Kirill Eremenko: This is episode number 297 with Data Scientist Ayodele Odubela.
Kirill Eremenko: Welcome to the SuperDataScience podcast. My name is Kirill Eremenko, Data Science Coach and Lifestyle Entrepreneur. And each week we bring you inspiring people and ideas to help you build your successful career in data science. Thanks for being here today and now let’s make the complex simple.
Kirill Eremenko: This episode is brought to you by our very own data science conference, DataScienceGO 2019. There are plenty of data science conferences out there. DataScienceGO is not your ordinary data science event. This is a conference dedicated to career advancement. We have three days of immersive talks, panels and training sessions designed to teach, inspire, and guide you.
Kirill Eremenko: This three separate career tracks involves, so whether you’re a beginner, a practitioner, or a manager, you can find a career track for you and select the right talks to advance your career. We’re expecting 40 speakers, that’s four zero, 40 speakers to join us for DataScienceGO 2019. And just to give you a taste of what to expect, here are some of the speakers that we had in the previous years. Creator of Makeover Monday, Andy Kriebel; AI Thought Leader, Ben Taylor; Data Science Influencer, Randy Lao; Data Science Mentor, Kristen Kehrer; Founder of Visual Cinnamon, Nadieh Bremer; Technology Futurist, Pablos Holman; and many, many more. This year we will have over 800 attendees from beginners to data scientists to managers and leaders.
Kirill Eremenko: So there’ll be plenty of networking opportunities with our attendees and speakers and you don’t want to miss out on that. That’s the best way to grow your data science network and grow your career. And as a bonus there will be a track for executives. So if you’re executive listening to this, check this out. Last year at DataScienceGO X, which is our special track for executives, we had key business decision makers from Ellie Mae, Levi Strauss, Dell, Red Bull, and more. So whether you’re a beginner, practitioner, manager or executive, DataScienceGO is for you. DataScienceGO is happening on the 27th, 28th, 29th of September, 2019 in San Diego. Don’t miss out. You can get your tickets at www.datasciencego.com.
Kirill Eremenko: I would personally love to see you there, network with you and help inspire your career or progress your business into the space of data science. Once again, the website is www.datasciencego.com and I’ll see you there.
Kirill Eremenko: Welcome back to the SuperDataScience podcast. Ladies and gentlemen, what an episode. What an episode I have prepared for you today. Ayodele Odubela is one of our speakers for DataScienceGO this year in San Diego and I literally just got off the phone with her from recording this podcast and this episode is going to blow your mind. Ayodele came into data science just over two years ago from a nontechnical background and the amount of success, the amount of projects that she’s done, the amount of things that she’s learned and already given back to the community is going to be super inspiring, for me it was super inspiring to hear about.
Kirill Eremenko: You’re going to hear about things like how and why Ayodele chose to do a Masters in Data Science and, a full time masters for two years and how she’s supplemented that with online education and why. Finding messy data on purpose in order to learn how to deal with messy data. We’ll talk about self-discovery, fortitude and passion. You’ll hear about some of the projects that Ayodele has worked on such as using SVM to support vector machines for detecting poisonous versus edible mushrooms, using random forests and decision trees for ranking wines based on the chemical contents, using the Naive Bayes to detect spam and real world project that she’s actually worked on, written a conference paper on, bullet stopping flying drones.
Kirill Eremenko: Yes you heard that right. Bullet stopping flying drones, and you will find out what role machine learning played in that. What Ayodele did with that and how they’re going to be applied in society once they get rolled out. Also you will learn how she got one of her data science jobs through Twitter and how you can replicate the same success, how you can expose yourself on different platforms to get hired basically. In fact, you will also learn that they’re currently hiring at MINDBODY, the company where she works and you’ll learn more details about that role. And many, many more things from soft skills to her presentation at DataScienceGO and lots and lots of other things. So a very inspiring podcast. I can’t wait for you to check it out. So without further ado, I bring to you data scientist, Ayodele Odubela.
Kirill Eremenko: Welcome back to this SuperDataScience podcast. Today I’ve got a very special guest joining us for the show. Ayodele Odubela. Ayodele how are you going?
Ayodele Odubela: I’m doing well. How are you Kirill?
Kirill Eremenko: I’m very well too and very excited as well to have you on the show. You are local to San Diego, which I was very surprised to learn just now.
Ayodele Odubela: Yes. So I’ve actually only been in San Diego the past six months, but it’s been a really nice change from living in pretty cold Denver, so it’s beautiful.
Kirill Eremenko: Wow, that’s good. I love San Diego. It’s the weather, it’s amazing there all the time. That’s so nice. How long did you spend in Denver?
Ayodele Odubela: I was there for three years, so I was just starting to get used to the cold and then I got spoiled a little bit.
Kirill Eremenko: What happened? Why did you move?
Ayodele Odubela: I actually landed in a new position, so I started with MINDBODY in March and that’s when we moved out to San Diego.
Kirill Eremenko: Oh, congrats that’s really exciting. I must say your LinkedIn is super interesting with the different projects and different roles that you’ve been in and you currently are in. When I read it, I was really pumped to see what will come out of this podcast. So it’s going to be really fun I think.
Ayodele Odubela: Yeah, I think so too.
Kirill Eremenko: To get started, tell me a bit more about this … We chatted about this just now before the podcast, but I’d love to learn more. So you have a podcast of your own and it’s about hockey. Out of all things, you chose to do a podcast on hockey. Tell us about that. How did you become a hockey fan?
Ayodele Odubela: Yeah, so I actually got dragged to a hockey game my very first time without really knowing anything about it previously and of all places, I actually got into it in Texas. And to a minor league game and fell in love once I figured out where the puck was moving. It was just so much faster sport than football, which I had gotten really accustomed to. So after a couple of years of being really interested in this sport, I started talking to a lot of other people, meeting other hockey fans and my boyfriend actually is the cohost of the podcast with me and we end up talking a little bit stats, a little bit about trades and contracts. And a lot of the other fun drama that’s really involved in national hockey in the US. So it came out of nowhere, but we definitely tend to just talk about, make a couple of jokes about [inaudible 00:08:02], analyze a little bit about what’s going on.
Kirill Eremenko: That’s so cool. That’s such an exciting thing. So what’s the podcast called in case we have some hockey fans in the audience as well?
Ayodele Odubela: Yeah, it’s called the Offensive Zone podcast.
Kirill Eremenko: The Offensive Zone podcast. Okay. Very cool. And how long have you been doing that for?
Ayodele Odubela: That’s actually about two years. We are about to start our third season, which is very crazy to me because we started it on a whim.
Kirill Eremenko: Wow, that’s cool. Exciting time. Well congrats. Hope that keeps going really well. That sounds like a cool passion project and completely unrelated to data science then.
Ayodele Odubela: Yup. I have done a couple of hockey projects doing some goal prediction. I’m really interested in looking at computer vision for trying to predict goals. Maybe the previous three seconds of the video or however long trying to get an understanding there of what leads up to a goal or what might lead up to a miss.
Kirill Eremenko: Okay. Very interesting project. Maybe eventually you’ll be able to predict the game in advance, the scores or something like that. That’d be valuable. I guess. I don’t know much about hockey, but it sounds like a really cool area to be in. Okay. So let’s talk about data science then. You mentioned that you moved from a nontechnical background into data science, right? It sounds like you love these things where there’s some uncertainty. Oh, hockey. Okay. I’ll go into that. Oh, data science. I’ll go into that. So what’s the story there? What were you in and how and why did you get into data science?
Ayodele Odubela: Yeah, so I was actually working in marketing and specifically social media marketing. In my undergrad I studied media professional communications and was working for a marketing agency when they had a little bit of an opening on the PPC side, so understanding paid media campaigns, that introduced me to the world of impressions and click through rates just as digital media was really starting to rise. And a couple of years after doing that, I ended up working for an app company, and I was putting on a lot of AB testing, so their push notifications, this was one we had just found out that using emojis would get people to click into those. I was doing AB testing on in-app notifications as well.
Ayodele Odubela: And as soon as that startup really started to go under, I actually decided to go back to school for my Masters in Data Science. This was 2016 when that was just starting to get hot and I started to notice an overlap in my skills and what I was really starting to do at work and how my digital role seemed to be more about analytics. The more I wanted to progress in the roles. They were looking for people who are able to analyze the results more so than create the content. So that’s when I switched into data science.
Kirill Eremenko: Okay. Well, so you looked around for roles and saw where the market is moving and decided to follow the market.
Ayodele Odubela: Mm-hmm (affirmative). Pretty much.
Kirill Eremenko: Okay. That’s very cool. You said you started your data science in 2016, is that right?
Ayodele Odubela: Mm-hmm (affirmative)
Kirill Eremenko: And also you told me you finished it like just this year, is that correct?
Ayodele Odubela: I finished it last December.
Kirill Eremenko: Last year, last December. So what is it like a two years master, oh no? Yeah, two year masters?
Ayodele Odubela: Two year program.
Kirill Eremenko: Okay. Full time?
Ayodele Odubela: Yes. So I probably would not recommend this to other people, but I was working full time and going to school full time in data science and it was tough but I survived.
Kirill Eremenko: Wow. Yeah, that’s the drive to be successful to learn new things. That’s really cool. Usually people take part time if they have work. What made you take the decision to do full time data science?
Ayodele Odubela: I wanted to get it done quick, so at least with a degree, but with work, I had the opportunity to work for a startup and I was putting in about 60 hours a week, but I felt like I had a really big impact. I was the only person that was data knowledgeable on the team. So it was on my shoulders a lot for fulfilling requests, doing a lot of the predictive modeling. Anything data related I felt was on my shoulders. So I had the benefit of having a lot of impact and I sacrificed sleep.
Kirill Eremenko: Okay. Yeah. Well, all right. And so was it worth it? Was a Masters in Data Science … This is a really good question actually. It’s quite a controversial question now because there’s so much online you can learn, so many things you can learn online. So what were the advantages of doing an in person … You said you did it at Regis University in San Diego, what were the advantages of doing it in person?
Ayodele Odubela: Yeah, so Regis is actually a really small school in Denver, Colorado and so some of my classes were in person, but some I actually was able to take online and just work on after work. I feel the biggest advantage for me was having these conversations in person with people about data science. There’s a lot of really difficult to understand concepts that when you hear it described in multiple ways from multiple people and talking through those problems that you run into, I found that to be the most helpful. I also think for me getting a Masters degree was really that foot in the door in a lot of companies. I think that is definitely in part because I have a nontechnical background. So if someone were to look at my resume and just look at a nontechnical bachelor’s degree and a marketing and analyst role, they may not necessarily think I’m qualified for a data science position. So I think that really helped get me to the next level. And when I was applying to jobs, I think that definitely made a big difference.
Kirill Eremenko: Okay. Just out of curiosity, what is your background in terms of bachelor’s?
Ayodele Odubela: Oh, it’s in digital media and communications. So it’s interesting actually. In my undergrad I was computer science for a year and a half, and I was not in love with the program, and I ended up switching over to a more general digital media role, but ended up switching over to a more general digital media and communications degree, but that included some courses in web design, and critical media theory, and a lot of things that I actually found really helpful to my role now, and since I work with a lot of marketing and advertising data and that’s in my wheelhouse previously. My undergrad actually was fairly important to my role and that communications aspect, and public speaking is actually a large part of presenting at my body right now.
Kirill Eremenko: Well those soft skills in data science, right?
Ayodele Odubela: Absolutely.
Kirill Eremenko: That’s super important. You know what this is, I’ve never encountered this before. I love this about you, that you are so open to just like, take, pick your things up and change and change and change. Starting from you said it was an IT degree at the very start, right?
Ayodele Odubela: It was actually just kind of general studies. So it was more of that communications in digital media. And I didn’t like the computer science things.
Kirill Eremenko: So you basically got out of computer science, went into digital media and then digital media communications doing that in those areas and then back into data science, which a lot of this actually really related to computer science. So like I like this about you that you can very easily, very adaptable, very agile in the sense of how you think about your career and your future and psychologically, how does that feel? Does it stress you out or does it like on the flip side, does it liberate you in some way?
Ayodele Odubela: I think there’s definitely a little bit of both. I have to give some credit to my past and having to adapt when moving from city to cities though. Growing up I was what I like to call a military brat. I went to a couple different schools for middle school picking up and moving was just part of life. So I think I learned to be really adaptable on that side. But even as far as my career, I think that’s definitely been somewhat due to necessity. When I was working in marketing, I was at a couple startups that ended up just running out of runway and having to try and find another role with short term notice, not really any severance package, you really start to think, where can I best leverage my skills that will put me in a place that’s a little bit more financially stable.
Kirill Eremenko: Yeah. Well, fantastic. No, that’s a great example. I think for the people who might be a little bit hesitant, they might not completely like what they’re doing and things like that. Just being open and to this uncertainty and seeing where life takes you.
Ayodele Odubela: Absolutely.
Kirill Eremenko: Cool. Okay. So, and with this Masters in Data Science, another thing you mentioned before the podcast, which I’d love to touch on, is you supplemented that with online education. Tell us a bit about that. So how did you do that and like what … How does that put you ahead or why? What was lacking in the actual data science degree that you were doing?
Ayodele Odubela: I think one of the things that was lacking, my program was very project based. So each week’s homework essentially would be one part of a larger project that was built over the course of the semester. I think what was lacking was that real connection to what real data might look like. So we had a lot of very packaged, pretty pre-cleaned data to work with. And one of the things when I started out, I actually took the like Data Science A to Z course and I really understood data. That wasn’t something that was really part of the program that even now, obviously those surveys, that data scientists spend so much time cleaning data but there’s little things I run into like Excel formatting and things that aren’t necessarily taught to you, but can be such a time waster when you get to your role if you don’t have experience working with it.
Ayodele Odubela: So that was definitely one of the reasons that I wanted to supplement my degree just working with things that are messy and finding Kaggle datasets that were imperfect and running into errors and trying to work through those.
Kirill Eremenko: That’s really cool. In the Data Science A to Z course, you’re talking about section three, right? Where we talk specifically about cleaning the data. Did you, like that was, the way I put that together was I took everything, all my experience I had back from Deloitte in terms of, alright, what messy data have I ever encountered and how … let’s make it as difficult as possible. So this is the biggest section in the course. And I knew I’m going to make it brutal. I think actually I mentioned that at the start of the section. Did you find it manageable? How did you get through that part?
Ayodele Odubela: I found it manageable over time but I really enjoyed the, I loved that it was difficult because I wish something that people would have told me more before even getting into data science was how hard it really is. And when I say that I don’t necessarily just mean technically, but explaining to stake holders that correlation isn’t causation, and you can’t say arise in this one aspect is related to one thing. I think early on, it’s easy to get disillusioned and say I’m going to have impacts, and I’m going to predict these really cool things without understanding how hard working with the actual data is.
Kirill Eremenko: Yeah, no, totally agree. That’s really cool. Okay. So finding messy data and learning how to work with messy data because projects can be pre-packaged too neatly in a real master’s. Okay. Any other reasons?
Ayodele Odubela: I just wanted to learn as much as I could. I felt like my program kind of left out some of the really interesting things. We didn’t go very deep in natural language processing and that was one that I really enjoyed just researching on my own and doing some of these supplemental classes. I really enjoy working with text data.
Kirill Eremenko: Okay. Really cool. Well, that’s yeah, text data is quite a powerful thing. It’s great that you’re finding these things that you do. What I keep like reverting to in my, like thinking about this is that all of this happened in what, two years that you’ve been in data science for two years. Is that right? Or two and a half?
Ayodele Odubela: [inaudible 00:21:56]
Kirill Eremenko: That’s crazy. Like you’ve already accomplished so much in such a short span of time. Can anybody do this? Is this available to anyone to become a data scientist from a nontechnical background in two and a half years and achieve this level of success?
Ayodele Odubela: I think so. So I don’t want to say, oh, there’s nothing special about me, but I think it was a little bit of just hard work and a genuine interest. So data science may not be the right thing for everyone. I thought that software engineering was going to be my thing and I realized it was not. I like the analysis and I don’t so much like the debugging and building a product and really thinking about things on that aspect. I think data science was the right intersection of what I was doing.
Ayodele Odubela: I was able to use some of my marketing experience and have a company take a chance on me because I had marketing experience, and I was analytical enough. But I don’t think it’s something that can’t be learned in two and a half years. For someone to get a job in this field I really think it takes that just highlighting where you’re unique and where your passions actually lie. And I think it’s that self-discovery, I don’t want to say, oh, if you’re not built for this then you shouldn’t do it. Even realizing now in my job, I’m a people pleaser and it’s difficult to be in data science and try to please everyone. So it just takes a little bit of fortitude and some passion, I think.
Kirill Eremenko: Yes, that’s really good. Fortitude and passion, that should be in the title of this podcast or something in the notes. Fortitude and passion. Okay. Very cool. So let’s talk a bit about some sample projects. You have some very exciting things mentioned on LinkedIn. Is that okay if we jump to that now?
Ayodele Odubela: Yeah, absolutely.
Kirill Eremenko: Okay. So first one is this one I found really interesting, I got to know how you did this. You said you built an SVM, support vector machine, that resulted in a 97% accuracy in determining poisonous and edible mushrooms. Why and how, the questions I have, why this project and how did you do it?
Ayodele Odubela: Absolutely. So that project actually came about, that was one from my master’s degree. So we were given a dataset. One of our instructors was really, really interested in hiking and had an odd fascination with hunter gatherer lifestyles. So we were looking up different kinds of edible plants, edible flowers. I landed on edible mushrooms and essentially the data that I was able to find had a couple of the different aspects of the mushrooms, the size, the chemical components as well as a couple other measurements. How many, I forget the name of it, the part on the bottom of the cap of the mushroom, and essentially what I wanted to end up productionizing was a tool that would use all of these components from our past knowledge and combine that in some way with the computer vision project that will let you scan a mushroom and tell you if it’s edible or not.
Ayodele Odubela: So this was that first step in that project not really going the computer vision route yet, but trying to understand what our likelihood is of getting sick if it’s happening to eat a mushroom on our hike. So well what I found really interesting about this one is that it didn’t take a lot of tuning for those SVM to actually perform pretty well. And I had a relatively balanced dataset, so I didn’t have to deal with a lot of those issues. But it was cool because this project was the first to really make me think about our evaluation metrics and accuracy obviously being the most popular, despite the fact that it’s a very accurate model. I don’t know if I would trust it. It’s something [inaudible 00:26:34] talking about me [inaudible 00:26:35] and I’m like, “I’m just going to pass on all.” But I can also see how that same methodology is applied to medicine and predicting disease.
Ayodele Odubela: And it really got me starting to think about the ethical aspects of data science and how we are also in the control of choosing what metrics we want to use, which is probably uncommon for most people in data science but if you want to use accuracy or precision or recall, that’s in the hands of the protect practitioners. So it got me thinking early on about the implications of that.
Kirill Eremenko: And it also puts into perspective that 97% might sound good for a business project, but when it comes to, is this mushroom poisonous or not, I wouldn’t like, 3% chance of you’re getting super sick or even having a lethal outcome doesn’t sound appealing to me at all. I would require a 99.97% accuracy at least. Very cool. Sorry?
Ayodele Odubela: Oh, I’m with you on that. I have not trust that 3% chance.
Kirill Eremenko: Very interesting. Okay. All right, so next one. This one I’m a wine fan, not fanatic. I like wine and this one sounds really exciting. So you created an algorithm to rank wine quality based on its chemical compounds using random forest. Tell us a bit about that.
Ayodele Odubela: Yeah, this is definitely one of my favorite projects. I am also a [wino 00:28:20]. So I had never really given a lot of thought to what’s in wine. I’d never given a lot of thought to the chemical components. And so this project came up just out of interest. I was looking at, I think at the University of California Irvine’s data repository and they have this really nice wine quality dataset. And essentially red wines are ranked, I know this sounds weird, but from three to eight based on how good they are. You look at the sulfates, citric acid I think is another one of the features, the alcohol contents another one of the features. So I just wanted to understand, what makes a wine actually good? We hear a lot about being a wine connoisseur some time may be fake or there’s a lot of controversy I think around that.
Ayodele Odubela: So it was cool to quantify it and so I ended up creating these random forests, but I wanted to have a better understanding of why certain decisions were made. So I ended up looking at just a single tree. And within one tree, what I was able to find that pretty much the root node or the top definition of what makes the wine good or bad is actually the alcohol content. So this one surprised me because I didn’t think it would matter as much in wine maybe as compared to like hard liquor or something. But essentially if the alcohol content is over 6.5, it ends up being on that seven or eight scale versus being in one of the lower sections for quality. So after wine, I think it ended up being folic acid and sulfates.
Ayodele Odubela: So a lot of chemistry things I haven’t really thought about in a while, if I get the chance to see the alcohol percentage in a wine bottle now, I’ll try and compare it to one that I like and I’ll choose the higher one.
Kirill Eremenko: Okay. That’s interesting. Tell me this like normally I thought normally wine is over 12% anyways, like 12.5, 14 or is this some special wines that go below 6%?
Ayodele Odubela: So I think there are enough that go below 6% that it was actually that ended up being the root node. So that leads me to believe in the ones that are higher than that it’s probably the folic acid and the sulfates that are pretty big features.
Kirill Eremenko: Okay. Wow. Okay, cool projects. So you use that in your daily life. Have you used the Vivino, the app that you take a photo of a bottle and it tells you the public rate or the crowd resource rating for it?
Ayodele Odubela: I actually have not.
Kirill Eremenko: Check it out. It’s called Vivino. It’s free and you just take a bottle, or you can go to a shop. What I do is I go to a shop and I take, you can change the mode from taking one photo to taking multiple and take five or 10 photos in a row and then just like on the fly, I think it uses … It does use computer vision to recognize the label and then it brings up the ratings. And they’re not like ratings from wine connoisseurs, they’re ratings from normal people like you and me who have drank the wine and just said, okay, what’s the score? What I think the score is.
Ayodele Odubela: That’s awesome. I’ll check in that.
Kirill Eremenko: Okay, cool. Okay, so that was project number two and project number three, you created a spam filter with 98% accuracy using the U phase in R. Was that another project in your university?
Ayodele Odubela: Yeah, that was actually another university project. That one kind of was the first foray I had into really imbalanced datasets. So as you can understand with spam, it’s usually less than 1% of your email may be spam. And that definitely relates to a lot of the work that I’m doing now where we’re trying to understand maybe what customer segments, someone might go into, but 0.002% of customers move. So that got me understanding what techniques might work within balance sets and what might not.
Kirill Eremenko: Okay. Well I’ve heard that Naive Bayes is really powerful for spam prediction, what would your comments be there? Like, why is Naive Bayes a good choice for projects like that or applications like that?
Ayodele Odubela: Yeah. I think part of that is the naive assumption there because it’s a little bit difficult when you’re working with those really, really sparse sets and not really relying technically too much on prior data. I feel like Naive Bayes just tends to be a better predictor when you’re looking at imbalanced data.
Kirill Eremenko: Okay. Got you. All right. So those three examples of projects, there’s plenty more that I’m sure you’re working on and different things. But what I’d like to know is, this would be interesting. So since you got into data scientist and you’ve been in a few different roles tell us a bit about like what was, what’s been your most exciting one so far? Well, of course. Okay. Let’s say this. What do you do at MINDBODY and before that, what was your most exciting role?
Ayodele Odubela: Sure. So right now I work with a lot of marketing, sales and customer service data. We’re really trying to understand our consumers better. So I’m on a project that’s about consumer segmentation and how they move between different segments. I also do a lot of work just fulfilling data requests for other departments. A lot of our customers are internal, so we do a lot of presentations and using some of the soft skills there. As far as pastorals, I would have to give it to my job at Astral AR. So there I worked heavily on machine learning. They are ‘drones for good’ company. So the drones they’re creating are resistant to firearms and being shot at or heavy or extreme kinds of weather. So essentially the drones are trained to understand what a weapon is on a sensor level.
Ayodele Odubela: So I did a lot of machine learning and understanding where metallic objects are including firearms, fire magazines, testing on different kinds of knives and the drones are supposed to be used in disaster relief situations or in law enforcement situations where they’re trying to get a better understanding of suspects and they’re able to deploy a drone, understand from a couple meters away if someone’s armed. And we as the public have the data on whether an officer knows if a person is armed or not in that moment.
Kirill Eremenko: If you’re listening to this podcast and you’re thinking it just took a whole new twist, then you’re sitting there like “Wow, what was just said? What is happening?” You are not alone. The first time Ayodele told me about this I also had my jaw dropped and yes, indeed the company’s called Astral AR, they have nothing to do with astrophysics. I’m not sure how they came up with name, but indeed you are hearing about bullet stopping drones and how they detect weapons. Are these flying drones?
Ayodele Odubela: Yeah. So they are drones capable of-
Kirill Eremenko: Flying bullets stopping drones. And you wrote a research paper about this. Is that correct? The conference paper?
Ayodele Odubela: Yeah. So I’m one of the co-writers on the paper. It’s called the Edna Bullet Stopping Drone. So we actually presented that at the IEEE conference last year for the global humanitarian. It was the IEEE Global Humanitarian Technology conference.
Kirill Eremenko: Wow. And you’ve already worked with law enforcement, is police going to maybe start deploying these things sometime soon?
Ayodele Odubela: Yeah, so when I was with Astral, we were working with the Austin police department and partly in getting officers trained to know how to use these, know how to pilot these. They’re very specialized systems that you actually are able to fly with just your thoughts.
Kirill Eremenko: Augmented reality, right?
Ayodele Odubela: Yes. So, and then the other aspect of that was really doing testing on the drone being able to withstand firearms and withstand bullets.
Kirill Eremenko: Wow. What is your life? This is crazy.
Ayodele Odubela: I know.
Kirill Eremenko: Okay. And so you used the word computer vision to detect firearms and stuff like that?
Ayodele Odubela: So the computer vision part was actually a really small piece. That’s actually more to detect anomalies and threats. So the computer vision behind the drones essentially is looking for someone in a threatening stance. So if they may have a weapon in their hand, but if you’re holding your hand outwardly from your body, there’s a couple of different stances that are picked up and might be anomalies. And a lot of the work I was doing was using radio frequency sensors. So similar to the TSA body scanners. And those are what’s actually determining this is a 45 millimeter weapon versus a Swiss army knife, or this is someone’s credit card versus a rifle.
Kirill Eremenko: Okay. Wow. Oh, that’s really exciting. Are you able to disclose or what’s in the paper in terms of what algorithm did you use for this?
Ayodele Odubela: Yeah, so what’s actually in the paper is it covers a lot of the different components of the drones, so exactly what it’s made of. There’s a metallic foam that they use to actually make it bulletproof. The algorithms, I can’t disclose completely what we use, but we ended up on using … I tested a lot of different algorithms against each other and ended up using a combination of decision trees and random forests to really get our final results. But our neural networks actually ended up performing the best, but in the context if these drones are actually sold to schools and police departments, a lot of that data is going to end up being public knowledge. So we decided that the trade-off in quote unquote accuracy for an explainable model was actually better in this case.
Kirill Eremenko: Does it better to do the explainability versus the accuracy?
Ayodele Odubela: We think about it in the very practical context lest there are going to be false positives and false negatives in the context that the drone thinks someone is armed when they are not actually armed. We want to be able to say why we don’t want to say the chemical component of item, whatever you’re holding is so much similar to this … So we really value explainability.
Kirill Eremenko: Yeah, yeah. Like, so neural network doesn’t allow that. It’s the decision trees that have that. Okay. Yeah. This concern is rising more and more, explainable AI and what do we do with neural networks, especially in an application like this where you’ve got a drone making decisions whether something is dangerous or not. I can understand where this decision comes from. Despite that you were able to achieve a 91% accuracy rate. Is that right?
Ayodele Odubela: Yeah. So that was really exciting, but I think it’s a really hard problem. It goes back to our poisonous mushrooms. Do I think 91% and 9% of the time it’s going to say someone isn’t armed when they are, it doesn’t cut it. And I feel like that’s the hardest part about solving this kind of problem is as a society deciding what is good enough. Do we say even though this isn’t perfect, it’s better than human judgment alone? So there’s a lot of implications and a lot of ethical concerns around that.
Kirill Eremenko: And it’s an example where humans can work with these drones, right? The concern here is of course the mostly in this situation, the false negatives that you have … let’s say if the drone has 91% accuracy and it does say that somebody has a weapon, then they don’t. Okay. Double check. No problem. It’s better safe than sorry. But the real problem is out of those 9% where it says if that person doesn’t have a weapon, you want to make sure to catch the false negatives when the person does have a weapon. Well you just have the human double checking those false negatives and double checking when the cases or looking at, I don’t know how few people would check that, but when the drone says there’s no weapon, okay, does the human confirm or not?
Ayodele Odubela: Exactly. [crosstalk 00:41:53] that going forward that’s going to be so much more part of any kind of AI system is we’re going to need to have human in the loop.
Kirill Eremenko: Yeah. Okay. How do you get a job like that? Ayodele really how, I don’t even know where to start. How did someone get a job like that?
Ayodele Odubela: You would not believe me if I told you I got this job on Twitter.
Kirill Eremenko: No way. You’re joking.
Ayodele Odubela: Yes. So the CEO of the company actually just saw me tweeting about data science and pretty much a couple tweets about data science and ethics and she reached out and we opened up the conversation about a lot of the machine learning I was doing in school and she was like “Do you want to work with sensor data?” And it just happened like that.
Kirill Eremenko: Wow. Congratulations. That is so cool. From now I’m going to use this as a case study. This is the best. Because I always tell people, make yourself visible, don’t just sit there and do your projects and like hang them on your wall. Put them out there, tweet, put them on LinkedIn, put them on, I don’t know, Medium. Put them on all these places where you can show your work, expose what you’re learning. Doesn’t even have to be ground breaking new stuff. Just like, okay, this is what I learned. This is the new thing I did. This is what I’m excited about. Somebody will eventually find … Companies are looking for good data scientists. That is inevitable. That’s going to just keep growing because data is growing.
Kirill Eremenko: The problem is that there’s an ocean of wannabe data scientists, of people who say they want to be a data science. It’s just because it pays well, it’s a cool trendy word and things like that. So it’s really hard for companies to pick out of this ocean, the right people. All you have to do is put your hand up. You only have to shine a beacon into the sky and somebody will come and get you.
Ayodele Odubela: Exactly. And I think that really just highlights the importance of being able to market yourself. I think that’s where a lot of my marketing skills came in and I’m like “Oh, use these hashtags, this’ll be fine.” But I think for a lot of people who are really analytical or come from academia or a lot of the hard sciences, being able to brag about your work and brag in a way that shows how passionate you are about the impact you get to have with your work is such a key that so many people unfortunately miss out on.
Kirill Eremenko: I’m reading a book now by Susan Cain called Quiet. It’s about how introverts get along in this world that currently looks so extroverted. And one of the things that she mentions is in the start of the 20th century we moved from a character, like a focus on character, how you are within your household, within your community, what values and traits you have. We moved to a kind of like what you could even call like a cult of personality that starting from this whole industrial revolution, not really industrialist, but basically when sales became more important, people like Dale Carnegie came along and things like that. It became more important the personality that you exhibit and how you get others to perceive you and in specific, particularly introverts and or more people that are more closed that are more to them, hold to themselves, they struggle with that because it’s not normal as you say, to brag about your work or it’s not a natural thing that happens for us.
Kirill Eremenko: Unfortunately that’s the world we live in and you got to get into it. You’ve got to learn how to show yourself and show your results so that people find you because they’re not going to find you just through knowing your brands or through the local community that you’ve always grown up in. We don’t live in that world anymore. So it’s just something that I think we all need to get accustomed to.
Ayodele Odubela: Absolutely. And I think one thing that might be helpful for some of those introverts is to think about it as you are being as efficient as possible. You can get to the same place and work so much harder without making those connections, networking, doing speaking events and just know that a lot of people who you see that might be extroverts are probably feeling a lot of those same fears. I definitely am one of those people where it’s easy to talk one on one, but speaking events, I can be vulnerable and say that’s difficult and I get scared and I get nervous. But think about it as you can be more efficient and get over some of those fears or you can do it the hard way and not really have to challenge yourself personally.
Kirill Eremenko: Yeah, yeah, totally agree. And speaking of speaking events, you’re coming to DataScienceGO and you’re going to be one of our speakers there. Congratulations.
Ayodele Odubela: Yeah. Thank you so much. I am super, super excited. I’ll be talking about fighting bias in AI and trying to highlight some solutions that help us analyze our models better and just be more critical about bias that we don’t really think about.
Kirill Eremenko: That’s very exciting. So fighting bias. It sounds like it’s … Because we talked about this, a bit about this before the podcast. It sounds like a passion of yours. Tell us about that, how… Bias and diversity and machine learning. How are you working on that? What is this project?
Ayodele Odubela: Yeah, so one of the aspects there comes from past examples. I think a lot of people might be familiar with the Tesla accident that happened I think a year or two ago in Arizona. I think it was Arizona. Yes. So one example that I had that shed some light about who this really impacts is I was giving a talk at the national society of black engineers last week. And at the personal development conference I had everyone raise their hands if they had issues with taking selfies at night and surrounded by a group of black engineers, almost everybody raised their hands. And I asked everyone in the room if they had heard about this accident and pretty much everyone raised their hands. And I asked them to think in 25 years if we’re going to see a statistically significant difference in who actually gets hit at night.
Ayodele Odubela: We are kind of on the understanding right now that self-driving cars are going to be more and more frequent. We are going to have to deal with them being on the roads. And that’s one of the things that I consistently think about that despite the fact that there are sensors, we know that sensors can fail. We also know cameras can fail. And I think that’s one of those measurement biases where the actual camera lens is the source of the hardware in almost every camera in modern day America or across the world is based off of the lens that was tuned specifically for lighter skin. So when the camera was being developed, they would do paper bag tests, they would take photos of slaves and usually their child takers with the children and actually tune the camera lens to the children’s skin color.
Ayodele Odubela: And what we end up with is a flawed hardware product that is the basis for more flawed hardware products. So I think in that sense, I do worry and I’m very concerned that we’ll see more people with darker skin are getting hit at night because we just can’t see them but we didn’t intervene early enough to really say we need to do more research and we need to make sure that this hardware that has the ability to detect people doesn’t make the wrong decision when it comes to people of different skin colors and how we differ.
Kirill Eremenko: Okay. And so you’re planning on using machine learning to fix that hardware problem?
Ayodele Odubela: I’m actually thinking about making it a hardware problem to fix. So part of this, I’ve actually started working on this project is collecting a lot more diverse data. So I’m actually working almost backwards, so collecting data sets of individual races, but being able to tune our camera to each of those. So if we have presets or if we have the things that our hardware aids for cameras to be able to recognize darker skin people better I think going forward, these need to be used so that we don’t end up with some of those problems.
Kirill Eremenko: Okay. Okay. Got you. And what will you share in your talk at DataScienceGO about this problem?
Ayodele Odubela: Yeah. So I’ll talk a little bit about what we can do now and then also a lot about explainable artificial intelligence and how there are practical solutions that right now we’re able to use to just understand our models better.
Kirill Eremenko: Okay. All right, very exciting. So you haven’t been to DataScienceGO before, in the past two years. Okay. So what have you heard about DataScienceGO and what are you looking forward to?
Ayodele Odubela: I’ve heard that it’s a lot of fun. So I’m really excited to network with a lot of the other data scientists there. Just looking at some of the other speakers there, some people I’m really excited to meet, but I want to talk with a lot of job seekers. I can imagine that we might still be hiring at MINDBODY, so really trying to also get a good idea of the job market in San Diego as well.
Kirill Eremenko: Awesome. Awesome. Very cool. And one thing I think you will like is that I didn’t realize this, but Pablos Holman who was our keynote speaker last year, he and this is a person that’s got 20 million views on his Ted talks and he’s worked with Bill Gates and we have quite a reputable person in the space of AI data science engineering. We were having lunch together and he pointed this out to me and I looked around, he was totally right, that the level of diversity at DataScienceGO is astonishing. It’s not something we were expecting that it just somehow naturally happened through our audience, through people who are coming there. But he said that this was one of like the top conference in the space of data science AI with like the highest diversity.
Kirill Eremenko: We had people from like 23 different countries come and from all sorts of nationalities. It’s like you, I don’t know, you find yourself meeting absolutely random people from all walks of life, from all colors of skin, genders, from whatever preferences and minorities from everywhere. And that is so exciting. I think with your talk will be very actual and relevant to this audience. So I’m very excited about that.
Ayodele Odubela: Yes, I’m super excited to hear that as well.
Kirill Eremenko: Okay, fantastic. And so let’s talk a bit about MINDBODY because you did mention just now, we talked about this, or you mentioned this before the podcast, you are hiring right now. You are hiring a coworker? So I warned you that because 10,000 people listen to this podcast as soon as we say something like you’re hiring a coworker, you might get a very large number of applications. So let’s make it easy for everybody who’s listening to understand if this is the right position for them and if they should indeed send you their CV. Tell us a bit about what kind of work is going to be involved and also what kind of, what are the requirements for the person that you’re looking for?
Ayodele Odubela: Absolutely. So we are looking for someone who is local or willing to live in San Diego, California. That is because this data scientist will work very closely with our consumer marketing team. So that means the MINDBODY app, our front facing application is the team that you’d be working with and working on that consumer product. Some of the skills required for the role are very basic level of sequel. Being able to pull data for customers is still really big for us. Also we’re looking for someone who has some marketing experience. So that’s either an understanding of a SAS company, metrics like lifetime value or click through rates. Any kind of marketing analytics knowledge would be really, really helpful for this role.
Ayodele Odubela: And we’re really agnostic about programming languages. If you’re an R Python user, we have more than enough tools to accommodate, but we’re looking for someone who’s really passionate any kind of interest in marketing is really helpful.
Kirill Eremenko: Got you. Okay. Very clear description. And how do people apply?
Ayodele Odubela: Anybody who is ready to apply can go to company.mindbodyonline.com/careers and you’ll be able to search for the data scientist role.
Kirill Eremenko: This is it. Let me check that again. So company.mind body dot what?
Ayodele Odubela: MINDBODY online
Kirill Eremenko: mindbodyonline.com/careers. Okay, let me see. If it loads for me just to the, okay. So everybody listening. Company.mindbodyonline – one word – .com/careers is where you can apply. Okay. Very cool. You even have a description of your perks. Very cool. Awesome. Okay, great. And so just to sum up, what is the mission of MINDBODY so that people can see if they relate to this mission?
Ayodele Odubela: Yeah. MINDBODY is really about connecting the world to wellness. So we have a combination of our apps and tools for fitness studios, hair salons, and masseuses. They’re able to interact with their customers better and we make it really easy to book an appointment especially if you’re traveling.
Kirill Eremenko: Very cool. Are you the first data scientist there or is it a big team already?
Ayodele Odubela: No, it’s actually a fairly large team. So we actually have about 40 people right now. But that’s broken up between business intelligence, business insights, data warehousing and a couple other departments.
Kirill Eremenko: Oh, so it’s quite a big company already.
Ayodele Odubela: Yes. So think we’re about 1900 people for reference.
Kirill Eremenko: Okay. Awesome. All right, well there we go. If anybody is looking for a job or knows somebody who is looking for a job in San Diego, then please refer them. Okay. I think we’ve covered on so many topics. Is there anything that we missed? Like I had a whole huge list of things to talk about, but things, it looks like we’ve gone through almost anything. Is there anything that you’d like to add to our discussion?
Ayodele Odubela: Yeah. Last bit is just that I am mentoring people right now who are trying to find roles in data science. So I want people to feel free to reach out to me on Twitter or on Instagram as well as I also mentor via the Sharpest Minds. It’s a very specific data science mentoring platform, but please feel free to reach out. I knew that something I kind of wish I had early on in my career.
Kirill Eremenko: That’s so nice of you. You’re only two and a half years into data science, but you’re already giving back to the community. That is amazing. I love that. Sometimes takes people decades to realize that it’s all about giving back. So huge kudos to you and I do hope people reach out and sounds like you do have a lot things to share from detecting poisonous mushrooms to bullet proof flying drones, it’s crazy. All right that’s awesome.
Ayodele Odubela: That could be an episode on its own.
Kirill Eremenko: Yeah, I can imagine. All right. And so you mentioned Twitter, Instagram, LinkedIn. I see you have over almost 4,000 followers. Let’s say everybody’s that 3,954 followers. Let’s tip it over 4,000. Yeah, I like LinkedIn. You get … That’s one of the best ways I think to connect because it’s got the professional-
Ayodele Odubela: I like LinkedIn and Twitter one and two for sure.
Kirill Eremenko: Yeah, definitely. And you tweet quite often, I am guessing.
Ayodele Odubela: Yeah, I’m pretty active. I’m probably most active in tech on Twitter. Some of those things can get reposted to LinkedIn, but fairly active in those conversations.
Kirill Eremenko: Fantastic. Okay. All right. Well there we go. That’s where people can find you and I guess one question I’m curious about is what’s a book that you’ve been reading or you’ve read that’s impacted your life or career? You have such an interesting story. There has to be something that’s impacted you and I’d love to know what it is.
Ayodele Odubela: Absolutely. I’m kind of going to use a very technical career one. it’s called Heard In Data Science Interviews. It’s by Kal Mishra and it came out just last year. So this has I think over 600 different interview questions with like really detailed answers. So when I was going through my job hunt this was my Holy grail and feeling really comfortable in how I answered some of those data science questions.
Kirill Eremenko: Wow. Okay. Heard In Data Science Interviews, right. I think I found it on Amazon already. 600, over 650, most commonly asked interview questions and answers by Kal Mishra. Okay, fantastic.
Ayodele Odubela: It’s great. It goes into some of those really specific questions too, like an NLP and other subject areas.
Kirill Eremenko: Okay. Very, very cool. So you basically provided a job opportunity and the solution to how to get it from podcast. Apply for the job, read the book. Very cool. Very cool. Sounds like you should write a book of your own. Like an autobiography or something. This is crazy.
Ayodele Odubela: It’s funny you mentioned that. Not plugging anything I promise, but I’ve been thinking about the idea mostly just geared towards people transitioning into technical roles because I feel it’s important for me to talk about these things and give back now while it’s still fresh in my mind, the struggles I’m still dealing with. So if that’s helpful to anyone, let me know.
Kirill Eremenko: Please, please do write it and when it’s done, come back on the podcast and we’ll talk about it and we’ll tell the world that it’s out. And yeah, that is such a common thing people transitioning from nontech to the world of data science, I wish more people did that because it’s possible. And examples like yours show that not just possible you can achieve great success in a short period of time.
Ayodele Odubela: Thank you for that.
Kirill Eremenko: Thank you. Thank you for coming on the show. So we’ve covered so many things. Super excited for your talk at DataScienceGO for everybody who’s going to be there is going to hear it. It could be a great continuation of the things that we talked about on this podcast and I personally look forward to catching up with you there in real life.
Ayodele Odubela: I do too. Thank you so much. I know that I’ve used a lot of your content to help me get ahead and it’s awesome to be able to chat and I’m so honored to be able to talk at DataScienceGO.
Kirill Eremenko: Fantastic. Thanks Ayodele. Hope you have a great day.
Ayodele Odubela: You too. Thank you.
Kirill Eremenko: Thank you ladies and gentlemen for being part of the SuperDataScience podcast today. Super pumped for you to have joined us for this conversation with Ayodele. I hope you feel as inspired as I am. My personal favorite takeaway from here has to be the bullet stopping flying drones. But in terms of valuable career takeaways, I think the whole notion of just being flexible and adaptable to changing all the time, to being accustomed to change. How many times did Ayodele change in her Masters? How many times did Ayodele change in her career trajectory? And that is what has helped her, not just get success but get success rapidly and something to consider. Where is a place in your life where you are maybe stuck or you have fear because you’re afraid of change and you don’t know what will happen?
Kirill Eremenko: Well just think of Ayodele’s story and that should inspire you to embrace change and jump into it, take the leap and see what happens. So that can be very, very rewarding sometimes. And on that note, as usual, you can get all the notes for this episodes at www.superdatascience.com/297. That’s www.superdatascience.com/297. There you’ll find the transcript for this episode and any materials we mentioned. Also the links and URLs for Ayodele’s profiles, make sure to connect with her on LinkedIn. Her Twitter handle is data_bayes. D-A-T-A underscore B-A-Y-E-S. You can also find it on the show notes, so make sure to connect with her and her great offer for mentoring. If you’re looking for a mentor then Ayodele might be a great person to connect with.
Kirill Eremenko: By the way, if you’re looking for a mentor, I highly recommend listening to what Tim Ferris, the famous writer and podcaster has to say about it so you don’t overwhelm your mentor and that’s the best way to build a relationship with your mentor. Just saying, if you’re going to reach out to someone about mentorship, first check those things out so that you can build a good well-grounded, fundamental mentor-mentee relationship. And finally as mentioned Ayodele is going to be a speaker at DataScienceGO and we’re going to have lots of exciting speakers like Ayodele. I would love to personally meet you there. We have hundreds of people already signed up already coming for DataScienceGO.
Kirill Eremenko: So if you haven’t gotten your tickets yet for DataScienceGO and you’re considering coming or you’d like to come or you’d like to hear more about Ayodele, this is one of those times to embrace uncertainty, take the leap and see what happens. Jump into it. So the website is www.datasciencego.com. Even if you have to fly from a different country, it is worth it. We had people from 23 different countries come last year and we’re expecting even more this year. So come see us in San Diego, 27th, 28th, 29th September. That’s already coming up very, very soon. That is what, when this podcast goes live, that’s two weeks away or a week away, just over a week away from when this podcast goes live.
Kirill Eremenko: So make sure to get your tickets if you haven’t gotten them yet. This is the last chance to transform your career and then DataScienceGO is not going to be available for another year. You’ll have to wait until 2020. So don’t miss out. Make sure you’re there. Make sure you skyrocket your career and take it to the next level. Once again, website is datasciencego.com. Get your tickets. And I’ll see you next time, until then, happy analyzing.
