HH-27-06-2021_mixdown.mp3-from OneDrive

[00:00:06] Hey, what's up, everybody, welcome, welcome to the comet and our office hours powered by the @ArtistsOfData Science. I'm super excited to have all you guys here it is Sunday, June 27th. That's been a good week. Hope you guys had a amazing week as well. We should be live streaming on a number of different platforms. So if you see this on a live stream, for example, on YouTube or LinkedIn or Twitch, don't worry, we'll take your questions there as well. I'll also add a link so you can jump right into the Zoome room and join us. They'll be in the comments wherever you watch the streaming. But the I'm super excited to have you guys here shout out to the Dev, Barath and Poonam. How are you guys doing, man? If you guys any questions, go ahead. Let me know right there in the chat and we can go ahead and answer those questions. But while we're getting warmed up for that, I thought something interesting would do to do would be to to look at some. I posted something earlier on on LinkedIn earlier this week, and it was about the three biggest myths that they noticed Data science aspirants cling to while they're breaking into Data science. And I had a huge response from a lot of awesome folks. And I know I'm going to be breaking LinkedIn AIs rule. They say when you're on LinkedIn live streaming to not talk about LinkedIn, but to talk about, you know, I'm not really talking about LinkedIn.

[00:01:33] I'm just going to share my screen and pull up one of these, uh. Comments and stuff that I had here, one of these of the post man, hot 93 comments, such amazing, amazing responses on this. And, you know, there's some great, great takeaways. And Chisti here that a great recap of everything that she's picked up from all these other comments. And really thank you. Thank you for that. But [00:02:00] here are some big tips that that she kind of pulled out from all of these comments. And I really, really agree with all these. To understand the problem should be the foremost step right. Model interpretation is important. Excel is way more useful than we may consider it to be, which it is, and the importance of Data engineering. But so many great, great comments here. And I'm wondering, you know, what are some some myths that you cling to while you guys are breaking into Data science? If you guys want to share with me, go ahead and let me know wherever you are watching this, whether it's here or on one of the streams, I'll be happy to to chat about that, Biem, and I'm ready to take some questions. I did have Parrot Bunim. How are you guys doing? Go ahead and let me know if you guys have any questions. I think there's a bunch of awesome people watching on LinkedIn live.

[00:03:01] Hey, guys, I'm going to go ahead and put a link to the Zouma room right there. You guys are, by all means, free to join in. Um, I'll go ahead and put that here. And go ahead and come into the room. Some questions coming in on LinkedIn already. Can we talk about model deployment? If you have very specific questions regarding model deployment, I could talk about that button. If you want to talk about the entire pipeline of my deployment, that's a huge, huge, huge topic. So the more specific your question, the better. I may be able to help you shout out to everybody on LinkedIn Alberto joining us from Buenos Aires. This is awesome and seeing all these wonderful people there. But the link is right there in the chat. So by all means, join in and be happy to have all of you guys here. I think it'll be stollen enough in the chat here, so if anybody has questions, go ahead, let me know. I'll give [00:04:00] a shout out first to a two hour delay of or pirate or Poonam, no question is off topic, though I cannot guarantee that I had a sufficient answer for you. I'll try my best. Somebody seen the link is in text format, uh, I don't know what to do about that. I just copy and paste the link and hopefully you can click it. Um.

[00:04:24] But. But yeah, um.

[00:04:29] Nobody has questions. Everybody is just silent in the room, the guys sitting there watching on LinkedIn come on in. I'm happy to have you guys here. Uh. So I do have a question or a comment here, I don't believe in silver bullets, but if there was one thing that you saw progressed your Data science career, what would you say it be? That's a really good question like that. And I would say the one thing that has helped me the most personally was just adopting a growth mindset. So growth mindset being that work that Dr. Carol Dweck from Stanford did, I remember I mean, I wish I would have been introduced this concept much earlier in life. I didn't come across this idea or notion or concept until I was in my mid 30s, probably 33 or probably 35 by this point, 34, 35 at this point when I heard about the growth mindset and. You know it. It's such a powerful tool. It's just a powerful belief system, powerful way of thinking that, you know, there are hard things out there that you have to learn to do as a Data scientist. But if you know. That you can learn them if you put in the effort, if you put in the work, that that knowledge can be attainable, then just become so much easier. That sense of imposter syndrome for me kind [00:06:00] of dissipated after that. I was like, oh, OK. Well, obviously, I'm learning things that are difficult. They're not easy. They're not meant to be easy. But I know that if I put in the work and I put in the effort that I can understand it and use it and implement it. And I think that would be the one thing that progress my career progressed. My entire mindset just progressed. Me as a person. Excellent question, though. What awesome. What do you think? And what's been something that you've, you know, progressed in your career or something that that that has really helped you move up?

[00:06:38] Yeah. The thing for me in my role, I mean, some of you know this book for my role, I'm not a data scientist. I'm the head of community comics. I'm focused on content and sort of that community marketing and communication. And I think just learning, learning and other people's language and really listening and using that language when I'm speaking to them. So that can be like internal stakeholders at a company or like people in the community that I'm working with. But being flexible in how I how I learned how other people and how other groups speak and think and then trying to be empathetic to. So I think it's sort of it's it's that openness to people seeing things differently than I do it just for my career in a creative space and a little bit more of that sort of space. That's super helpful for me to just just take that in. And it helps me now. And that comment, sort of like what we're doing with community, is quite new over a comment. And so I think that's helped me sort of build this bridge between the rest of the company, whether it's sales team or the product team or whatever, and build that bridge between what we're trying to do with community. And this is learning how to speak all these different languages. So, you know, I think that's been part of my growth mindset is to like step into other people's shoes and use that from my past. That I used to be is a mental health counselor for a brief time and learning, learning [00:08:00] those empathy skills and super, super important for me. And it's my job is very, very people centered. So I think that's one thing that I've learned the most

[00:08:08] Absolute love that and part of this you got a question here in the chat if you want to just on meet yourself, maybe even turn on a camera, if you like, and and ask that question, go for it. Otherwise, I could just read it out. But, you know, having some participation is a lot more a lot more fun. But had to just, you know, while parathas, getting ready here to ask this question, I think they have hopefully found that useful. So the book I'd recommend, there's a couple of great books that can really help you with adopting this growth mindset and learning how to learn. I guess those two kind of things go hand in hand, like the things that really help me progress. My career was adopting that growth mindset and adopting the belief that on a long enough time horizon I can learn anything in that time horizon isn't really that long. It's just effort intensive and I can learn anything I need to learn. It doesn't it doesn't matter if it's brand new to me or not, and just learning how to learn. So a couple of great resources I can give to you. One of them is absolutely free. Dr. Barbara Oakley, she did the most popular online course in history on Coursera called Learning How to Learn.

[00:09:14] She's also got a book in mind for numbers. And later in July, you'll see her on my podcast that was able to bring her on and record an episode with her. So definitely check that resource out. And you know, the book Mindset by Carol Dweck is also a good one, as well as Limitless by Jim Quick. Those are all great books so far. It doesn't look like Pirate wants to himself for his question. So I'll just ask it right here. What are the typical challenges that you or your organization faced in model deployment once you did validate a model on some curated test Data and had it up and running? Any specific example? So for me, challenges were I [00:10:00] mean, I was the first data scientist this organization ever hired. And it's kind of a legacy company, right? Some manufacturing organization. Nobody knew anything about machine learning or Data science and organization. And it was just me and one scrappy, uh, software engineer who had interest in machine learning, trying to put this thing together. So the challenges for us mostly were just technical architecture and infrastructure type of challenges, like getting the model, obviously finding good Data, getting the right Data model, developing features.

[00:10:36] All those typical challenges existed for the model development process. But the deployment was just figuring out, OK, how are we going to like how are we going to serve? This thing is just a regular API. OK, great. If we have just the regular API, then, you know, the model is anticipating the role vectors to come into a certain way and those role vectors have to go through some transformation. Great how we can do all that. Right, because we used raw data to, you know, create a Data model, do some feature engineering, build the model. And now the model anticipates whatever Robeck is going to predict on to be in a particular type of format, particular column, so and so forth. So, um, how do we handle that with new requests coming in? Because they come in, they've got to be transformed. They've got to get served. The model models spit back a prediction, then integrating that into the product. There's numerous, numerous challenges. Um, so hopefully that gives you an idea. Poonam, I see you have a question here as well. Uh, put them go for it, if you wouldn't mind. Just a meeting yourself and making this more of an interactive. Conversation, I'd love that.

[00:11:48] Sure, thanks for this opportunity. My question was, you know, a lot of times the content in Data science gets too technical for the non-technical [00:12:00] people who are just trying to absorb the subject. So my question was, are there any tips, any ideas, you know, which would help us make the adoption better and our application in the real world practice?

[00:12:16] You just have to practice as much as possible because just kind of reading and watching, like videos and stuff like that, that's very passive type of way to study. So as much as you can try to. Do hands on stuff as much as possible, right? Really try to go more for practical type of examples, right. Try to find connections to what you're learning in the real world. Try to read case studies of how people have used this particular thing. You know, for example, you're studying this to say, I don't know statistics. Right. OK, great. How was statistics used to do a particular thing in a particular company, in an industry and just read as much case studies and then try to recreate those on your own? That makes sense. I don't know if I'm missing the question here or not, but it's just practice that you had. The way you absorb the technical content is just through practice and repetition and a lot of it and finding connections to things in the real world as much as possible, because otherwise it's literally just a Greek on a page, right? I'd love to yeah, I'd love to hear from my LKA on this as well, um. A couple of things I love to hear from from LCO and also shout out to Matthew Plaza, so LKA to to answer this question here. I'd love to. So when I pick your brain on two things, one of them was poolrooms question. One of them was part of this question, which I could restate for you, um, in a minute here, but wants to know how we know any tips for people from a non-technical background and now doing the formal course in Data science. How do they absorb all the technical [00:14:00] content? Do you have any tips?

[00:14:02] Ok, so yeah, I guess is this the question that you were answering with your practice? Because that would be thing say to. Yeah, so you are practicing and not being afraid to to do your own kind of challenge, not just follow the example from any cause or book that you found, but actually like at looking for Data says that you think are interesting or that you have come into contact with earlier on or even your or your own Data. Maybe you're like somebody who likes working out and you track it. Then you have this Data and start working with this and like because you will always come into challenges that are. Different from the ones in the courses, because in the courses, everything is prepared for you and you have this nice looking datasets and if there is something that is not prepared and they guide you and take you by the hand through all of this, and if in the real world this will not be the case. So I guess that that's what I would say. And then don't be afraid to use Google and Stack overflow and reach out to people if you have questions and just. Yeah, don't give up easily. It takes a while to to get through all the technical stuff, but it's worth it. So just practice.

[00:15:20] Awesome. Go for it.

[00:15:22] I also think one thing that can be helpful when you're learning new concepts generally is to sort of create your own glossary or index of those things and sort of you know, they take the time to like hand write out or type out the definition and and then connect to those real world examples that you're working through in those projects. And that practice is you're actually able to instead of just passively reading those concepts on the page, like I have a glossary that you have technical concepts that you want to go deeper into and then try to, as you go through your examples of practice, try to connect those things back, because it's almost like [00:16:00] you can you can take those technical things that you're struggling with and put them in a place that you have a resource now that you can kind of go back to. It can look different as a glossary, but a way to sort of like take those out of the context of like, oh, it's just so much, you know, in this one thing, it's sort of like, oh, that a little bit. And then go through it bit by bit or connect it to your real world practice. I think that can be super helpful and something I've done with like non-technical things that's been helpful for me in the past.

[00:16:25] Yeah, like that. Almost like setting up, like you mentioned, like flash cards of some sort. So spaced repetition is a good technique that I've heard people use so that that could help you kind of just get exposure to material. There's, um, there's a lot of great software out there. I think got one of them is escaping me. I'm sure it'll come to me, but but it's just type in space, repetition software and and that's phenomenally helpful. Mozart or any of these responses helpful or is just kind of already something you knew or you're looking for something different here.

[00:17:00] So definitely this helps. I mean, of course, we are aware about a lot of it, but I think this is, you know, seeing is believing. So you do it yourself and Harp and so on. And that makes it the best experience. And I will definitely try and incorporate this in my day to day practice. So I wanted to thank all of you.

[00:17:21] Awesome. Thank you very much for joining us. Maybe, maybe, maybe you hang around and listen. If you have any other questions. Back to Elche. There's a question here from from authorities asking about typical challenges that you or your organizations faced in model deployment. I was giving him an example of just being the first data scientist in a company that had never done any type of deployment of, you know, machine learning models into production. For us, some of the biggest challenges were just the the just the architecture and the infrastructure. And how do we serve it? How do we accept requests and things like that, or do we do it real time or are we going to match it? How are we going to do the inference? What are some challenges that you faced?

[00:18:01] Ok, [00:18:00] so maybe first there's context in the company that I work for, we don't really deploy the models, but rather we deploy a model building engine so we make an engine that can build models on in our case time series Data. And then. Because this Data is so, yeah, it tends to change really fast, because it's literally happening over time. You need to retrain the models very quickly. So what we do is we deploy that engine and instead of making a request to your model, you make the request to the engine, which builds a model on the spot and then applies it to your Data. But other than that, I think the challenge is quite similar. So we try to look at what the customers would be using or the people who use the the models would be using in software already. And typically this is like B.I tools like leaking all tricks and well, Ultragaz is more of a automation tool then. But like what else is there like power by and Tablo, these kinds of software and we try to integrate with them as much as possible. And also we do this through a rest API that we offer and then the rest API is also exposed. So that means that if if some customers use their own in-house application, they can easily integrate with that. And so I think it's looking at the specific industry that you are trying to work in order specific type of data and then see what do they use already and how can I integrate with these things so that I don't have to be yet another tool for them to use. But I fit easily into the process that they already apply in their day to day.

[00:19:43] I like that a lot. I was actually joining Vinicius this session and he was talking about as a data scientist, we should not become just technical commodities. We should be integrated as part of the business strategy and become strategic partners for the business. And you really echoing exactly what he what he was saying. Shout out to to [00:20:00] our friends, Comet Emelle for getting us set up on this. I know you guys have a lot of questions on model deployment, what to do with once the model has been deployed to production. Let me just go ahead and share my screen on the show. You guys, something real quick. And also, there's like 30 of you guys watching on LinkedIn, but there's only eight people in the room. Guys, you guys are more than welcome to come in the room. There is a link right there in the chat for you. But Comet Emelle has some really awesome stuff. One of the hardest parts about getting started machine learning is like, what the heck is a programmer to do? How do I test high parameters and notice what impact that my models and so forth? This experiment tracking tool is absolutely phenomenal. I definitely recommend checking it out. If you go to resources and you just look at their blog, it's super easy to set up. It's free. They're doing what GitHub did for code, but for, you know, machine learning teams, they also do a model production monitoring. So this helps you monitor your models once they have been in production. There's a really cool tutorial they did recently, and it was exploratory data analysis with Sweetwaters and Comite, I'm sure asking if you can find a link to that.

[00:21:11] Just definitely go ahead and then hook us up with that and I'll share that across all the different platforms we have here. But, yeah, definitely an awesome tool. You guys check it out. I highly recommend it. In fact, I use it at work myself. Um, also a couple of questions here coming in from LinkedIn. Uh, friends, if you guys got any questions here in the room, please let me know. I'll go ahead and agree to the Q and for the thirty some people on LinkedIn, you guys got to join in. All right. So a couple questions. Ones coming from AJ is Static's plays a major role in real life Data science projects. I'm assuming by static's you mean statistics. And yes, it does look like it's the foundation of pretty much everything like Data [00:22:00] science machine learning is essentially just. Linear algebra and statistics applied to business problems, so, yes, it does play a huge role. Question here from Alberto, your thoughts on the importance to having future stories? I think feature stories are an awesome concept. I haven't had much of an opportunity to play around it to myself just because I just haven't tried to use case for them in my particular industry or rather my role. Um, they are something I really, really want to look more into. I know there's a cool, cool product out there called Feast, I think for feature stores pioneered by Gosuke, which is one of my favorite companies. Um, but yeah, I don't really have too much of a, you know, opinion on on the importance of feature story. I just don't know enough about the Melco. Is there something you've worked with at all?

[00:22:54] Certainly, no, it's one of those things that's on my radar for a while, but I haven't had a chance to dove into it and learn about it.

[00:23:01] So I definitely I mean, if if anybody wants to come in the room and talk to us about feature stories about me, you're more than welcome to I just don't know enough about them to really have an opinion on them. Um. All right, looking through for questions here, no questions on YouTube, no questions on Twitch LinkedIn you guys are alive. I see you guys there. Thank you for joining us. Javed wants to know the impact of Data science and block chain technology or block chain and data science. Can these two technologies together come into play in the future? I think they can. Give me one second. I'll show you what I'm interested in right now. What I've been reading, uh, a book that I've been a. Getting into his block chain Data Analytics for Dummies, because I am one, but definitely recommend checking this out if you want to see the intersection of block chain with Data science, my good friend Carlos Mercato, you should follow him if you have not already. Carlos Mercado is [00:24:00] is huge into the blocks in technology. He's doing a lot of interesting projects as well.

[00:24:07] So definitely check his work out. I mean, there's definitely potential for intersection there. Absolutely. And the sooner you can get on that, the better off you're going to be. Right, because here you are combining skills right now that are both valuable and unique that not a lot of people are putting together. And if you can come together at that intersection and master it, you're going to set yourself up for some amazing success. So I would say if you if you personally have an interest in both Blockin and Data science, study both of them and then find the intersection and then do something about that intersection, which is, you know, maybe read a book like this. Maybe follow Carlos Mercado, maybe check out some of the LinkedIn courses from Jonathan Reichental and tell he's got some amazing content there as well. So hopefully that was helpful to you. Charvet shout out to Toure. Haven't seen Toure in a very long time. Toure, cheers. Looks like you're having a good time and a nice looking bar of some sort. I mean,

[00:25:13] It's been a while. Sorry, I've been working for, like, what, five weeks now? Great. Today.

[00:25:19] Wow. 18 hour days.

[00:25:21] So now I'm back on the free time again.

[00:25:25] And you moved out to. Did you move to like Northern Europe, right?

[00:25:31] No, I'm actually living in the south now in southern France. So I figured out that 27

[00:25:37] Degrees and the cold beer, you know, at my local bar

[00:25:40] And such and such a difficult life, my friend. Oh, it's rough to do it. Somebody has to have somebody has to write down and shout out to Matthew Blah's and the house. Good friend of mine, Matthew Belleza. Yeah, right. I met when we were still taking more questions. Thank you guys so much for being so active on LinkedIn here. Like I said, you [00:26:00] guys got to come in and join us. Man, if anybody else in the chat has questions, please let me know. Shout out to Asuman. Good to see you again, my friend. It's been a while Somnath. We got uh well bonamassa a question already, but hey if you got more questions bring them on. And yes, so Robert Robinson in the chat mentioned the software that I was thinking about, so Poonam, this is called it's called On Keet and Kaixi and this is the software for space repetition. So definitely check that out somehow, my friend. How are you doing? It's been a while. Doing good. Doing good. Hi. It's been a while. Yeah, man, how you been? What's he been up to?

[00:26:45] Yeah, so I was gonna ask. I recently finished my Data science boot camp, and that's why I was so busy last month and finally finished it, Capstone and everything. So I was wondering like, what next? What do you got? Well, I'm looking for a job or opportunity. And in the meantime, to take a bunch of. Mathematics and classes and all that stuff, but in the meantime, what do you like after after is having a Data science certification sort of what would you suggest how to proceed or what should I look for?

[00:27:26] Projects, man straight up. That's like the number one thing you get to do is put what you have learned in these amazing boot camps and things like that and apply them. In real world projects like that, you've got to do that. So have you been doing any of that?

[00:27:48] I've been I've been doing I actually I I did I did apply some of the some of the material to my previous work related stuff, [00:28:00] and I did some blog about that as well. Mm hmm. Uh, yeah. Yeah, I'm trying to do I try to apply and I have several projects on on the pipeline that I am trying to take to the next level. One of them at a time serious. I just asked a question to ask about that and then and then some of the deep learning stuff. But yeah, yeah. I'm looking forward to add more projects to the portfolio.

[00:28:29] Yeah, definitely. I mean, I mean more than doing like the fancy cutting edge algorithms. I'd focus just on having a really, really well structured project, making sure that the repository structure is nice, clean professional, using something like cookie cutter data science as the template for the repository structure that just gives a level of polish to it. Right. And it communicates something to potential teammates and potential hiring managers like, oh, this guy gets how to structure his thoughts and gets how to organize code. Right. Uh, so that's that's huge. And just using. I mean, just something as simple as making sure your your code has comments, make sure you're using dark strings, just anything that you can do to give yourself that professional edge by projecting an aura of professionalism. I think that is much more important than any particular algorithm or any particular data set is just communicating that you have what it takes to be a science professional and you do that through some of the quote unquote little things. But there are huge things, right? We talked about repository structure. You talked about commenting, documenting code.

[00:29:41] Let's talk about using helper functions. Right. Let's talk about having a clear entry point into a project like this is where you start, where your next is, where you go after. That's where you go after that. Well, documenting your thought process. OK, I was doing this thing and I saw this. Therefore, I think this. Right. So [00:30:00] I know it's kind of generic, but just like, you know, it's one thing just to have a graph out there with an unlabeled X access, unlabeled Y access, unlabeled graph itself and then have the reader try to interpret what is going on. Like, that's not a good, good look. Right. Make it easy for whoever is reviewing your project, understand the importance of everything that you're doing. And you do that by just almost like a brain dump. Right. Just clearly stating what it is that you see why it's important and how it's impacting your decisions going forward in your project is not helpful. I don't know if you're if you're asking the question about how to do a project. I just went off on a tangent. I didn't anyway.

[00:30:41] No, that that's helpful, actually. So that now reminds me to go back to back get up and try to see if

[00:30:49] You have anything I can do. Yeah, you'd be surprised. I just use these touches right. Like that. This is what it takes. I need to look like a professional with the projects. Right. It doesn't like it literally doesn't matter which algorithm you use, doesn't matter which dataset you use. What matters is the execution of that project and how professional you can make it look. Right. Because you want somebody that's either a hiring manager or a potential teammate to come to your project and be like, oh, damn, do this guy this guy works like I want to work with this guy. Right. I like how well done he did this thing like, oh my God, working with him would be a dream. I'm not going to have to have nightmares about cryptic code that he's writing because he documents everything so well. It makes the project stand out. OK, any tips or or, uh, for for four, we can stand our project and shout out to everybody on LinkedIn for some, somehow I managed to block myself from commenting on my own LinkedIn post. I don't know how I did that because I think I accidentally hit mute on myself. Um, but I can I guess I can be myself. There you go, guys. Come join [00:32:00] us. Come join us. OK. Any any tips on projects.

[00:32:05] Um, yeah. Do do things that interest you. And and I think also when you get better at it and do more complex projects, then still you have to be aware of the value of the earlier and maybe in your own simple project that you did, because that's that's a mistake that I made in the beginning, like thinking of a project that I did two years earlier that was like very simple. And it's not at all technical in terms of the techniques that I use. So why would I think about this one? But sometimes that's what people are interested in if you're trying to get a job, because sometimes simple is enough. So even if it's something that, you know, you can do better still you can think about it, you can even just say so like, oh, this was a fairly simple project and I learned about it. And even now I know that I would approach it differently. But back then I did that. And so just because you're further along, don't let go of what you did earlier. I think that's something I can add.

[00:33:05] Yeah, I like that. And simple, like, if you can come up with a simple solution to a tough problem by reducing how complex your solution is, that just makes it look so much so much more intelligent as well. Awesome. Go for it.

[00:33:19] Yeah, I think something you were saying, Harpreet, about this made me think of a more general sort of point about that detail oriented focus around your projects in a more general sense. I think there's a common thing, a thread that I see in a lot of either whether its content creators, people who write technical content or do these projects is that there's this sort of 80, 20 thing where you do the projects. You get maybe the first version of the blog post out, you've got all your code or you're just kind of all over the place. And that's the 80 percent of the work. And it feels like, OK, I did it. The model I trained, the model, ran it, ran it through all the sort of steps that I'm supposed to. And now it's like I'm done. And [00:34:00] I think what really separates people in most fields really is there's that 20 percent extra of taking like, OK, now I'm done. But like, what else can I do to optimize this? What else can I do to better communicate this? How can you trace the lineage of what I'm learning and communicate that all of those pieces are that that 20 percent that like you're on to say you're on like a you know, I just drove cross-country in the United States not too long ago. And it's that sort of like on a 10 hour car ride, you know, the first eight hours you're like, I'm motoring right along. And then those last two hours are just like so terrible and feel so long.

[00:34:33] But you've got to learn how to do them. You've got to train your muscles. So for me, for instance, it's like, OK, we wrote the content, we wrote the blog posts, like, now what's the best way to distribute that and how do we push that out, that extra 20 percent as opposed to just finishing the going and be like throwing up and saying I'm done with it. And I think like across different fields, across different ways of executing like that extra 20 percent of wrapping that in the packaging of like, you know, whether it's the process or the code in the documentation like that extra 20 percent is what's going to put your over the top of people. And that's of detail oriented mindset. It's going to be because, you know, there's probably hundreds of thousands of people who can run these models, train these models and get them to work on a validation set or whatever. But I think that's that extra 20 percent. That's the Harp is sometimes the hardest part and the most grueling part and the most. I really got to share this document this and clean this code up, but you got to do it. I think that's like I've learned that more than anything. That's what separated me in my career, in my field, is that I'm willing to go in the muck and do that extra 20 percent. And so you might it might look just simple and easy to people on the outside, but it's like it's super important that that gets done.

[00:35:41] Yeah, absolutely. Man couldn't agree with that more so human. I'm I'm hoping hoping to get some good takeaways here, my friend. And I mean, looking forward to checking out some of your projects. Man, don't be afraid to to share your screen, pull it up and then show us what you got. Be happy to look at it. Shout out to everybody on LinkedIn. Still joining us, AJ [00:36:00] Verma is asking, what is ensemble learning? I mean, you could also put that into Google, but essentially just stringing together the results of multiple models to come to a prediction value. Right. So if you got a number of classification algorithms that you're using and you want to, you know, essentially do this like let's say you have a classification problem and you want to do some ensemble learning more than you can fit the different models. Right. Maybe logistic regression, random forest SVM and then take the majority vote of those classifiers to serve your final prediction. Right. You can think of the analog for that in regression as well, where you average out the values to sort of the predicted value. I don't know if that's a trick question or not. OK, what are your thoughts on the sample learning? Shout out to Dave Mangalyaan. Good to see here again.

[00:37:02] Um, yeah, I think you said it exactly, just multiple models, maybe different techniques, putting them together or the results that come out of them by some sort of gathering and then using the results that comes from all of them, maybe as a way of reducing the risk of one model, typically making the same sort of mistake in another model might not make that type of mistake. So, yeah, yeah.

[00:37:28] This example I work where we have we're working on a regression problem and I, I essentially took the results of three different models, averaged them together, and that's what I'm serving as my prediction. Right. And you know. There's multiple reasons for that, I think, OK, highlighted some some really good ones, but, you know, we're we're one model might be overcompensating on value, another might be under compensating. And, you know, it just it whether you use [00:38:00] or narcissism depend on your use case and depend on if it's going to help you get a better prediction or not. But I mean, at a high level, that's what ensemble learning is. Um, looking for any other questions in the chat here, I don't see any other questions. Um. Dave Heideman. Doing great. How do we spend?

[00:38:26] We just focus on finishing the projects like AIs, it's my first project on NLP, and it's quite challenging.

[00:38:36] Yeah, what's toxin's about that? Like what's the what's the problem statement you're attacking

[00:38:42] Like it's like giving up tax or this kind of text, like a book or just part of a book. And then the end result will be like what their meaning. Is it gender bias or not. What's the presentation of the male or female at that?

[00:39:06] That's really interesting. I like that that's that's a very interesting problem statement is this for a personal project or a project? I work,

[00:39:14] I like it's a work.

[00:39:17] Nice, that's cool, man. Sounds like a fun thing to do like that,

[00:39:22] Like we're now in the like in the final stage or in a stage like visualization and deployment like like that. Because are just done with the pipeline, although the pipeline is not that. And I really is strong, but. And we just like. However weak, and the secretary of the pipeline that would be there just straight through, just until the end result would be just three or four months later.

[00:39:58] Yeah, I don't know. That's [00:40:00] cool. Shout out to Matthew Belleza. I saw you there and he disappeared. How are you doing, man? Good to see you. I know. I know. I know. You join in on some happy hours and some some officers and stuff like that, but I never get a chance to actually talk to you except when it's the comments and stuff. I mean,

[00:40:18] Yeah, really good. Just been cranking a lot lately. It's been a big learning curve for the current job that I'm in. So, yeah, we've we're using a lot of like in cloud stuff. So it's a lot of trying to rationalize our Data models for our legacy system. And then my great Data so they keep you busy on the weekends.

[00:40:38] Dude, I've got the same challenge going on at my current company. I mean, what are you looking into in terms of tooling on on Azure to to to work on some of the stuff you're doing?

[00:40:49] Well, most of us were just beginners at this point. Most of us, whether it's like the Data and the engineering and the engineering team or like the defense team that I'm with, we're all used to, you know, SQL server building out the triggers and all that sort of stuff. So it's been kind of a learning curve, just trying to learn this part. SQL and all this stuff mostly right now is just trying to get in Data models like migrating our Data models from a server side to the cloud and getting used to try to update the data governance like that in purview and whatnot and keep the minute they get the Data lineage going. So we're not quite yet at the point where we can make percent, where we can operationalize our models or not. But we're getting there. We're getting there. It's a process.

[00:41:33] Yeah, sounds like identical to the situation I'm dealing with. I work because I came through and, you know, developed a machine learning model put into production and it's doing great. Other parts of the organization want it. But then we're like, we have no good data architecture or infrastructure or governance or quality in place. So we're also looking to Azure purview for for governance and lineage and things like that. What are some of the challenges that you've been I mean, you talked about a few here, [00:42:00] but in terms of you said you mentioned working out like know legacy systems, moving to the cloud environment, like what are some of the cultural challenges you faced with with people in your company?

[00:42:13] Well, the cultural challenges is, of course, when you're always moving from legacy systems to like a new like system, people like are invested in the old order. So they're just like, OK, it's not that they're stubborn or anything like that. It's just this is how we've done this for the last five years. I'm used to it. This new stuff is scary, but the most of it is trying to. I spend most of my day in the Data science team, spent most of our day trying to explain what we're trying to do, trying to get the business requirements for them and not just trying to do that, but also explain to them the values. So, I mean, some of the stakeholders do have times where they're asking us, hey, you know, we want you to answer this question. So we go back to the system and we find out, you know, there isn't really tables or a database or an injection where with logic has built up for that. So most of the time I'm using what most time is is just trying to be able to translate this out from technical requirements to requirements that the business can understand and question us on and keep us accountable.

[00:43:13] I like that, and I'm definitely going to have to connect with you offline at some point so we can, yeah, we can chat more in depth because these things are challenging, especially for someone who's just got a traditional Data science background moving into a position where we're doing Data strategy, Data management. And it's a whole different world. And it's not easy if you're not familiar with it, because as Data scientists like we're end users of Data, I like to work on the stuff upstream. It's yeah, it's different with a bit of a challenge for you. Like personally, like, you know, because I know you also have a background in mostly traditional Data science type of stuff trying to adopt the mindset of working [00:44:00] on data governance. That's not as sweet and sexy as machine learning is.

[00:44:05] No, no. It was kind of a challenge because I do a lot of consulting on the site before. It's just like the consultants usually are not big, big, big business, especially. So you tell them, hey, I want to cluster this and I want to segment customers. OK, sure. What do you need to do it here? It's more like do we have the time and resources because the tickets are accumulating to and to like a very high level. And do we have the time and is it worth money to do it? So the main challenge is just trying to even make sure that we even got the definitions down because a lot of the Data quality was so so all over the place and it wasn't 100 percent clear. So mostly it's been just trying to make sure that we find definitions or analytics terms that the business can agree on and then also be able to translate that business logic into the contracts and trying to meet the requirements of the engineers who always want to make sure everything is right. I get it. I have a programing background and in the business who says why I'm out getting my analytics on Tablo. So it's a challenge just trying to balance the two.

[00:45:10] Yeah, a very challenging. I can attest to that. Thanks for thinking about you. Rodney is asking, would maturity votes fail under circumstances, certain circumstances? Yeah, I guess it would fail when you have an even number of models that you're taking the majority of because, you know, you probably would need an odd number just to ensure that there's a majority. But, Rodney, you're more than welcome to join in. I know you have a link to the we have a link to the actual Zouma with my friend, so please join us. He said he's going to join us after coffee. So hopefully the coffee is done, my friend. Matthew, also, like you're one of the consider you one of my brothers in the philosophy because you're always commenting on my philosophy posts or or things of that nature that in the post. So let's [00:46:00] talk about the the I guess, the importance of being exposed to topics outside of Data science.

[00:46:13] Well, I mean, I see everything kind of as like there's no really boundaries. So, like, the stuff I learned from when I was in the military, the stuff I learned from when I was doing marketing stuff and then a consultant and then the stuff I learned from philosophy stuff, there's no real line in my head. There's just like, what can I use from here that I can apply to here? So a good example, like I learned about Data recently and I said, oh, like when you were talking earlier about trying to understand information. I said, as soon as I hear something from like you're here or some somewhere else, I find a notion, my notion, and then I just dump it in there. And then I just use like Lake. And then later on I go back and stand. But for me, it's most fun. For me, the philosophy is very important. It helps keeps you centered. I mean, it in difficult situations are really trying to balance the ideas of the business. It's really helpful. It's like, OK, you're going to have to sacrifice one thing for the other. How do I do this in such a way that retains my credibility with one of the parties and keeps my credibility with other parties and still gets the job done and delivers value. So, I mean, philosophy is absolutely essential. It's the core of like trying to understand what you're doing. It's I mean, otherwise you're just doing the process. I mean, my reasoning process is you're just saying, OK, I'm going out. Great, great. But the reason is it's like, are you considering it in the larger business aspect? And the philosophy helps me consider that not just in terms of the Data we're down delivering the business value, but also there's also a political element to as well.

[00:47:44] Man, I love that, I love that. Thank you very much for that insight. Shout out to Rodney Berard, the beardless Rodney Biard in the house. So, Rodney, go ahead, talk to us like about ensemble models.

[00:47:58] So this is just something I've [00:48:00] been thinking about is when you use in ensemble models, when you use majority rule type. Things to determine what what prediction you're going to go with, there seems to be a connection to social choice theory and and in social choice theory, you get breakdowns of some of these voting rules. And I don't think that's something that machine learning people have thought particularly deeply about. So that's that's where that question was coming from.

[00:48:35] Yeah, I remember I actually did bring this up in previous office hours a few weeks ago. Social choice theory. I didn't get a chance to dig into that. But I mean, talk to us about that at a high level, so.

[00:48:50] So in political systems where you have voting, right, you get this result. It goes back to this French aristocrat in the 18th century called the monarchy to condo say, and he points out something called the voting paradox, which is when you have more than three or three or more alternatives, that majority rule can file basically. Yeah, so. And then more three, three people. So if you've got. You know, yeah, it's so more than more than three underlying models and more than three alternatives, you should potentially begin to see problems with majority rule. But I've never seen any work done looking looking at the question.

[00:49:40] Yeah, it must be something that I went to read this book as models of cooperation, what they call it, and then collective action problems in this book, The Model Tinker by Scott.

[00:49:52] Yeah, I haven't seen that yet.

[00:49:54] So it's an excellent book, which, by the way, you can find his conversation with me on my podcast as well. But yeah, he [00:50:00] talks about, uh, something similar to what you what you're discussing. Yeah, it's slightly

[00:50:05] Different to the collective action problem, but. It sort of goes in that direction, I guess it's something there's something I've noticed that there's these results in mathematics on on these voting paradoxes that the ensemble learning people just don't talk about. And and one would expect that when you're using majority rule that you're going to run into issues under certain circumstances. So so the question is, have people run into those? When when doing that and the question is, has anyone seen any literature on this and in like published literature or even working papers

[00:50:51] That say, let's change that, Rodney Lewis. Have you published something on this man? I read.

[00:50:56] I'm sick of it.

[00:50:58] Oh, yeah. So, uh,

[00:51:01] Yeah, this is this is a great book, Modern Thinker. It's just all like this is a book about just a ton of different models across different disciplines. Super, super interesting right now. Yeah. Uh uh, who just. Yeah. Ensembles. Also in the book, Dave talked about that you had such a good, good book. He talks about all sorts of stuff like systems, dynamic models, game theory models, uh, power law distribution, long trails, linear models, network models, broadcast diffusion, entropy, all sorts of interesting stuff in this book. Highly, highly recommend it. Um. Uh, game theory, game theory, super interesting, I'm fascinated by that stuff. Shout out to everybody else joining in. I don't see any questions on LinkedIn LinkedIn audience is waning. We're down to eight, nine people. So if you guys have questions, now is going to be the time to ask before we start slowing down and wrapping it up. And if anybody has questions right here in the Zoome [00:52:00] room, please, by all means, go ahead and meet yourself and and go for it. Um, don't ask questions here on. LinkedIn YouTube twitch. All right, well, hey, man, that's awesome session. Thank you so much for, uh, for for joining us, everyone. Um, somebody is asking, uh, if I am updated on trends in game three. Unfortunately not.

[00:52:26] Um, I just have a general interest in it. Like if I come across game theory, like I mean, I studied economics in, uh, in school. I really enjoyed it. And then game theory is just I played a lot of poker, so that's when I got interested in the game theory. And Rodney says he works in game three, which is awesome. Um. So, I mean, I'd love to learn more about a man like stuff is fascinating. I interviewed Kevin Zillman, uh, we talked about game theory on my podcast. He's got a series on Big Think. Yes, on big thing about game theory. So definitely check out that episode with him. Really interesting stuff. Um, last question here coming in from Jakov and Dehlin on LinkedIn new to the field of data science, is there any tutors available for newbies? Um. Don't know of any tutors available for newbies, if you're interested, you could sign up for my program, Data science dream job. Look that up, go to Daudi JoCo forward slash free dash training. Um, you'll learn a lot of great information on there that you're interested in the program. By all means, come in and join us. We've changed thousands of lives. I don't see any of the questions from anyone else. Anybody else got questions here. No question.

[00:53:46] Oh, well, let me I got to tell you about this book that I'm reading this week is kind of meta, but I'm reading a book called How to Read a Book, which is super, super interesting. My friend Jewelry's told me about this book. And, [00:54:00] uh, yeah, it's it's it's really cool because he'll tell you how to read just not only a book, but then he goes like, you know, how to read a practical book, how to read poetry, how to read history, how to read science and mathematics. And it's, uh, I I've been putting in some of the, uh, suggestions they talk about, um, in my current reading. And I'm already starting to become like. It's better I reading this book, aims to teach you how to read, not for the sake of reading, but teach you how to read for the sake of understanding. So really, really good book. Um, you can find a PDF version of this online. This book is like almost 100 years old, I think. Great. From like the 1930s or something like that. Um, so there's a free version online. You can definitely find that, but I recommend this one. Joe, thank you so much for, um, give me the recommendation on this. Um. Yes, OK, go for it. Question.

[00:54:55] Yeah, so it might not be very technical, but I noticed for myself that I struggle with this maybe kind of a luxury kind of problem, but I'm interested in too many topics and I don't always know, like how to handle that or how to select which ones to go in or when to move on and when to dove deeper in a topic. Or like if I decide to look into something, how to keep track of all the other things that I'm putting into backlog, so to speak. So if anybody has any tips on that, because in my case currently it's just a gigantic Excel sheet with a list of things I still want to read, which is way larger than I will ever get to.

[00:55:36] So, yeah, I suffer from the same problem. Yes, I'm just interested in so many different things. How do I figure out what it is that I want to pursue? And it's just whatever it is that that I mean. So there's there's definitely stuff that you read or you're doing because it pertains to some work that you're doing, but then there's also just as being a smart person, intelligent person, you just have curiosities and different things. Right. I'll [00:56:00] just dig into one as I fancy it. Right. Like, and just explore it and think about it and try to find connections with other topics. Right. So that's one thing he talks about. This book is called Synoptic Reading, and it's trying to find connections between things that you're reading so that you develop a better understanding. But I just think if it's this stuff for your own curiosity, I'd say just follow it wherever it leads to. Like, there's no need to be attached to some completeness of learning it. Right. Like, for example, like, I don't know why, but recently I've been really into, like, you know, Sean Carroll's work, talking about many worlds and and parallel universes and shit. I just like I don't know, like it just seemed fascinating to me. And I started, you know, thinking about it. And I was like, oh, man, this is cool stuff. Interesting stuff. I'll probably drop it after a week and then, you know, move on to something else. Like for some reason I was really into information theory a couple of weeks ago and I was just watching this course on great courses about information theory. I was like, oh, this is fascinating stuff. So I just I just go wherever the interest take me, I don't really care about organizing it or things like that. I know that's not the answer you're looking for, but if anybody else has tips

[00:57:14] Like follow your guts a bit, just. Yeah, yeah.

[00:57:18] I like I like, um, Alan Watts. He has the same philosopher Alan Watts. Do things that are delightful to you. You thereby become delightful to others. So just do what you like man and look at, look into what you like and don't be attached to any outcome from studying that particular thing and just enjoy the the act of learning something new and trying to connect it with other things. Um. Right. Speaking on that tip of Sean Carroll, the podcast interview he did with Lex Friedman was really good because he started talking about artificial intelligence and quantum theory and things like that was just fascinating stuff. I mean, and for me, it's like with philosophy, like [00:58:00] I'm all over the place of philosophy, like I'll be studying metaphysics and then ethics and then going back to other types of things in between. Like I recently just got really into Taoism, like, you know, after being into stoicism for for so long, I started getting to Taoism is just fascinating stuff. Yeah, awesome. What do you do when you're, uh, pursuing many, many topics?

[00:58:24] The question I think kind of I kind of just the echo. What you're saying is I think like it's about developing trust in your intuition, something like this middle level, I think, and. Yeah, because I think I think that I think for me that the sort of the insecurity is borne out of like a mistrust of myself. And I think I've learned to trust my own mind that it's going to do that associative work. And once you what you kind of start doing that associated work, you like anything else, you practice it like you. You learn a bunch different things. You practice you you develop that associative tendency and then you can start trusting those connections that you're making. You can start trusting the way you pick something up and put it down. And you don't feel like someone else's ideas in a book are so much better than yours that you have to finish the whole thing in order to take something from it. I think it's like a mistrust and insecurity that is like super relatable and common. So it's not like a negative. It's just like something that I think we all have to work through in different ways. But I think it's that sort of it's that sort of self trust that goes a long way for me. Yeah. That sounds that sounds like a good point, because sometimes they need to feel like I need to finish something because I started it might not always be the best choice.

[00:59:43] Yeah, that's just, again, that attachment to the outcome, the attachment to this idea that I need to complete something that's just for I mean, obviously, if you're working on something and it's for work, completed work stuff, but if you're just doing stuff on the side, that's just to to gain interest. Like, for example, like [01:00:00] I don't finish all the books that I read. Like, sometimes I'll just get the main point without having to go through the thing in its entirety. I'm never just attached to OK, I have to read to the last page list. I'd love to hear from either Rodney or Dave on this topic. So, uh, by all means, if you guys want to to jump in, let me know. Dave, I see, uh. Put something in the chat, jump in between books.

[01:00:27] Sorry, Dave and I have been having a side conversation, so we're probably both distracted. Could you just bring us up to speed on where you

[01:00:37] Were at Data? It's just that, uh, elkies talking about just how do you manage all these competing interests that you have intellectually, how you prioritize what to go focus on next?

[01:00:50] Um, well, case in point, distraction.

[01:00:56] Yeah, that's true. Manage the distraction, which is hard when it's hard. It's taking a very long time to figure out how to do that for myself. Does not look like there's any other questions in the chat or here in the Xoom room, guys, we're here every Sunday, 11:00 a.m. Central Standard Time. The link to register is going to be right here on LinkedIn. But I'll shout it out to you. It is a bit liberty l y Ford Slash Comet Dash Emelle, Dash 08. Shout out to Comet Emelle for being so generous with with with this initiative and making it possible for us to have this space together. I'm looking forward to having you guys here, you know, more of you guys here and more big guys joining us on LinkedIn. And we had like 30 people at one point. And that's huge man like that. That's awesome to see. But that doesn't look like any other questions you guys do join us. Would love to have all you guys here. Um. Take care, guys. Have a good rest of the afternoon. Evening, [01:02:00] morning. Whatever it is, wherever you are. Take care. Remember, you've got one life on this planet. Why not try to do something big? Cheers, everyone.