stanford — machine learning and deep learning by Andrew NG. Those courses are quite famous. I really recommend everybody to first grasp those concepts before you go on Kaggle. It will be easier to learn. Because I was changing my programming from Matlab to Python, I also did a course on audacity on Python. (49:43)

Alexey: Do you remember the name of the course? (50:25)

Tatiana: I think it was “Python for machine learning” or “Python for data science”. They were covering things like Pandas, NumPy and all those useful packages. You will not know all packages in Python, it’s impossible but those that are mostly used. (50:27)

Alexey: One month ago we had a guest on this podcast who recommended the same course. (50:45)

Tatiana: Next, some people take courses on SQL. It’s important if you want to work as a data analyst or you are going to work a lot with SQL databases. It’s straightforward to learn SQL. It’s also nice to do it systematically. There is a two weeks course that you can take. For interviews, you need to do some coding exercises on time. If you don’t have a computer science degree, or you had quite a lot of time ago, it’s nice to check the course 6.006 from MIT about data structure and algorithms. It covers Cormen’s famous book on data structure and algorithms. Then I started to do some courses for fun like a specialization on self-driving, but this is not needed if you don’t really want it. (50:55)

Tatiana: Then Kaggle. It helps you to practice. Then there are a few books that can help you to understand how to make end-to-end machine learning pipelines. You also need to learn system design — there is a course “Grokking system design interviews”. It’s more for interviews, but it helps you to understand how to build large scalable systems. There’s also “Grokking machine learning design”, but to be honest I didn’t like it that much. It’s quite shallow, it’s not really going deep enough in my opinion.

Alexey: “Grokking system design” — this is something you take towards the end of your studies, when you already feel comfortable with other things. (52:46)

Tatiana: They are more or less in chronological order. (53:00)

Alexey: Yes. I understand. (53:03)

Tatiana: Python, machine learning, deep learning, Kaggle, the MIT course on data structure and algorithm, system design, machine learning design. Then if you want to learn how to put it in production, you can learn Flask or you can learn fastAPI. There’s a course on JavaScript — depending on what you want to learn. But that’s after you mastered machine learning. Finish it and you will understand what is the next step. (53:05)

Starting with Kaggle

Alexey: We have a question. What level of python programming one should have before starting with Kaggle? (53:40)

Tatiana: Zero. I know it because that is what I was doing. It’s okay. I got Kaggle masters. You will make it. All you need is to start. That’s how you learn it. (53:49)

Alexey: Yeah. If you compare Python with other languages like Java or JavaScript or even R, in my perception, they are more complex than Python. Python is quite simple in a sense that there is not much there in terms of syntax. It’s not overloaded with syntax unlike other programming languages. (54:08)

Tatiana: Yeah. I learned Java some time ago. I can also compare. I did not like Java, I liked Python. (54:35)

Improving focus

Alexey: Do you have any tips on how to improve your focus? (54:44)

Tatiana: Some people practice meditation for that. There are different types of meditations. In some, you focus just on breathing, in some you focus on your body sensations. There are analytical meditations when you just about some problem and focus your attention on that problem. I think analytical meditation is very useful for learning how to think about some technical problems without distraction. I had this skill from school. We used to have school Olympiads in physics and math. You have four hours and very challenging problems. You sit there for four hours, trying to get that medal in the school Olympics. That really helps you to focus straight for four hours on those physics problems. (55:02)

Alexey: I still cannot imagine it. I have just one kid. And it’s so easy to lose focus when there is a kid. When he starts playing, he is very loud. Even if I am in a different room, just hearing that something is going on, maybe furniture falling is failing or something like this. It’s very difficult to stay focused. I don’t know how you managed to do that. (55:55)

Tatiana: I can cut that off. As I told you, I can paint and not hear anything. Maybe I have a natural tendency for that, but you can also focus with Sankalpa. You only focus on your hands and legs. If somebody is happening outside, you make a mental note, “I notice that sound”. And you are coming back to your practice. With mediation and learning to focus, It’s not going to happen in a week or two. Don’t trust anyone who will tell you it will. If you do that for a year, you will see the result. You can train your brain to focus. It just takes time. (56:27)

Tatiana: It’s not easy, but it’s worth it. If you can control your focus, your attention and your mind, you can control everything. There is a saying that “the person who control himself controls the world”. (57:09)

Alexey: There is no silver bullet. It’s not like you take a pill and everything is great. You need to work. (57:27)

Tatiana: There are some magic pills. I have not tried. I know in the US, some students take it when they prepare for exams and during the exam. That helps them to focus but I do not recommend this way. I never tried but I know that exists. (57:35)

Astroinformatics

Alexey: We have a question from Saurav. What is your view on Astroinformatics? Do you know anything about this? (57:56)

Tatiana: I was doing just some projects in astronomy. I don’t know much. It was quite a long time ago. (58:07)

Alexey: You mostly focused on lasers in your research. The question is if machine learning is gaining momentum in this field. (58:15)

Tatiana: I know about astronomy. We published a paper about machine learning in astronomy. Before COVID, they planned to open a new telescope in Chile. It’s much larger than Hubble. They are going to collect a lot of data. The task challenge was to classify signals from different objects — can be stars or other objects that you get from images from that telescope. Indeed they started to look into machine learning approaches. There are papers about that as well, but that’s pretty much all that I know. A lot of fields are now trying to apply machine learning — in physics as well. However, it’s a question, how successful it will be. (58:32)

How background in Physics is helpful for transitioning

Alexey: Another question. For some people math might be a serious roadblock. I think for you it wasn’t because you had experience in physics. You also took part in competitions when you were at school. For you math was not the problem? (59:29)

Tatiana: Well, for me it is just a school class. (59:46)

Alexey: What part of your math background helped you the most for your transition to machine learning? (59:48)

Tatiana: Well, the first year of university in physics. Physics uses much more complicated math than machine learning. For me it’s trivial math compared to what we do in physics. (59:55)

Alexey: I remember that differential equations that we studied in the second course were quite difficult. None of this is needed in machine learning. (1:00:11)

Tatiana: Everything is linear like almost. (1:00:24)

Alexey: Maybe there are some applications of differential equations in time series. But in general, in industry, it’s quite uncommon to see advanced math. (1:00:25)

Tatiana: If your industry is aerodynamics. (1:00:41)

Alexey: Okay, that’s a different story. (1:00:45)

Tatiana: Just not machine learning yet. But I think machine learning can benefit from using more sophisticated math as well. It just did not come there yet. (1:00:47)

Alexey: Now we have deep learning. You just throw more data, more hardware at it. And it just magically works. Nobody knows how and why. (1:00:55)

Tatiana: I can tell you what I was doing in lasers. I would take a system of differential equations which also has some parameters. All parameters have some physical meaning, but some of them are not trivial to measure. What are you going to do? You are going to try to twist those parameters to feed the experiment. You are doing the same stuff like in machine learning, but with a system of differential equations — by twisting some parameters. (1:01:04)

Alexey: Another question. With this background, what did you need to focus on? What would you recommend to people with a background similar to yours? Math was not a problem. What was a problem for you? (1:01:31)

Tatiana: Not really a problem but I was not programming on Python before. I was programming on Pascal and Java. For me it was learning python. I never had computer science degrees, so I had to learn all those algorithms and data structures. That’s why I really recommend this MIT course. I am still practicing on LeetCode to get a grip of doing that coding questions. If you have a computer science degree, you probably learn a lot of those algorithms and practice them. I never had that practice, so it takes me time to get good at that. (1:01:50)

Leaving academia

Alexey: We have a question from Mike. What was the main reason for you to leave the academy? (1:02:33)

Tatiana: I had two maternity leave. If you are not active for five years, you can forget about your career. I decided that I want to keep learning on maternity leave and keep growing and developing myself. That was a convenient way to do it because you could do it from home. You do not need anything. One of the resources that I did not mention, you can get some access to free GPUs, using credits in Amazon cloud and Google cloud. They occasionally give away credits on Kaggle competitions. Also, if you register an account on Google, you can get $300 free credits. You can ask for a GPU there and use it. (1:02:40)

Tatiana: So I was completely independent. I could do that alone from home. All I needed was a laptop. Two maternity leaves was the main trigger for that.

Alexey: I have no idea how it happens with lasers, but you need some special equipment and a lab where you experiment, right? (1:03:38)

Tatiana: Yeah, you need a lab. In Nano photonics, it’s usually a “class 10 clean room” — you dress up like doctors on coronavirus in China: with a mask and glasses — you have all this. They should avoid contamination. On top of that, there is a lot of chemistry in making lasers. Pregnant women are not allowed there. I learned that the hard way when I got to know that I cannot do that when I got pregnant. If there are ladies who are watching this, please think about pregnancies possibilities and choose a career which is okay with that. (1:03:49)

Preparing for interviews

Alexey: We should be wrapping up. Do you have any last tips before we finish? (1:04:34)

Tatiana: I will give some tips for preparing for interviews. The best tip I can give is to practice. Do mock interviews. Use pramp. But the best resource is your network. Talk to people who are looking for a job and interview each other. It’s nice to see it from the other side of the table. Another resource, for Russian speaking communities, we can send a link to some telegram channels. There people meet to do mock interviews on system design, on coding, and on machine learning design. (1:04:40)

Tatiana: When you have an interview soon, start practicing Leetcode. They ask a lot of coding questions from Leetcode. It’s hard to solve them in a short time if you haven’t practiced before. It requires a couple of months of practice. It’s better to start it in advance. When you are closer to the interview, get a course on Grokking system design interview. Unfortunately there are no good courses yet on machine learning design. Those are the main things.

Tatiana: Even if you don’t want to start a job immediately or you feel you are not ready, you should be ready to send your CV. It’s okay if you are not completely ready before you go to an interview. You try and fail. Before you fail around 5-10 interviews, you probably will not pass them. It requires practice to gain that skill of understanding what the person wants to hear and what signals you want to show and how you are going to show them.

Alexey: That was also hack #11, if my notes are correct — “look for failures”. (1:07:12)

Tatiana: Apply to Google tomorrow. It’s okay if you fail. You are going to learn something along the way. (1:07:20)

Alexey: Okay. The last tip is to apply to Google. (1:07:26)

Tatiana: Well, not necessarily Google. But if you are applying to big tech companies and you fail interviews with them, you actually learn as well. It’s a good learning experience. (1:07:28)

Alexey: Where can people find you? (1:07:38)

Tatiana: I have LinkedIn and what else. (1:07:41)

Alexey: Twitter, Instagram… (1:07:47)

Tatiana: I am not Instagram. I am on Facebook but I do not check it. I use LinkedIn and Twitter. And I will be answering some questions in DataTalks.Club next week. (1:07:49)

Alexey: Thanks a lot for sharing your hacks with us, for sharing your experience, your knowledge. I am sure it will be helpful for many people. Thanks everybody for joining us today and watching our chat today with Tatiana. I wish everyone a great weekend and see you next week. (1:08:10)

Subscribe to our weekly newsletter and join our Slack.
We'll keep you informed about our events, articles, courses, and everything else happening in the Club.

DataTalks.Club. Hosted on GitHub Pages. We use cookies.