scott-clendaniel-2020-05-15.mp3

T. Scott Clendaniel: [00:00:00] Going back to that problem solving, how we're gonna solve a problem. How can I support every other member of my team? Where can I improve myself in being able to contribute to that team? And also helping my teammates be able to contribute to the problem. And definitely focusing more on the holistic approach. You have to have a vision and an ability to get things done. I'm talking about a vision for how am I going to solve this problem. How I going to move my career forward? How am I going to move my move forward? How am I going to be able to help the organization as a whole?

Harpreet Sahota: [00:00:45] What's up, everyone? Welcome to another episode of the Artists of Data Science. Be sure to follow the show on Instagram @TheArtistsOfDataScience and on Twitter @ArtistsOfData. I'll be sharing awesome tips and wisdom on Data science, as well as clips from the show. Join the Free, Open Mastermind Slack Channel by going to bitly.com/artistsofdatascience. Where I'll keep you updated on bi weekly open office hours that I'll be hosting for the community. I'm your host Harpreet Sahota. Let's ride this beat out into another awesome episode. And don't forget to subscribe, rate, and review the show.

Harpreet Sahota: [00:01:35] Our guest today is a leader in the Data science space with over three decades of experience serving in various roles in business, analytics and artificial intelligence.

Harpreet Sahota: [00:01:43] He's a data scientist who specializes in predictive analytics, machine learning and training teams. Currently, he's a chief data scientist of the Strategic Artificial Intelligence Lab at Legg Mason, based in Baltimore, Maryland, where he's aiming to create cutting edge artificial intelligence that can be made accessible to all. And specializes in designing and developing new machine learning departments for mid to large size organizations. Before that, he served as vice president of analytics at Morgan Stanley, where he focused on fraud detection and prevention for the firm's 50 trillion dollar per year wire transfer business. He's also served as a lead artificial intelligence consultant for Analytics Edge LLC, where he was involved in a wide variety of Data science projects for 25 clients, creating nearly 100 predictive models for industries ranging from banking, cellular health care, hospitality and various nonprofits. He serves on the world board of directors for ICOM Digital Analytics, as well as a chief data officer on the Board of directors for the D.C. region at Gartner. He's got some amazing experience training teams in Data Science and has authored Harvard Innovation Labs Experfy Artificial Intelligence Course. He's also been invited as a guest trainer for 3M Asia and as a guest lecturer at the University of Maryland and John Hopkins University. So please help me in welcoming our guest today - a man whose mission it is to make Data science accessible for mortals - T. Scott Clendaniel.

Harpreet Sahota: [00:03:08] Scott. Thank you so much for taking time out of your schedule to be here to him. And I really, really appreciate it.

T. Scott Clendaniel: [00:03:13] You're quite welcome. Hey, I just heard that description. I don't think this guy can hold a job.

T. Scott Clendaniel: [00:03:17] What's up with that?

Harpreet Sahota: [00:03:19] He sounds like a very, very talented Data Scientist.

T. Scott Clendaniel: [00:03:23] Well, we'll go with that. I like that better. OK.

Harpreet Sahota: [00:03:25] Well you've got some tremendous, tremendous experience. Let's talk about a little bit about how you first heard of Data science, how you got into the field.

Harpreet Sahota: [00:03:33] What drew you to the field and some of the challenges you faced while you're trying to break into and create your own lane in Data science?

T. Scott Clendaniel: [00:03:40] Sure, yeah. I would love to be able to tell you that this was a carefully crafted strategic plan across 34 years. Unfortunately, I would also be completely lying because that's not what happened. I was actually when I started at the University of Baltimore, I was majoring in strategic planning, of all things. And I had met with the co-op department, which was sort of like internships. And they had said, hey, I've got this wonderful opportunity for you to go work with SteadyCorp, are you interested. And I said, yeah, that'd be fantastic. We set that up for me. So I go in and I'm sweating bullets. I'm 21 years old now, I'm like I don't know if I can do this or not. But I'm going to really try. So I go into the interview and get all the way to the final question. And the gentleman asked me "Gee, your major seems to be strategic planning. Why would you be interested in focusing on marketing analytics?" Well, my jaw almost hit the table because the one thing that the placement officer had not told me was the job was actually for marketing analytics. I was like, well, it's so important to be able to track your return on investment and ability to do things in a market place and set up metrics. And so that's why I was interested. And somehow that worked so well during analytics from that point on.

Harpreet Sahota: [00:05:00] That's one hell of a way to think on your feet. And get an answer out. That's awesome, man. So how much more hyped has I become since he first broke into the field?

T. Scott Clendaniel: [00:05:12] Well, way too much. I would actually say when I first started out, people talked about analytics and eventually they talked about business intelligence, but they used to refer to this field by some wacky names. For most the time I was in it, it was data mining, which actually started out as a pejorative. The origin of that term was actually people who were fishing in databases for relationships that they could sell to clients that weren't really supported by the data. And somehow that nickname attached to the entire field.

T. Scott Clendaniel: [00:05:41] So for the first 15 or 20 years, that I did this, it was data mining. Or the other one, which always makes me chuckle KDD, which was knowledge, discovery and databases. And one of my favorite sites out there is still KDNuggets, which is sort of like the CraigsList of analytics.

T. Scott Clendaniel: [00:06:02] Not big on the graphic design, but absolutely excellent material. But when I started out doing data mining or KDD or whatever you wanted to call, you had to beg clients to tell you anything or or to listen to you. And their response was always something along the lines of, oh, that's too geeky for me or oh I don't want all those statisticians running around telling me how to do my job. Yeah. Maybe there's something in it. Maybe there isn't - I don't know.

T. Scott Clendaniel: [00:06:31] The pendulum has swung completely the other way. Yes, you can apply deep learning to patching postage stamps or whatever bizarre application people come up with from this and that. And I'm really concerned that we're going to enter an A.I. winter, which will be the third one we've experienced, because I think it has been so overhyped that there's no way that the field can possibly live up to the expectations. So the short answer is way too much, way too overhyped. And yes, I'm really worried, too. It's interesting.

Harpreet Sahota: [00:07:02] I had no idea that data mining was used as a pejorative term like that. That's really, really interesting...

T. Scott Clendaniel: [00:07:09] Yeah, I, I didn't find that out until I started telling people I was doing that.

T. Scott Clendaniel: [00:07:14] And occasionally I got a raised eyebrow. It's like, what's up with that? But that's where IT came from.

Harpreet Sahota: [00:07:18] And that's interesting that I went from geeky to extremely sexy. In a short - you know, in that amount of time.

T. Scott Clendaniel: [00:07:26] My wife is still not convinced in my case about the sexy part, but I'm working on it.

Harpreet Sahota: [00:07:31] I think you yourself just bring Sexy back to data science

T. Scott Clendaniel: [00:07:34] Oh there, stop - You do go on. So.

Harpreet Sahota: [00:07:39] So you mentioned A.I. Winter that we might be entering into a third one. I think a lot of my audience is pretty new to Data science.

T. Scott Clendaniel: [00:07:47] Sure.

Harpreet Sahota: [00:07:47] And the field. Do you mind briefly just taking us through - first of all what the concept of an AI winter is and maybe just touch briefly on the last one that we experienced.

T. Scott Clendaniel: [00:07:57] Absolutely. This is also a great way to start a bar fight among A.I. geeks is to start arguing. When I went to it's happened. It didn't happen. But let me give you the least my interpretation of it. And A.I. winter is a period of time where the field of artificial intelligence sort of goes fallow. Where not a whole lot of development goes on and people start to lose faith in the field. The original concept, or most of what we call artificial intelligence now was developed back in the 60s based on the theory of a neural network reflecting the basic biological structure of synapses in the brain. And that if we could imitate mathematically what happens in the human brain, we could therefore come closer to achieving more human intelligence in terms of problem solving. But neural networks did not catch on very quickly, and there was a long period of time. 60S, 70s and early 80s were neural networks. We're sort of out of fashion. And it was very difficult to get people to adapt any type of predictive analytics technology. Now, in the 80s, it sort of came roaring back and there was a period of time in the late 80s where people are like, oh, neural networks are the solutions to everything. But at that period time, the theory was this all you needed was three hidden layers and a neural network. So you should be able to solve any problem because one layer was a point, two layers was a line, three layers was a shape. So if you got to a shape, you could basically bound any group of points in a hyper plane and be able to say, hey, I got it. And then it took off again. But neural networks were not nearly as successful as people thought they would be.

T. Scott Clendaniel: [00:09:28] So we went all the way through the 90s and the early 2000s. And what really caused things to come roaring back after that second A.I. winter so to speak, was the discovery of deep learning specifically for problems like computer vision. So if I need to identify the kittens in the YouTube videos, its very hard to do that with traditional statistics and adding additional layers to a neural network can help. But when you got to computer vision and several research papers that came out from Google, especially around 2010, it was, hey, now we've been able to find out how to add additional layers to a neural network and make them deeper. Hence deep learning, which is basically just very sophisticated neural networks with lots of hidden layers. We can get all sorts of performance that we've never had before. And so A.I. took off again. The problem is that people seem to have missed the fact that even Google, who is a huge proponent of solving everything about 10 years ago, has come out and said, well, yeah, they're really helpful for a lot of problems, but it's certainly not the solution to everything. And we're going to take a step back and not try and solve everything with neural networks or deep learning. And that seemed to go right over the heads of all the A.I. journalists. And that wasn't covered very much, though. They were big on creating the hype cycle on the way up. But when the cycle goes back down the other side, not so much. So I'm kind of concerned that we have now oversold what A.I. can do.

Harpreet Sahota: [00:10:54] So where do you see the field of data science, machine learning, artificial intelligence headed in the next two to five years?

T. Scott Clendaniel: [00:11:03] There are definitely applications where A.I. will continue to take off, but I think that the best opportunity in the next two to five years have less to do with autonomous vehicles and have more to do with solving more traditional problems. So, for example, big leaps have been gained and being able to diagnose disease, especially in terms of automating some of the processes that used to be done in radiology. So if you're going to make a Career choice, and you haven't started schooling, radiology is not the first choice I would recommend; because I think it's going to take over a lot of those slots. But more importantly, more traditional problems can be solved. And they're not nearly as sexy, but they have a lot bigger payoff. So which of my customers is going to open my e-mail? Which of my customers is going to buy? Which product? Recommender systems. From what you've seen from Amazon's been doing that forever, improving those types of areas. I think that the biggest applications are actually going to be on the cost savings side and eliminating waste and solving lots of classic classification problems, which my customers is going to buy. Which of my customers is going to [inaudible], which my customers might be a credit risk? Those type things are much lower hanging fruit, but they don't attract nearly the attention. But that's why I see the next three to five years having the biggest opportunity. And I think autonomous vehicles might be a little bit further down the road. But I don't know.

Harpreet Sahota: [00:12:27] So in this vision of the future. What do you think is going to separate the great Data scientists from just the good ones?

T. Scott Clendaniel: [00:12:35] I think whats going to separate the great Data scientists from the good ones is taking a small step back from the belief that everything can be solved just by throwing a another Python library at it, by adding package after package after package to problem solving. I think that we need to go back to say, let's look at this from the standpoint of what does the organization really need? What is the problem we're trying to solve? How are we going to define criteria for success? How are we going to say when good enough is good enough, as opposed to ultimately reaching for some unreachable state of perfection and moving more towards what happened with software development and more of an agile based approach and iterating through, I think great data scientists are going to become much more focused on how we're gonna solve this problem. What are our criteria for success? What stages can we do this in? And let's put on our Problem-Solving hats and stop trying to make code by itself, solve everything. 

Harpreet Sahota: [00:13:39] Very much in line with that myself.

T. Scott Clendaniel: [00:13:41] Oh, great.

Harpreet Sahota: [00:13:42] So let's talk aboutleadership in Data science. So that's it for you to be a good leader in Data science. And how can an individual contributor embody the characteristics of a good leader without necessarily having the title?

T. Scott Clendaniel: [00:13:58] I think to be a good leader, you have to first learn how to be a good team member. You need to be willing to focus on the greater good, for lack of a better phrase. In terms of going back to that problem solving, how are we going to solve a problem. How can I support every other member of my team? Where can improve myself in being able to contribute to that team? And also helping my teammates be able to contribute to the problem. And definitely focusing more on the holistic approach, but also sort of forever learning approach to life. I think those are very important. You have to have a vision and an ability to get things done. And I'm not talking about a vision for world peace, although that would be fantastic. I'm talking about a vision for how am I going to solve this problem, how I can move my career forward, how am I going to move my move forward? How am I going to be able to help the organization as a whole?

T. Scott Clendaniel: [00:14:48] As we've already discussed, I am older than dirt. But when I first started in this field, there were several folks that said to me, in business you should always be prepared to earn the organization who has hired you 10 times the amount of your salary back to the organizations. Now, maybe that was overkill. Fair enough. But not a whole lot of folks seem to be really concerned about how are they going to provide, let's say, a two fold return of their salary to that organization. If you start thinking about problem solving in that way, it does wonders to focus the mind as they say on the type of skills that I need to develop to be able to provide leadership for the organization.

Harpreet Sahota: [00:15:27] So for for someone who's, let's say, the first data scientist in the organization and they're kind of responsible for building up to Data Science practice, what are some of the challenges that you would see them facing and how do you think that they could overcome those challenges?

T. Scott Clendaniel: [00:15:42] There's been way too much of that. As a matter of fact, I can't tell you how many consulting assignments I have started off where someone wants the sun, the moon, and the stars - and within a week. I come back and said, gee, you guys know that you already have a dashboard report that solves this problem. And you've had it for years, but no one else is looking at it. You need to be really careful. If you're the first data scientist in an organization to make sure that you focus on a crawl, walk, run approach. The way to be successful is, one, make sure that whatever the first project you work on is going to be a success or as successful as you can make it. Because if an organization is very new to Data science, if the first project blows up, the conclusion tends not to be, oh gee, there was a mistake with first project. The conclusion instead tends to be, I told you guys that this Data science stuff is a bunch of hooey. And isn't worthwhile and that we shouldn't have hired this person and we shouldn't have a department. And it is extremely difficult to overcome that if you'll allow that to happen. So all of your efforts should be on making sure that you set expectations that you can reach and that you start out by saying, what are things that I can solve? What are the biggest pain points in my organization? How did they find that? How do they measure that? What can I do to solve one of those? Not by creating the most complex model. But what's the simplest, most understandable? Can I solve this one with a Three layer decision three, if you can get that into place first before you start fooling with stuff that is going to be almost impossible for naysayers in the organization to understand.

Harpreet Sahota: [00:17:14] If I understand what you're saying correctly - in order to set yourself up, as you know, in order to set yourself up to be successful in your career. From the get go, you should focus on having a lifelong learners attitude coupled with a bias towards action in order to solve problems not in the most fancy way possible, but in the most parsimonious, easy to understand manner so that you could sell your solutions to whoever your audience is.

T. Scott Clendaniel: [00:17:39] Yes. Simplicity is ridiculously underrated and not simplicity for simplicity's sake. But complex systems tend to fail. Complex systems are hard to explain. People do not support what they don't understand. Instead, they fear what they don't understand. Whatever we do in Data science, it's actually working with people. You need to start there. Are the people that I'm trying to help. What are the problems and how did they define those? How can I help people using the technologies I have, as opposed to let me take the technologies I have and force them on the people around me? The second approach does not work.

Harpreet Sahota: [00:18:18] What's up, artists? Be sure to join the free, open, Mastermind Slack community by going to bitly.com/artistsofdatascience. It's a great environment for us to talk all things Data science, to learn together, to grow together. And I'll also keep you updated on the open biweekly office I'll be hosting for our community. Check out the show on Instagram at @TheArtistsOfDataScience. Follow us on Twitter at @ArtistsOfData. Look forward to seeing you all there.

Harpreet Sahota: [00:18:48] What would you say the hero's journey looks like for a Data scientist or anyone in a data related role. That's may be going from an individual contributor level to chief executive level. Maybe a chief data scientist. Chief analytics officer, not one of those type of roles.

T. Scott Clendaniel: [00:19:03] Again I'm going to go back to the people skills. I know that we have all been smothered with articles, presentations and training on storytelling, but people seem to misunderstand what that storytelling aspect is.

T. Scott Clendaniel: [00:19:17] It's not going out and pretending you're Steven Spielberg and create a great movie about it. It's identifying what is your role? What is the problem you're facing? How do you solve it? What the result is going to be?

T. Scott Clendaniel: [00:19:31] It's taking a situation out of the world of let's explain lots of formulas and throw lots of complicated graphics on the screen and saying, you know, from a very simple point of view, what are we trying to fix? And I recommend that we use this model. Using this Data, on this time frame, And here are the goals that I understand. You have to start from the perspective of the audience you're trying to serve. You don't start from the technology and figure out where or apply it. You're out. Well, the problem is and then figure out which technology will help solve that problem.

Harpreet Sahota: [00:20:04] I like that a lot, especially that part about storytelling. What are some...

T. Scott Clendaniel: [00:20:08] There seems to be this feeling among Data scientists that storytelling is cheesy or that storytelling is beneath them, or my brilliance should overcome the need to tell a story. I am an expert. Well, OK, you know, no one wants to listen to you if you come home with that attitude.

Harpreet Sahota: [00:20:25] I think that's a very, very underrated skill, is the storytelling aspect. Can you talk to us a little bit about the importance of storytelling, what somebody could do to develop their storytelling skills and maybe, you know, do you have a framework for storytelling that you use?

T. Scott Clendaniel: [00:20:39] Absolutely. Essentially, it is who is your hero? And your hero is always your audience. What is the challenge that they are facing? What do they need to overcome? What tool or technology are they going to use to overcome? How is that going to happen? And what is the celebration or result of overcoming that probably going to be at the end? That's a very basic outline, but it does wonders to connect with your audience. A lot of people don't go into Data science because they just can't get enough relationships with other people. I hate to say that, but it's true.

T. Scott Clendaniel: [00:21:10] But you need to be able to figure out if you want to get the goodies, the desert in life, you need to figure out how to eat the spinach. And unfortunately - I actually like spinach but anyhow - people need to start with that in mind. And understanding how to help people is where you start and the storytelling approach to make it in terms of pieces that people can understand. Why do people subscribe to Netflix and why do so many people go to the movies when they're not Shut down during pandemic's at least. Why do people resonate with that, especially when they're completely fictitious stories?

T. Scott Clendaniel: [00:21:49] Well, because the human mind on a social basis is geared to understand stories and to resonate. So storytelling eliminates the noise of technology and complexity and pulls out the signal of what is important and how it's going to solve the problem. So if you use storytelling from that perspective, the same way you would look at Data, I need to separate signal from the noise, present the signal, eliminate the noise. Storytelling is the way to communicate the signal and helps block the noise of technology overwhelm.

Harpreet Sahota: [00:22:24]  Does the way you tell a story...

Harpreet Sahota: [00:22:27] Does that differ based on let's say if you're talking to your manager versus maybe talking to a roomful of executives? Do you have any tips for Data scientists?

T. Scott Clendaniel: [00:22:37] Absolutely. Go back to the example we just talked. That was a great question. What are the different components? Your audience is the hero. You are not the hero in the story. The audience is the hero. The audience has the journey. The audience has the problem. How are you going to help the audience be successful? In a lot of cases, people just don't care what algorithm you use. I hate to break this to folks. Now, there are certain organizations where you've got rooms full of people who love to discuss the algorithms. Okay, great. Fantastic. But you've got to know who your audience is. And to use the example you laid up before being the first person in the organization. How are you going to make the organization the hero? How are you going to help them overcome the challenge?

Harpreet Sahota: [00:23:24] So what other soft skills would you say that candidates are missing that are really going to separate them from the competition? We've talked a little bit about, you know, communication skills and storytelling. Are there any other skills that that, you know, you think Data scientists nowadays are lacking that they should really ramp up on?

T. Scott Clendaniel: [00:23:44] Yeah, you know, everyone wants to say that the skill set that they have is the most important because it helps their egos and I guess I'm no different. But the advantage that I had of coming into the field did a lot of disadvantages. Well, the advantage I had was having studied strategic planning and having an MBA, the degree itself I don't recommend a whole lot of people go out and study business or get MBA, that isn't the point.

T. Scott Clendaniel: [00:24:09] For a long time, I couldn't figure out why I had taken that degree and ended up in the field that I'm in. I was like gosh, I just wasted so much time. Worrying about this, about that all this stuff with advanced accounting. And my goodness, what I didn't realize was this.

T. Scott Clendaniel: [00:24:24] I was building problem solving skills and that's what seems to be missing. Understanding the problem, defining the problem, defining the criteria for success and laying out a plan to get there. That's what you really need to do. Those are the soft skills you need to have. Keep working with whoever your audience or your client is to make sure that they're comfortable with the definition of problem, because many times they aren't. And if you're coming into an organization or if you're applying for a job in an organizations like, oh, yeah, I read this great article about LSTM and we incorporate that in our next plan. Something is really cool technology. My cousin Fred was at M.I.T. and they talked about this lots. We need to have that in my organization. And I have no idea how to apply it, run for the hills. Because what's gonna happen is if you do get the job and if they do hire you, they are going to turn on you.

T. Scott Clendaniel: [00:25:19] When they figure out that LSTM has nothing to do with the problem they're trying to solve. They're not going to blame themselves. They're gonna blame the new Data scientist that they just hired.

T. Scott Clendaniel: [00:25:28] Be aware of that before you start.

Harpreet Sahota: [00:25:32] So what are some questions we could ask ourselves at the outset? You know, when we're starting a project, maybe questions we can ask ourselves and questions we can ask our stakeholders. That can really help us clarify exactly what the problem is. And then...

T. Scott Clendaniel: [00:25:47]  I got a great one for you.

Harpreet Sahota: [00:25:48] Yeah, definitely.

T. Scott Clendaniel: [00:25:50] Let me tell you what I used to do, which was a bad idea and then let me tell you a much better idea. What I used to do is to ask people at the end of the year, what do you think would get you a great bonus? What do you think would get you a promotion? What do you think would get you a big salary? And people hemmed and hawed and felt really uncomfortable and they didn't really want to share. I didn't go so well. So instead.

T. Scott Clendaniel: [00:26:20] The question I started asking was first, what do you think we get your boss a giant promotion or a big bonus, or a big salary increase or senior management at the very top of this organization? What? And oh, my goodness. The story starts to spill. Oh, you wouldn't believe this person or that person is really into this or that. And all they talk about is blah, blah, blah, blah, blah, blah, blah. Or ask them what is the biggest pain point that your manager, your managers manager seems to talk about all along?

T. Scott Clendaniel: [00:26:50] And they will tell you more than you want to hear. But those are the problems you should be looking at. Those are the problems you should be trying to solve, because wherever the pain is, is where the opportunity is. You need to smooth out the pain and fix the pain, not, oh, it could be this or it could be that. Or, you know, rainbows and unicorns find the pain, diagnose the pain checks, the pain kind of ties and tea out of your ear.

Harpreet Sahota: [00:27:15] Your framework for storytelling because the hero in your story is not you. It's your audience.

T. Scott Clendaniel: [00:27:20] Absolutely.

Harpreet Sahota: [00:27:21] So you shouldn't try to pick up projects that you think are interesting because there are things that you want to do. You should to take on projects that are going to, like you said, solve a pain point for the organization so that you could get that 10 X return on your salary that they're investing in you.

T. Scott Clendaniel: [00:27:35] There is a hidden Data science message in the movie Dr. Strangelove of all places. So you go into the movie and, you know, all the trippy graphics and all that kind of stuff. That's all well and good. But actually, if you look at the discussion between the ancient one and strange, the ancient one turns, to him and says you still haven't figured out the most basic lesson in all I've tried to teach it. And he's like, well, what is that? It's not about you. And that really resonated with me because it isn't and it isn't about any of us. It is about the team we're trying to support the organization. We're trying to support where we're trying to get promoted, where we're trying to get our salary increases, where we're trying to do whatever that is what it is about. The part for you in any situation as how can I contribute to solving that problem? What can I do to improve myself to become better at solving those kinds of problems? What can I learn that would help prepare me for the next time a similar problem comes up?

Harpreet Sahota: [00:28:33] Speaking of, you know, what to learn or how to to learn what we should learn next.

Harpreet Sahota: [00:28:38] I feel like there's a lot of boot camps that cover a lot of the technical aspects of Data science.

T. Scott Clendaniel: [00:28:43] Yes, there are.

Harpreet Sahota: [00:28:45] There's there's not enough.

Harpreet Sahota: [00:28:47] I think that help develop the business acumen or product sense for Data scientist. How do you think a Data scientists could develop and cultivate a business acumen or a product sense for themselves?

T. Scott Clendaniel: [00:28:59] Good question. Any type of management type of course, in terms of helping solve problems. That's great. I think that actually there are lots of lessons available within computer science curriculums. It's just that people want to jump past that. A great exercise is actually we go through the books that have been published. There's a whole slew of them now on how to pass the Data science exams to get into Google or Facebook or wherever. And where I think it's really helpful in studying those books is they lay out a specific problem and they're looking for you to be able to solve it. So it it's not the answer so much as is what is the process that you went through to get to an answer? What did you think about what did you consider? How did you lay out some options? So even within your own curriculum, I think there's a lot to be learned that people keep focusing on the syntax as opposed to what's the problem? How do I define it and what is the thought process I use to get to that design thinking? I think it's fantastic. Can't recommend it highly enough. Anything that focuses on general problem solving, that's what you want to do. And and if you're truly geeky like myself, think of it as a video game. If this were a video game and I was trying to get my character from this level to the next level, what are the problems I had to solve and how would I go about doing that? And if you think about it that way, it pulls you out. Which line of code? Which package you make an import, which sort of software used more towards? What do I have to do to get the problem solved? What do I have to do to make my audience the hero in their story?

Harpreet Sahota: [00:30:34] You hear a lot of well, at least I do with my mentees. I hear a lot of you know, the first question they ask is, which algorithm shall I use? Which dataset should I use? And I'm like, how about you ask yourself, which problem do you find interesting that you go solve... 

T. Scott Clendaniel: [00:30:46] Thank you!

Harpreet Sahota: [00:30:50] Flip it on it's head, because that's the most challenging aspect, I think, of Data scientists not necessarily...

T. Scott Clendaniel: [00:30:56] Are you familiar Familiar with a gentleman by the name of Leo Breiman - famous researcher from California who helped develop CART and Random Forest.

T. Scott Clendaniel: [00:31:03] One of his theories came out around the year 2000 and 1999 was called, or the phrase that he popularized was the multiplicity of good models and cutting to the chase. What it says is if you do all your Data definition and your problem definition and your feature engineering correctly, all kinds of algorithms can solve the problem. Not every algorithm can solve every problem, but lots of algorithms can solve the problem. If you solve the problem correctly into your feature engineering, that's where you start. And so lots of algorithms will then solve the problem. Now, there are exceptions to that. If you've got like a P greater than N problem, in other words, you've got more variables than you have records. You're pretty much stuck with support vector machine. If you're going to try and find kittens in the YouTube video - I like kittens -  I think that you're in a situation where you're gonna have to do some type of deep learning. But for lots of other more common business problems, all kinds of tools will do it. And so Breiman had raised this idea that there's a multiplicity of good models or many good algorithms you can use to solve the problem. Get the other stuff fixed first, then pick the algorithm.

Harpreet Sahota: [00:32:06] Absolutely. Love it, yeah. Because feature engineering is really, you know - being able to build out the complexity from that real world Data generating process. Like that Raw data is not gonna do you much good at all. You need to use your ingenuity to build out features, capture that complexity in the form of new features so that you're setting up the dataset for success.

T. Scott Clendaniel: [00:32:24] Absolutely agree with you.

Harpreet Sahota: [00:32:25] Can you share some tips or words of encouragement for our listeners who's got like a couple of decades, maybe 10 to 20 years of non Data related experience under their belt and they're now trying to break into Data science? What challenges do you foresee them facing and how can they overcome these challenges?

T. Scott Clendaniel: [00:32:43] Sure. One thing to do is make sure that you're doing it for the right reasons. And I don't mean to be all sort of philosophical here, but I think it's great if you want to join the field. Again, my background was, you know - I tease with people, but the first half my career, everyone told me I couldn't do this because I didn't have a PhD in statistics. And the second half, of my career everyone told me, I can't do this and I don't have PhD in computer science. And yet, you know, I've been doing this since the Reagan administration. So somehow it seems to be working out OK. It's more about how are you going to get into the field. And part of the challenge you're going to run into is the fact that fifteen years ago, our field changed a lot because we brought in so many software developers. And I think it's fantastic. They're amazing people. They have skills that I do not have and I wish I did. And we brought all those people in the field which is fantastic. Unfortunately, what happens as a side effect of that is everybody became convinced that you have to be a pure developer to be able to make a contribution in the field. So first of all, you need to be aware that there is a mindset among a lot of people who hire for entry level folks that you really have to developer skills, be aware of why you're entering the field, because if you're not willing to put in that work to get some of those fundamental software developing skills, it's going to be very hard to convince folks that you're a good fit for the position.

T. Scott Clendaniel: [00:34:04] I'm not saying that's right. I'm not saying I approve of that. I'm saying that that just is the reality. But if this is something you love and you want to be in it because you don't think you'd be interested in anything else, that needs to be your why, that needs to be your motivation to carry you through what you need to learn. There is a fair amount to learn. If you're jumping into it because it's your third cousin twice removed, told you that it pays a lot of money. It's gonna be hard to keep the motivation and you're likely to get bored and frustrated and leave the field. So maybe this isn't a great fit for you. So make sure that your why, whatever your why is, is enough to fuel you to get you through the difficult patch to get you to where you want to be. But look, I am not a genius. God knows, I wish I was. I hope that I'm at least average IQ, maybe. And what has propelled me through this is the focus on meeting people's needs, problem solving, and figuring out the technology to get through this. If I can do this, you can do this. If I can get through this, you can get through this. I think those will be my words of encouragement. Find your why and make sure it's the right why and use that to propel you through the more technical stuff.

Harpreet Sahota: [00:35:08] Are you a fan of Simon Sinek?

T. Scott Clendaniel: [00:35:11] I don't know Simon Sinek.

Harpreet Sahota: [00:35:13] Oh, man, well he wrote that book. Start with why,  I think you'd really, really...

T. Scott Clendaniel: [00:35:16] Excellent. I didn't -  I had never made the connection. Thank you.

Harpreet Sahota: [00:35:20] So a couple of follow up questions based on what we just talked about. So what advice or insight can you share with people who are breaking into the field and they look at these job postings and some of them want the abilities of an entire team wrapped up into one person?

T. Scott Clendaniel: [00:35:37] Oh yea, matter of fact most of them do.

Harpreet Sahota: [00:35:38] And they end up feeling dejected or discouraged from applying. You know, what words of encouragement do you have for them?

T. Scott Clendaniel: [00:35:44] There is a huge challenge in that because it's a relatively young field. People are terrified of making a mistake. So they try and create a job description. That is what I will call boss proof, meaning that their boss or their bosses bosses are gonna come in and say, oh, well, I read an article in Forbes that says, you know, you need to be able understand convolutional neural network to be able to take the job. And they say, OK fine, we'll put CNN on the job description. Then they pass it on to the poor H.R. person. Now, most human resources folks are experts in human resources. Not experts in data science. So somebody shoves a poorly written, overly detailed job description to them and says, go find these people. And there are so many people applying for so many jobs, the only choice the human resource person has is use the applicant tracking system to look for keywords. Throw a bunch keywords in there that try to match what the hiring manager is looking for. And so you end up with job descriptions that look like an entire computer science department. So what I would suggest is, one realize very few people have all those requirements. I would say if you have more than half and it's something you're interested in, send in an application, see what happens. That's the first thing. Don't rule yourself out. Second of all, what I found to be really helpful, especially on LinkedIn, is take your 10 favorite reasonable jobs that you would like to have, that you think you might be a possible step and dump out what the keywords are from various descriptions.

T. Scott Clendaniel: [00:37:10] Then figure out which ones showed up in the most job descriptions, a job you like. Don't worry about location, but do worry about comparative level or your entry level mid-level, senior level. Then start going down the list and figure out what do I have? What do I not know if it's an important bullet point in nine out of 10 jobs that you want to apply for and you don't have it. That's a problem. So that might be how you prioritize what you're going study, what you're going to do, some experimental projects or what you're going to visit a local A.I. and Machine Learning club to discuss, those type things. What are more articles I can read on that? Then go down the list. But if you have more than half, try it out. I will tell you that it is a scary world out there in some cases. I turned down for a job once not too terribly long ago, where they said, well, well, how much work have you done with autonomous vehicles. I was like, well I work in financial services. I've never worked with autonomous vehicles. I don't know, that might be a problem.

T. Scott Clendaniel: [00:38:08] You're going to run into that.

T. Scott Clendaniel: [00:38:10] The trick is to apply to enough places and to work on your skills that so the skills that you have and what you love to do matches what the job is. And you're going to run into some cases where you really want to be working on identifying the kittens in YouTube videos, the hiring manager wants you to have something completely different.

T. Scott Clendaniel: [00:38:28] That's going to happen, but you keep trying.

T. Scott Clendaniel: [00:38:31] The job search process is horrible. I don't have anything positive to say about it. I think what people misunderstand is they think they're having a hard time and everyone else is fighting this easy. I don't know if anyone likes the job search process.

Harpreet Sahota: [00:38:44] It's definitely very, very strenuous. But, you know, if you take your approach to it and gameify it, treat it like a game and just take every rejection and every setback as a learning opportunity, just change your mindset about it and turn it into a positive, right.

T. Scott Clendaniel: [00:38:59] And if you're in a larger organization, you're usually much better off trying to make a lateral move to a job that looks more like what you want in the organization you're already in, as opposed to making a leap to a different organization.

T. Scott Clendaniel: [00:39:10] If you haven't built up any experience

Harpreet Sahota: [00:39:11] I really liked that advice about just taking inventory of 10 different job postings and highlighting the terms between them that are common and now, you know, what it is that you need to study for. That's actually advice that I'd given to one of my mentees just yesterday.

T. Scott Clendaniel: [00:39:27] Fantastic.

Harpreet Sahota: [00:39:28] You've got a ton of experience and finance and fintech. Do you have any suggestions for finance or fintech Data science projects that an aspiring Data scientists could tackle?

T. Scott Clendaniel: [00:39:37] There are a couple of data sets out there on Kaggle regarding credit scoring. Thats always a good one. Fraud detection is a really big one. What I would suggest you stay away from is the stuff that starts to dig into advanced economics or time series prediction of stock prices. Because it's just really, really hard and there tends not to be the level of Data you need to come up with the great prediction. As opposed to credit scoring, fraud detection, credit profiling. Those types of projects would be closer to fintech. Chat bots are ridiculously popular. It's not a field that particularly lights my fire. But there's a lot of opportunity for it in fintech, so that would be an a good one.

Harpreet Sahota: [00:40:17] Do you have, like any good case studies that somebody who's into fintech or finance should check out?

T. Scott Clendaniel: [00:40:24]  The kaggle?

T. Scott Clendaniel: [00:40:25] Credit scoring is a good one. The Kaggle Aetna challenge for good drivers. That's another good one. Those would be two to start. I don't know of a ton out there, but let me give you a good example from fintech that's really applicable to a lot of different fields. And that has to do with credit scoring. For example, if you look at the total amount of debt that somebody has, total revolving debts all your credit card type stuff, that actually is terribly predictive. If you look at the total credit lines that somebody has. So if you add up the total credit limit on other revolving credit, that doesn't tend to be too predictive either, much less than I would have imagined it would be. But if you take the ratio of how much they owe to how much they could go if they wanted to. That's hugely predictive. Now, here's the challenge. If you just throw those numbers into most traditional statistics packages, it's going to say, no, there isn't a real high correlation on either one of these. The odds of it calculating the ratio on its own is hard. They do not do a good job at. So that's an example of if you follow just purely classical statistics and threw those two  inputs out because each one by itself isn't terribly predictive and didn't try a ratio, you'd be out of luck and you would have missed one of the most important predictors of overall product performance. So that's why I say the problem solving, which are these things are related. What might I try? What are some different combinations? I throw out about 80 percent of everything. I try because I have learned over the years that trying to outguess the Data is usually a fool's errand. It is better to think through the Data. Try a bunch of different stuff. See what makes sense. See what holds up. See what validates and use that. Rather trying to massage your own ego and think, oh well, I'm so smart I got to figure this out without testing. That tends not to work. Let the data lead you. 

Harpreet Sahota: [00:42:13] Feature engineering to the rescue, right

T. Scott Clendaniel: [00:42:15] Absolutely. 

Harpreet Sahota: [00:42:16] Develop good features, it's important. So having to help so many, you know, 100 predictive models say this is something that I think a lot of people who are breaking into the field or maybe brand new to the field don't have a lot of experience with. And that's the things that we need to do after a model has been deployed and put into production. So what are some things that we need to be cognizant of and monitor and track once the model is deployed, both from the Data scientists perspective and the business perspective?

T. Scott Clendaniel: [00:42:47] Plan for obsolescence is my biggest lesson there. So when you think about model risk management, all models deteriorate over time. The model looks fantastic. Today may look horrible six months from now. Rather than trying and just sitting in fear that the models are going to deteriorate. Plan for it. How often are you going to check the model. Are you keeping, in essence, a control group of records you don't score? Are you using a population stability report to see if the world has changed? Is the performance - You know, it's going to deteriorate at some point. So how often are you checking and what's your plan to replace the old model with a new model? What is your champion-challenger test plan to be able to say, OK.

T. Scott Clendaniel: [00:43:24] I came up with two new models that might be able to replace the old model that's starting to fail. How are you going to test all three together and compare scores? How are you going to replace that? Have you already worked with the technology group on when it's going to be replaced? Do you have the resources for that. That at all needs to be baked into the plant? Don't wait for disaster to happen. Know that you're going to have model drift almost no matter what you do, rather than try and fight gravity. Say May. OK, so I'm just gonna plan for that. And here's my schedule and here's how I address it. Here's a test for it. And this is going to do. Models are not one and done and planning for that up front. Also, you're going to have much less problem with people saying, see, I told you that this stuff doesn't work. Well, it works for a period of time and the process is always updating. What I like to call non-stopimization. In other words, you're always gonna optimize. You're always going to improve. You're always going to try and be better. But you know that models are going to deteriorate. So here's how they're replacing them.

Harpreet Sahota: [00:44:22] Also, a very good approach to have in your life as well. So what advice do you have for Data scientists who have who feel like they don't need to learn anymore? They are ready to grow any more, they don't need to improve any more, that they've already learned everything they need to learn to be successful. What would you have to say today, scientists in that mindset?

T. Scott Clendaniel: [00:44:42] I would not tell that person anything, because if they have that mindset, they're certainly not gonna listen to me. I have run into those people in my life and me trying to explain to them all the reasons why that isn't true. It's usually a waste of breath. And it's not a useful exercise. One thing that they may want to think about is pretend they were in the industry 20 years ago. What are the skills they would have years? And would you still use them today? Well, no, because this doesn't last. I got this year. Okay, great. What makes you think that that isn't going to happen again? Ten years from now, there is no point in time where you can stick a pen in the bulliten board and say, OK, progress stopped here. Nothing has improved in my field. Since the rate of acceleration is increasing geometrically in this field. So if you don't like constantly learning, this is really a bad field to go into.

Harpreet Sahota: [00:45:28] Yeah. Learning never, ever, ever stops as Data scientists. What I absolutely love this field. I used to be an actuary, I used to be a biostatistician and.

T. Scott Clendaniel: [00:45:36] Oh, jeez, that's heavy duty, my friend.

Harpreet Sahota: [00:45:38] But I didn't feel like I was continuously learning, continuously growing, wasn't a challenge. It's almost like OK cool. I passed a bunch of exams. And is that all there it is. I mean, that's the one thing I love about science is that you have to have to always learn.

T. Scott Clendaniel: [00:45:55] So what's the old joke about a PhD? You learn more and more about less and less until you know everything about nothing. You can really do that in this field because everything changes. I, I would say that the industry has changed more in the past seven years than it had in the twenty seven years prior, at least from my career experience.

Harpreet Sahota: [00:46:22] Are you an aspiring Data scientist struggling to break into the field well then checkout dsdj.co/artists yo reserve your spot for a free informational webinar on how you can break into the field that's gonna be filled with amazing tips that are specifically designed to help you land your first job. Check it out, dsdj.co/artists

Harpreet Sahota: [00:46:48] Last question here before we jump into a lightning round.

T. Scott Clendaniel: [00:46:51] OK.

Harpreet Sahota: [00:46:51] What's the one thing you want people to learn from your story?

T. Scott Clendaniel: [00:46:54] There is a real benefit to taking your ego out of the equation and being humble and being willing to continually learn. And so learn that you need to help people. And that's what it's all about. If this was impossibly hard, I could not do this for a living. I will tell you a personal story that seems to resonate with a lot of people. When my son was itty bitty, he was about three years old. My wife at the time kidnapped him. I was living in Hawaii and she took my son back to Baltimore and said, you know, if you want to see your son, you're gonna have to come back to Baltimore. I was going to have to give up my career to be able to do that. Now, what is going - this is horrible. So she says, I'm going to put our son on the phone. And by the way, I told him that the reason you can't be home right now is because you're packing his toys. It's like, why did you tell him that? Here he is. He gets on the phone and this tiny little boy's on the other end of the phone from 6000 miles away. That goes Daddy, are you done packing my toys yet? I was just crushed. I was like, what am I going to do? I had to give up my entire career, pack the house, move back to Baltimore and all of the financial services companies that I would have worked with back in the Baltimore area had moved to Delaware because the change in tax laws, I had to work as a temporary secretary for several months trying to figure out what I was going to do. And then I was like, I wonder if I would be interested in this artificial intelligence stuff, because it helped me a lot. So maybe I can help other people with that. So I had to recover from all those obstacles and become hopefully a leader in the field of artificial intelligence, if I can come back from that.

T. Scott Clendaniel: [00:48:30] You guys can be successful too. 

Harpreet Sahota: [00:48:32] That hits me because, like, my son was just born a week ago today.

T. Scott Clendaniel: [00:48:36] Oh, congratulations.

Harpreet Sahota: [00:48:38] Thank you. Thank you. Yes. So, man, I really hope that my wife will never do that to me.

T. Scott Clendaniel: [00:48:44] And he's fine.

T. Scott Clendaniel: [00:48:45] I'm fine. He's fine. His mom's fine. Everybody's fine. It Has a happy ending. But it didn't seem very happy at the time.

Harpreet Sahota: [00:48:50] So let's jump into our lightning round. OK. T. Scott. Clendaniel, what does the T stand for?

T. Scott Clendaniel: [00:48:55] Timothy.

Harpreet Sahota: [00:48:57] Timothy. All right.

Harpreet Sahota: [00:48:58] So what are the two five letter words that really grind your gears and why?

T. Scott Clendaniel: [00:49:03] Model versus magic and the fact that people don't know the difference.

Harpreet Sahota: [00:49:06] So what is an academic topic or just an area of research or interest outside of Data science that you think every Data scientist should spend more time researching on?

T. Scott Clendaniel: [00:49:19]  Graphics.

T. Scott Clendaniel: [00:49:20] Because the idea of using a picture to communicate an idea does wonders to get furthering your career and to help people on the other side.

Harpreet Sahota: [00:49:27] So what is your favorite question to ask during an interview?

T. Scott Clendaniel: [00:49:31] What is holding you back from hiring me? I'm sorry you meant this kind of interview. No, I have no idea. But at least I'm honest enough to admit it.

Harpreet Sahota: [00:49:42] No, that's that's a good question. I like that. What is holding you back from hiring me?

T. Scott Clendaniel: [00:49:45] What concerns do you have that I might be able to address? Because what it does is it forces the person on the other end of the line to say, oh, OK. And if they say none, that's good to know. If they say something that I can address, I can address it. And if it's something I can address, it's probably not a good idea to go forward.

Harpreet Sahota: [00:50:01] It's very, very interesting that you say that, because that is also one question I ask in my interviews. It's pretty much one more time I'm spending with you. The more and more I realizing how similar we are. It's actually and I'm looking at myself the next 10 years.

T. Scott Clendaniel: [00:50:16] There you go.

Harpreet Sahota: [00:50:18] Yeah. I Like it. It does. It's a great question because it allows you to go inside their head to see what gaps and concerns they had.

T. Scott Clendaniel: [00:50:24] Absolutely.

Harpreet Sahota: [00:50:25] You could then flip it and be like, all right. Well, actually, we didn't talk about this thing. And here's what experience I have with that. So what's the weirdest question that you've been asked in an interview?

T. Scott Clendaniel: [00:50:34] I'm going to have to go with the reinforcement learning for autonomous vehicles in that the same person had used one of those animation things you can use with one of those go cameras in the interview and put first a pirate hat on his head and then a cat on his head and said, doesn't this look cool. And then ask me about reinforcement, personal once, like, I don't think this is going to turn out well.

Harpreet Sahota: [00:51:00] So what's the number one book you'd recommend our audience read and your most impactful take away from it?

T. Scott Clendaniel: [00:51:06]  In search of excellence

T. Scott Clendaniel: [00:51:07] Old book from the 80s, had a huge impact on me. Bunch of excellent McKinsey consultants went on a search. The top performing corporations in America. Coming up on 40 years ago and found out that what made them successful is very different from what most people think, including the other members of the consulting firm before they started the study.

Harpreet Sahota: [00:51:26] Interesting. And what would you say was your most impactful takeaway from from that book?

T. Scott Clendaniel: [00:51:31] Just how important customer obsession is and that it is not silliness and it's not being a jolly do gooder. It is knowing your customer and how to help them solve what's causing them pain and different applications of that.

T. Scott Clendaniel: [00:51:51] Another one that came out of that was an MBWA - management by walking around. Don't just listen to the executives who are on the next level down. You need to be able to get on the front lines and visit and listen to customer interactions or you don't understand what your business is really doing.

Harpreet Sahota: [00:52:08] Yeah, I'll definitely add those to show notes. So if we could somehow get a magic telephone that allowed us to contact 20 year old T. Scott, what would you tell him?

T. Scott Clendaniel: [00:52:17] He wouldn't listen to a thing I said. So probably not much.

T. Scott Clendaniel: [00:52:21] Humility is a lesson that I wish I had learned a lot earlier. Once upon a time, I thought I was a smart person until I started working with, like, spooky, smart people. And then I was like, wow, I better find a better way to differentiate myself because I'm not nearly as bright as I thought I was.

Harpreet Sahota: [00:52:42] So what is the best advice you have ever received?

Harpreet Sahota: [00:52:46] Treat others the way they wish to be treated. Not the way you wish to be treated, which is the golden rule. So what do they call it, the platinum rule. You really need to listen to people and peoples what it's all about, whether no matter what you study, no matter what makes you happy. If you can't learn to help solve other people's problems, you're in for a long, hard road.

Harpreet Sahota: [00:53:09] What motivates you?

T. Scott Clendaniel: [00:53:10] What motivates me is the desire to help. Lots of people can be made happy from lots of different things. If you grow up in one culture, one type of music might make you happy. But if you grow up in another type of culture, another type of music might make you happy. So many different things can make people happy. If you learn to become happy in the journey of helping others, then you can be happy and they can be happy at the same time. You don't sacrifice one for the other.

Harpreet Sahota: [00:53:35] Speaking of music, what song do you have on Repeat right now?

T. Scott Clendaniel: [00:53:40] Tubthumping by Chumba Wamba.

Harpreet Sahota: [00:53:44] That's a good one. So how could people connect with you? Where can they find you?

T. Scott Clendaniel: [00:53:51] Please follow me on LinkedIn. I kind of maxed out connections, but I am very active in LinkedIn. I am constantly on there trying to share articles, tutorials, free resources. I have really made an effort to try and spread the tools to Data science community and I hope to do that.

Harpreet Sahota: [00:54:12] Awesome, well. Scott, thank you so much for taking time out of your schedule to chat today.

T. Scott Clendaniel: [00:54:17] Very welcome.

Harpreet Sahota: [00:54:18] I really appreciate you being here.

T. Scott Clendaniel: [00:54:21] All right. Have a great week.