comet-ml-oh-2-feb142021.mp3

[00:00:09] What's up, everybody? Welcome to the comet and our office hours powered by the @ArtistsOfData Science. Super excited to have all you guys here today. Man, I hope you guys have a wonderful weekend. Thank you for joining me on Sunday, whether it's Sunday morning or even Iraq and whatever it is. Thank you for spending part of your evening here with me and my friends from Comet and how we got Ayodele, my co-host in the house without Curtis', Gabriel and Koshland. Thank you so much for hanging out. Hopefully you got a chance to check out the Data Chedid conference. It was one heck of a event. I really enjoyed my presentation and putting in the the work to create that presentation. If you guys haven't checked it out yet, it is available on YouTube and. Yeah, and I'm excited to be here. I really. How are you doing. Oh man. Check, check. That hair looks looks nice.

[00:01:06] Had it. I felt the need to change things up a little bit since we're all at home. Right. Yeah.

[00:01:11] Yeah. I was gonna ask you how your week's been made. It looks like it's been been pretty interesting, this blue dude yourself or did you get it done?

[00:01:19] I did it myself actually a couple weeks ago. It's been it's been something I've experimented with since it.

[00:01:27] Is that something that you used to do back in the days as well? I change your hair color and stuff like that

[00:01:32] A little bit. Not nearly as drastic or as often. But I definitely think, like, uh, I'm like, what's what's the harm in seeing me there? I'm like, it was on a screen every once in a while, right?

[00:01:44] Yeah. Yeah. When I was in high school, I was like I mean, I grew up kind of an interesting area, interesting neighborhood where it was quite diverse. And I kind of sat on the edge between like, you know, like, you know, punk rocker and like, you know, just like a wannabe thug type of thing. And it's just because, you know, the area I grew up in, so always changed my hair, all these different colors and all this weird stuff. But I'm super excited to have you here. How the rest of your week been?

[00:02:12] It's been good. It's been busy, but it's I think it's always a sign of good things about,

[00:02:17] Uh, do you learn anything new, do anything new this week

[00:02:22] Or. The biggest things I've learned? I'd probably say I spent a lot of time like reading about envelopes, actually. So it goes hand in hand with, like, everything comment. But I'm just trying to get a better understanding of all of the tools that are out there. We cover certain areas and MySpace and other companies cover other areas like cloud Data, hosting and production and stuff. So it's trying to get a better understanding of the landscape of what Emulous really is.

[00:02:56] So for anybody who wants to go out there and learn a little bit more about Mellops and kind of get their head wrapped around what it is that this thing is, what's a good resource that you came across yesterdays.

[00:03:08] Yeah, first thing there were a couple of medium posts that I thought explained it really well. So if I can, I'll I'll try and drop some of those links in chat so anyone can check those out. But I really started by trying to get what, like from end to end of the model building to production monitoring. What does a normal setup kind of look like? So I've done some of those down.

[00:03:36] Yeah, definitely. And I guess like when we think about what Mellops is, what is Mellops, right? Is it is it just like the portion where once it's deployed now we need to monitor stuff like model drift and Data drift and things like that? It is not the only thing that MLA focuses on. Or is it just the entire chain of events from ideation to deployment?

[00:03:58] Yeah, I would say it's more the entire chain of events and how each specific tools and frameworks. So for a lot of like large companies doing data science, you may have like a data lake or whatever their cloud data services is really everything from there, like pipelines to the production, ordering at least from everything that I've tried to gain so far.

[00:04:23] Yeah, that's such a fascinating area. Right. I think it's something that I definitely want to really up my game on and learn more about just because it's cutting edge in a sense. And I feel like there's opportunities to really make a positive contribution. Um, and it's just a signal that data science is just growing and becoming more and more important. And yeah,

[00:04:47] I think it's a key for like a lot of the reproducibility that, you know, it's not just data science to Emelle, but we've seen like problems in just industry science with reproducibility in to paper. So I think that's kind of our way of dealing with that somewhat.

[00:05:05] Yeah, because, I mean, people. People don't realize that Data size actually is a science, and for a science to actually be science, there needs to be reproducibility and falsifiability baked into whatever it is that you're doing. Right. So like having I think this is one step in really solidifying Data science AIs the actual sciences reproducibility. So it's it's definitely an exciting piece of the puzzle. I'm happy to be learning more about this. So, guys, hey, welcome to the comet. Officer, happy to see you guys here to see Quentin here. Quentin, I think we were connecting for you a LinkedIn and you're visiting, I think, from France. So thank you for for coming here. Hey, so let's open the floor up. If anybody wants to go for a question, go for it. And then while our community member is asking the question, if you've got a question and you want to hold your place in line, just type in. I've got a question right there into the chat. So open it up to anybody questions on anything whatsoever. Or I could flip it out and ask. I ask you guys, you know what? What have you been learning this week? Has anybody been trying to pick up a new concept or a new area of of skill? Are you guys struggling with their how you guys feeling with it? I think for for me personally, it's something I've been really interested in is, you know, I picked up this book that I had recommended, this database's book, and then I posted about it on LinkedIn and people were mentioning the statistics book and the linear algebra books. I've got them all and they're all sitting here. But this has been great databases, like just a great refresher and a really interesting way to to learn up on that concept. Curtis, I saw that you were unmetered. Go for it.

[00:06:50] Oh, yeah. So I've been doing a lot on also. I've been doing a lot on them a lot like one of my most recent clients is they do a lot with em, a lot. So it's kind of like a platform for model monitoring and stuff like that. So I've been covering a lot around a Mellops model management and yeah, what's mainly around like financial area. So that's kind of like a new sort of realm for myself where you can work.

[00:07:23] You learned up on it about this week. What can you share with the audience?

[00:07:27] Oh gosh. Um, boy. So that meant that the process is like going from concept, right. Until deployment and management and production and the whole loop of how and Mellops goes around and why we can't be just using it, why Mellops can't be done in the same way as Devil and the little discrepencies like we're also managing Data and code in Data science, whereas in tablets and stuff like that you're just managing code and. Yeah, yeah, yeah.

[00:08:08] I think it's the biggest thing about machine learning systems is the input isn't just code, it is the Data itself. And Data itself changes it because it's a representation of the real world in the real world changes. So we need to have those type of things built in to make sure that we have reproduced reproducibility and consistency.

[00:08:28] It gets especially and cool that you're exploring this in finance, especially since it's one of the more regulated industries than the vast majority of us work. And so have you experienced any, like, specific issues around Mellops or maybe finance specific right now?

[00:08:51] Hands on? Not not really. But I have been just going through because the company is based in in America. I've been going for different documents like R11 11 and different like model management guidelines or or requirements set by the government in the United States. So it's just like touching around things at the moment and also developing content for at the moment, not necessarily any hands on experience, but I've read about experiences that one that may come up in like a financial situation. Yeah, no, no, no.

[00:09:33] Do you have any experience is a

[00:09:35] Less formal related, but most amazing to me is a credit reporting agencies. So every individual in the company is that even technical folks had to go through training and pass a screening test, which is like the Fair Credit Reporting Act here in the states that consumer reporting agencies have to abide by, essentially. So the biggest impact I noticed on our technical workflow is that we had an. Incredibly heavy focus, uninterpretable models, so we were not in a work that really put any kind of box models into production

[00:10:17] At all, even if it's given better results than black, simple linear regression, for

[00:10:22] Example. Yes. So and because we are they were regulated by and the main issue there was that we had to really easily be ready for auditing. So be ready for all the extra and then essentially have a legally defensible decision. So less about if a black box model could be more accurate. But if we couldn't legally defend that decision, it was almost always opting for something interpretable like a linear model.

[00:11:04] So for people who are interested in going into the finance space or credit reporting space for Data science, what are some points that they should probably be aware of, like what kind of tools or techniques or specific types of algorithms that they should really focus their attention on so that they can make an impact when they get their first job in that that field?

[00:11:28] You know, I would say work on, you know, like everything about the go to industry and all models. So we talk about linear models that like generalized linear models and additive linear models and understanding. There's a lot of depth there. Same for like decision trees going kind of deep into a tree and a little bit of ensemble methods. I found those were the most commonly used and the easiest to kind of document and defend when necessary.

[00:12:06] So, yeah, so people would probably want to get some good resources with respect that he got. Like, besides your course, you got an awesome course on LinkedIn learning. Do you cover any of these other, uh, parts, these generalizing your models?

[00:12:22] Less so in the course of it is kind of an introduction, but I know there's a lot that I that I was using kind of to have in my back pocket at my last role, especially since that there were after taking like the exam, they're like personal finds that technologies can face if we do anything serious, I guess because we have data about people's personal information. So I know there is definitely some resources out there. I'll probably be able to grab one in just a minute.

[00:12:58] Yeah. So one of my personal favorite resources is the Pennsylvania State University. So Penn State University, their statistics department, all of their courses are pretty much open source. They're not like on video or anything, but they have all the lecture notes online. If you just search for Penn State statistics, you'll find the whole curriculum of courses that you can refer to and hands down. My favorite resources are so good, so definitely do a research into that. If you do like Penn State, quote unquote, Penn State and quote unquote gloms, I think it'll take you like statistics five or four or something like that. And it's a very, very good resource. All right, guys, so if anybody has a question, go ahead and admit yourself. I see we got a couple of new members here. Are you joining us? And good to see you here. Good to see you. I'm glad you're able to to make it. I know it's a bit more favorable time for you as opposed to the Friday sessions, which I think like 40. Misty.

[00:13:58] Yeah. Yeah. Thank you so much. I'm really happy to be here and be a part of this conversation.

[00:14:04] Yeah, that definitely happy to have you here. How's your week been. Yeah, it

[00:14:07] Was, it was good. I've been learning ah I've been learning the three so yeah I've been learning some new stuff

[00:14:15] Dri has offered mostly visualization.

[00:14:19] Yeah.

[00:14:19] Yeah, yeah. So how's that, has that been for you.

[00:14:22] It's been okay. Like I have a computer science background but I'm not a big fan of JavaScript so you're doing a great job as well I guess. Dimentia says JavaScript for these three. So I'm trying to learn that also there are many resources for the same. So I mean, they have new versions coming up quite regularly. So now the version is number six. But we have resources for version four, always in five. So I have to look at the documentation as well because the syntax changes and there are a few changes. Yeah, I'm trying to catch up.

[00:15:00] Yeah. Mean like I mean, you're still still super young. You got a whole lifetime of learning ahead of you, so that's awesome. Me personally, man, like I. OK, so that turned 38 later this year, some significantly older, I think, than most most people here or most people who are currently breaking into Data science. My background came from just like proper academic statistics and math and stuff like that. And it wasn't until like the last two to three years that I started having to learn a lot of software engineering kind of best practices. And it's been a super hard uphill struggle. And you know, me personally, I always feel like be like this. I wish I would have done this stuff when I was in my 20s or I wish I would have. I wish my statistics department was more computationally focused so that it would have better prepared me for for the future. But the other constant feeling of man, do I know enough stuff yet, what do I you know that it's it's it's rough. I know a lot of people deal with that. Do you ever do what you do and how do you how do you handle that consistently?

[00:16:06] I think part of it for me is being able to like, fact check myself. So, you know, when you are trying to break into a new career, I think early on, it's very much like telling yourself, hey, someone's paying me to do this. I'm legit enough to, like, do this. But the other half is kind of getting comfortable with not knowing a lot. And I think we've at least created a good community in Data Science and Emelle. I think that if you don't know a specific model or algorithm, most people aren't like, wow, how could you not know that? But more. Yeah, this is a weird thing we may only use in this industry. Of course, if you're on the product side, you haven't seen this and then it goes into like a good relationship about all and learning and being able to just exchange information. So I think that's helped make it less intimidating is knowing that the vast majority of people, even if they know X, Y and Z, more than LinkedIn, they would be more than happy to at least talk to you about what it is instead of kind of shaming you for not knowing. So I think having those experiences over and over again helps.

[00:17:19] Yeah, I that's the favorite. My favorite thing about this field is that there is so much to learn, but it's rather than being stunned until I can actually like oh my God, there's so much I need to learn where I start, like you just start with the basics, the fundamentals and incrementally to start learning and getting better and better. And I don't know, like me personally, I feel like I'm LinkedIn there's just a lot of, like, virtue signaling by people that go in order to be there. Scientists have to know this and this and all this other stuff. And it's like, I know all of this stuff and I'm a data scientist. And if you don't know this stuff, you're not a data scientist like me. That's there's a lot of that. And I think that really discourages people. And it's like, oh, man. Like, you don't need to know everything all at once right now in order for you to start making an impact. And you definitely don't need to start comparing yourself to every other data scientist out there. And you have like a measuring stick, like looking like this real data scientist to my not. And we will all go to these types of imposter syndrome bouts. Without a doubt, there are a career, but I think it's important just to focus on where you are, get good at some of the tried and true fundamentals and the basics, and just incrementally start just getting better every single day. Right. Pick, pick one thing a month that it is that you want to master. Right. And if you do one thing a month for the next year, you've just got to get at 12 different things and you've built your toolbox up that much more. So if we were to think about just some of the beer, I want to say beer, but just the ground level fundamentals that you need to know in order for you to really start progressing in this field. What do you think those would be?

[00:18:56] You say it's kind of the stereotypical kind of Venn diagram you might see, but it is true. It's really the core math. So probability and statistics, I would say a little bit of linear algebra, especially if you are ending up working on, like, deep learning projects. And I think something that it's not left out a lot. But I like to include really strong SQL skills when I talk about coding as well. That was something that when I moved from grad school to industry, I really wish to see a lot more. That was the biggest hill to climb, despite the fact that I had studied the core concepts. It doesn't make a great school master drill. And then going along with that is really this business sense and intuition. So I always like to mention, especially if you are changing careers, it's so much easier to change your job function within the same kind of industry or change your industry and Harp maintain the same job function. But changing both is always incredibly. The hard sell, I always suggest, especially if you are coming from an industry like health care or finance, but just want to change your job position maybe from being more on the ground to doing like Data science and animal work, using everything that you already know about that industry and being able to rely on those insights will make it easier to just learn how to get great at retrieving data and creating models and software engineering, because you kind of already know the the subsets of the business. So I'd say those are the top things to try focus on

[00:20:45] School without a doubt. And I think it's super, super important to know when to learn. And I think if you put in just a month or two of concentrated effort, you can get some really quick returns on on your time investment that will immediately start making me more and more effective. So I see there's some people unmuted here. So you've got a question. Let me know. Um, otherwise, just to reduce background noise and start, uh, just meeting people. So if anybody has a question, go for it. So I see here I see your name. Oh, all right. So if anybody has a question, go ahead and just type it right there into the chart that you got a question I'll ask you. The Q is asking SQL is what SQL structured query language. Yeah. So that I think is absolutely key. And yeah. You know, that's something that I wish I had learned in school as well. Like, you know, like I have to do grad school statistics. They don't teach school. That's one thing that that I that I wish I had learned a lot earlier in my career. Um, but the thing is, man, it's it's not too difficult to to learn.

[00:22:01] I always suggest, like, if you the best way to actually learn in school is if you have like any kind of Data that you think might be interesting, finding some toy Data said even it's there's a lot of places you can create, like just kind of a really tiny database for free if you like school and just put like Data off of Kaggle into it. The reason sequel's hard is because the nature of Data and most companies is not usually well structured and it's usually messy, not because learning SQL itself is hard. So the biggest gap for me was, yes, I had taken some online courses on SQL when I was in grad school, but got to industry and expected that most of these basic queries would work. And it turned out I had to make four different kinds of Joynes to get the actual data that I needed. And so it wasn't that it was difficult, but it's almost impossible to teach you SQL about a specific orders kind of Data. So there's so much relying on other people at your organization to say, hey, what are the specific homes they need to be joining on? And having a good idea of the landscape of your actual Data is something that's hard to gain from any of these like tutorials. So it's less that you have to memorize every single thing and more. You have to build that intuition for trying to understand why things are laid out in a certain way in your database and getting them together. That's the that's the difficult part.

[00:23:52] I think that's excellent insight. And that's one thing that makes a super tough, is that it won't the code won't always break or you'll get a result and it might not be the right result. So you have to be super, super careful with the way you write your queries and really understand it. And I think the hardest part for me about SQL is you have to really use a lot of your imagination when you're doing Joynes and you're in your head like I don't know of. I mean, I know there is there's Azure Data studio that's available, but unfortunately, it doesn't work with, like a bunch of different types of SQL engines. It's mostly for Ms. SQL Server. But Notebook's, I think would be really helpful if we had some type of notebook environment where we could do SQL with so that we can kind of see intermediately what we're doing. We're doing these queries. What's the Data looking like? Because I find I find that I have to use my imagination a lot when I'm doing a sequel because I have to imagine in my mind, okay, wait, this is a table. These are the columns. These are the ways I'm doing whatever Joynes and things like that. Did you struggle with the same thing?

[00:24:57] Yes, I very much. I have taken to a lot of complicated Lagu Data scenarios, whiteboarding and imagining what my. Table should look like I'm kind of working backwards from there, because you're right, it's you don't get to see these tables as you are writing a query. So it's definitely lean on some of that imagination's skills.

[00:25:23] Yeah, yeah. And it'd be great if we had, like, a sequel notebook that would just make life so much, so much easier. So we got a bunch of people joining in. Thank you guys for for joining in. Looking at the chat now. Just catching up. I see there's a couple of questions that we'll get right to. If you guys have a question, go ahead and put that right there into the chat to us. You got your hand up. I will get to question after we handle Quentin's question here. Um, so Quentin is asking, what are the must haves in your portfolio? Um, so what do you mean by must haves? Quentin, go ahead. And yourself. Yeah.

[00:25:59] Hi. Hi, everyone. Yeah. So basically I'm at a stage right now where I'm trying to build a compelling portfolio that is trying to convince what I know and I'm trying to I'm trying to figure it out like what you mentioned before, you know, that we have so many things to learn. And it's complicated sometimes to know what should be the next step, which should be which we should be learning. So the meserve's is like, OK, we don't know everything and we can keep learning in the next experience, in the next business or for ourselves if we are freelance or whatever. So the thing is, the kids like us. That's for for example, our priorities are key skills that we need to have in the first place. What all those skills like, what are the skills that we have to put in the portfolio and what are the skills that we can say, OK, with learning in the future? I will be able to learn in the future and your business, but I don't have them right now.

[00:26:57] And that's a good question.

[00:26:59] Yeah, that's very good. You handle ideally and let me just go off on that.

[00:27:03] Yeah. I think the biggest I must have is being able to show that you can clean and manipulate data. So I'm kind of going general here for like Data science job titles and being able to show that you can get Data from maybe disparate data sources. So combining multiple datasets on kind or something like that, I think that's one of the biggest. The aspect after that, I would probably say being able to interpret model results. So showing that, you know, yes, you print it out like your classification metrics. But even if you can go into detail in a couple of sentences, so what? So what does this mean for my modeling process? What does this mean for the potential product? And I would say the it's not a must have, but if you're maybe targeting more machine learning focused roles, I would say try to kind of show how if you can deploy a model in any way. So even if it's like a Django Web app with Piscine or making basic like predictions in browser, showing that you can have a model that is alive and working somewhere, I would definitely say is a nice to have, but it's something that would probably give you bonus points compared to the vast majority of other portfolios out there

[00:28:37] And does excellent points. And to that I'd even I'd like take it one step higher level, like not not so much in those technical details here, but a high level. Can you pave a way from ambiguity to a decision? Because I think a lot of the times in the real world, you don't have the answer to anything up front. You're not going to know how something's going to pan out or work for you. So being able to concretely state what it is that you are trying to achieve, what result it is that you're trying to work towards, and then from there lay out a path. OK, great. This is the thing that I want to answer and then lay out a path to get there. Right. And along the way, obviously, you're going to have to clean Data. You're going to have to model the Data come up with some type of, you know, do your exploration, your visualizations, interpret and and evaluate your models as well. So I think that is also a must have. So everything I said, plus clearly defining what it is you're trying to do and clearly defining what it is that you're going to stop or what it means to have a result or no result, if that makes sense.

[00:29:45] Yeah, completely. And actually, thank you guys for the answers. And actually I think it's completely in line with the subject of the dedicated conference where you were you spoke in the sense that the main point is that our storytelling basically is to have a narrative for the business and to keep in mind that all what we are doing is for the business in the first place. And apart from having a lot of practice, we know like we have to grind, we have to be programing, we have to do a lot of projects to be comfortable with this. Apart from the mere practice standpoints, would you guys think of any recipe to practice this business acumen, the fact of like what exactly what you mentioned. Right. And be to having a clear decision, making your material. You know, like do you have a full recipe to practice that specifically

[00:30:42] For recipe to practice business? That's a good moment. I don't I don't know that I have a recipe, but what I do is I read just a lot of white papers. So a lot of companies, they always talk about what it is that they're doing. Right, because the culture of Data science work is just everything's open source. So as much as I possibly can, I'll read up on, like real world case studies. And, you know, there's like a case study that Airbnb had done where they had talked about what they learned after one hundred and fifty failed models. Um, and, you know, things like that or like Netflix will always have you know, they'll always post something about how they're applying the science machine, learning to make better recommendations. So I think that's a key point. Is this reading up on industry white papers when they talk about this is what we try to do. This is, you know, the problem statement and what we're hoping to achieve for our end user and just being exposed to that as much as possible, I'd say ideally. What do you think? And if anybody else has some tips on this, I'd love to hear as well. Yeah.

[00:31:46] If you can share the resources as well, like you were mentioning white papers, you can share as much as you can as well would be nice.

[00:31:53] Yeah, I would say my strategy is fairly similar and that the easiest way, especially right now, since pretty much everyone is remote and to learn from what these teams are doing is by going to their blog or checking out these white papers, because that's one of the few forums that's public that we kind of get to talk about our technical failures. Right. So learnings after these models fail a lot. But I would say another aspect, at least as far as my journey would be having these conversations with other industry people. So especially when I was at a prior company dealing with a lot of issues in model failures and trying to make good models, I became kind of really ingrained in a lot of these other technical machine learning groups and community groups and would ask, so, hey, you are at X company, how do you deal with model failure? How do you deal with these projects not being kind of what you expect? So I've had to get a lot of those learnings from other people. And yeah, I think forums like this are great because we are able to provide things like resources. But it's a little bit harder right now to call it a friendly machine learning at another company and ask necessarily, you know, what what they're kind of struggling with. So why do you think that there could be a lot more work around you, help had having better resources to practice this kind of business acumen? I think so much of it comes from experience that you may not have built great tools to kind of help people learn that aspect.

[00:33:46] Yeah, and I think and I think, like you, you can always benefit from other people's experiences, but there is nothing better than your own as well. Like you understand things in different standpoint when you actually make the mistakes yourself. And and I don't think there is any shortcuts for that. At the end of the day, you have to make the mistakes and you can get a bit faster. Is you benefit from other people as well.

[00:34:10] Yeah, and I think that's a huge benefit of like conferences, like conferences. That's where people present what they've worked on. So if you go to YouTube and just look up like I'm blanking on the name of it right now, it's like they're not going to try to attempt to remember the name. But there's a bunch of things called PyCon or something like that, PyCon or something like that. But there's a bunch of presentations like that. If you just go to YouTube and type in like, you know, Spotify, quote unquote, Data science and you'll see a bunch of presentations that Spotify Data scientist had done or Netflix Data scientists have done. So that's a great resource as well. And yet just papers, so many white papers. I've got so many just saved on my my hard drive. And a quick thing to do is like, for example, if you're doing, um. A Google search, you do quote, like I'm talking like in that search here, so we do quote Data science, quote case studies, and then you type in file type calling PDF. Right. And I'll pull this up for Gosule quick just to show you how I'm doing it. And this is what I kind of used to, the type of search I do to help me find case studies. Right. So we could do Data science case studies like, you know, file type PDF. And this will restrict my search to just. White papers and the like, so here's one case studies, a machine learning, machine learning, case studies.

[00:35:39] Right. So if you go here a case studies and machine learning and what I do is I'll not only just read through some of this stuff here, but like go to the references section and see if they have any references to, you know, this looks like it's a graduate thesis. Probably not the best thing, but, yeah, just doing a search like that, case studies and you kind of like you. Typically, how people are on the Internet, you kind of go through a rabbit hole until you find a gem, so it takes a little bit of effort before you find a paper that really is interesting. And I try to do that with my newsletter because I do so much research out the week, just reading up on case studies and stuff. And I try to link back to something in my newsletter. So hopefully if you're part of that, you'll be able to see that. But here's a link right here that I could share for you or even if you have an industry that you're particularly interested in. Right. So you can look up case studies and like e-commerce or case studies in in manufacturing. And what you're looking for is like key terms that jump out that you can then go research more about. Um, yeah. So the answer to the question wasn't too abstract.

[00:36:48] Yeah. You answered my question completely. I mean, it's we are part of a world that is an abstract concept. So that's that's one of the thing. But I don't know. You answered my question perfectly. It's a good resources.

[00:37:02] Yeah, definitely. Go. Somebody else was.

[00:37:06] Yeah, I had a follow up question. Yeah. OK, I can read about these case studies and, you know, I can learn, but how do I showcase this? Like obviously projects as a in a showcase these students. Like what I'm trying to find, you know, some solution. I could tell my readers, like, you know, all this kind of stuff in a project. But are there any other ways I could show this business acumen?

[00:37:30] Yeah, I mean, through the project itself. Right. Like like that's how like everything you do in that project from how you clearly define the problem statement to how you are thinking through and sharing your rationale for why it is doing what you're doing that's showcasing that that business acumen. I tell you. What do you think.

[00:37:48] Yeah, I'd say yeah, your project, first of all. But I think in addition to that, this may be part of your interview process. So if you get to the point where you are talking to but I've had this come up in technical interviews as well as speaking with management, but how you can understand what the business problem is for a specific model or project. They want you to do a lot of the questions I have Geithner are around, OK, here is a situation. Here is what we want to do with these predictions. Kind of describe how you go through the modeling process. So that might help kind of bolster what you can show in your portfolio. But being able to clearly explain how you go from here to potential goals to the process that you would actually take in Data cleaning model building as well as moving into production. So if you get interview questions like that, it's a good chance to be able to show off that you do have some really strong business acumen there

[00:38:52] And even understanding how to develop KPIs and understanding, OK, if my machine learning algorithm achieves this level of accuracy or whatever it is that you count as a model evaluation metric, how does a, you know, a move on that needle correspond to a business KPI and try to find the connections between those two? Right. So definitely do some research into metrics and KPI framing and generation. Like, how do I you know, obviously topline revenue is something all businesses care about. But businesses in particular industries, they generate revenues in different ways. Right. So if somebody has like a software as a service platform. Right. The way they generate revenue is can be different than somebody in a manufacturing industry. So, you know, for example, like the software as a service, like what's the big one? I tell you in SAS, I think like customer acquisition cost and things like that. Right. Whereas manufacturing, it might be just labor costs. Right. So, you know, for example, customer churn is also a big one for a software as a service. Right. So just understanding those things very quickly, I'm predicting churn. And if I am trying to predict churn, then what does a what does a successful intervention look like? If I'm able to prevent X number of customers from churning, how does that relate back to top line revenue? What do you think?

[00:40:18] I do? I think you're spot on with that. I think that's an area that at least in several of the interviews I've had, we tend to go in depth on because it's not just about the model building process, but truly a deep understanding of the business. So being able to directly correlates KPIs and then measure if you're actually getting movement on these KPIs. So if you are, let's say you build a model that you put in production, are you actually starting to see results that you would expect? And if not, and that is often the case. What steps you take to. Iterate again, so knowing, let's say a model fails, what do you do to debug, what do you do to try and identify potential reasons for the failure? Is it that you had a mismatch? And what kind of Data you expected would go into the model when you were building? Being able to identify these parts and then go through that iteration all over again is something that I've been asked to talk through. Been in Data science interviews.

[00:41:30] And I can imagine the follow up question here as well. We're working on a take home like Nottingham project, but we're working on a portfolio project that's not even like a real business. How do we how do we even apply these to our to our portfolio projects? And the thing is, like, it's that's fine. Like if you're a portfolio project. Yeah. It's a fictitious environment, but that still doesn't prevent you from clearly defining what a KPI is. And it doesn't necessarily have to be like the right or wrong KPI because you're operating in a sandbox environment. Just the fact that you've put some thought into it and clearly said, OK, these are these are the things that we want to measure. And we know that if we move the needle on the model accuracy by this amount, then this is what we expect to see happen in in this particular KPI that we are hoping to influence with this model that we've built.

[00:42:20] One of the things I seen on, I would say really like extraordinary portfolios is going through this process of writing out what steps you did and what that kind of model looks like. But I've seen people go even further by adding is kind of what if scenarios or case statements saying what if one scenario might be you put the model into production and it does not move your KPIs by 15 percent or more. And so they'd under that section right through. OK, if this is the case, here are the steps I would take even without having to go through those steps with their actual model or another what if scenario. This model does improve our KPIs. By this. I've seen some people say let's long term look at this we put in so they put into their portfolio. This is how often infrequent they would collect new data to update the model. This is how they would track accuracy drift when the models and production. So those are things they might just be a little text segment as part of your portfolio. But when a hiring manager or reviewer is looking at that, they're like, OK, you've thought a lot further than just a single model in your in your portfolio.

[00:43:41] And if you guys were training it on my presentation, I dedicated this ties back to the analysis plan and the executive summary. Right. That this is where you would include that type of information that I was talking about. Thank you so much. So just looking into the chat, we got some questions. But first, I remember Tó had his hand up, so we'll go to talk and then after tomorrow, we'll go to a Carone crooner's question touristed.

[00:44:05] Yeah, it's funny because the original question, I had kind of disappeared, but I'm really curious more about when you talk about your project, how you perform. OK, I'm extremely new to this. I'm trying to understand the process. And from what I hear and the more I'm listening, it seems to me that the workload related to the actual algorithms, which are our language, whatever language you do, is that you basically do design conceptualization, Data, structuring, et cetera, to put in place first. But the actual algorithm process or the programing part is not really the biggest part of the job. And then the next step is the analysis and courses that get an output. And then you have to kind of go through a lot of analysis to get to review and then go back and forth, back and forth between programing and to kind to this project, to the expected results that you had initially. Is that a fair concept? And if so, where if you took 100 percent of spread over that kind of workflow, where most of the time

[00:45:25] That's absolutely good assessment. And I mean, just like any any science, you have to think about how it is going to proceed with this problem and that coding is a significant and important part of it. But I would say it's pervasive throughout the entire end to end the project. But I guess the if I understand your question correctly, it's where do you spend time in this in this pipeline in terms of understanding of the problem and actually hands on work?

[00:45:59] The to do that, basically, it's like you basically start hear stiff competition and then you go through the process. Now I do the same steps similarly in a much smaller scale, but I am less transactional data from my audience. These can range from one hundred thousand three, four, five million transaction lines or Data with 42 seconds. Now, that's my experience I use for this because that's what. You see, I have to clean it, I have to break it up, there's cells with specific break that structure that I would need to break down into smaller parts. So I do this by creating columns for me to add Data its structure. And that takes about 40, 50 percent of the actual several years. I have my my portfolio copy paste adjust and double click and it just copies the cells. And then when I run my cuticles, which is my summary's etc. again I go back. So the actual part of doing formalism shows very simple tasks. That kind of thing takes up very little portion of my time. It takes time is the preparation and analysis work after and then going back to the formulas, adding new columns to break down further, etc.. But then I go into the next problem, which is that I get a lot of data out, which I don't have to request additional action from the operator of the system where it comes from, what does exactions me and then I can start to. So for me, this is my project is how it works and the split of work is more, like I said, 50 percent in the first part, 10 percent of the actual programing, etc. And then the last 40 percent is the analysis.

[00:48:03] Yeah, I think that's fairly accurate for Data science and I think it's going to depend on who the data scientist is. Me, for example, like I just spent with the vast majority of my time thinking about what it is that I need to do and how I'm going to execute on it. So I think it was Einstein that says, give me an hour to solve a problem and I'll spend fifty five minutes thinking about the problem itself. I'm no know Einstein by any means, but I tend to find that that's kind of how I work, is I if I've got like, you know, if I time box a problem and I've got four to six hours to work on this problem, I'll probably spend a good three hours just thinking about, OK, what is it that I actually have to do. Right. And just distilling that complexity down into actionable steps and breaking it down. OK, great. If this is my end result that I'm looking for, let me work backwards from there and plan out small, discrete tasks that I could do along the way that are going to move me from absolute nothingness and ambiguity to what it is that I'm wanting to achieve. So I will I will spend the vast majority of my time just thinking about what it is I have to do, breaking it down, putting it into discrete chunks and then just executing I. What about you?

[00:49:20] Yeah, I would say it's hard because I feel like it depends so much, but I think maybe an initial twenty percent of my time really trying to formulate that problem, understand what I need to do, and then between like 20 and 30 percent of that time is spent like cleaning, manipulating Data and making sure that's what I'm using for training is what I should be using for training models. And then coding is about another 10 to 20 percent. And actually building some of those pipelines in model building and the rest is pretty much analysis and kind of the iteration of analysis. The model wasn't great, but going back and doing those tweaks and the rest of that is really just iterating do the remaining steps.

[00:50:16] Yeah. When you have a framework in place, it just makes it so much easier. So that's why it's good to establish a set of principles by which you execute on your work so that you don't have to reinvent the wheel every time. And, you know, like I talked about in my presentation, a couple of frameworks. One that I've been using a lot of recently is called Quadro. And the federal framework just makes it really easy for me to to to execute on stuff. And before Khadra, it was cookie cutter Data science. So these frameworks in place that just make the actual execution of things easier and more streamlined and make the pipelines kind of fit together. So next question I got up will go. We'll do two more questions and we'll call it a day unless there's more questions. You know, if you guys got questions, now's the time to put it in the chat to hold your place, because after we do crooner's question, we're going to go to Pasha's question. So if you have a question put into the chat right now, who would place otherwise will and after these two koruna go for it?

[00:51:17] Oh, yes. First, I would like to do a follow up on what was the discussion earlier. So I basically struggle on the timeline as well and my project. First, I start pilot research on the project, so I find out while working on it that this might not be the good approach or while I'm researching, there are multiple models out there and multiple research papers that I can implement. So I just struggle. I get a hard time on deciding what I should do first or should I basically start?

[00:51:56] So you start with the baseline model, the simplest possible model that will get you a good result, and then you iterate from there, right. So you have a suite of possible solutions that you're going to test out. And you start by defining this for yourself. Right. You have to have a contract with yourself at the beginning of any project in the form of an analysis plan, which you say was clearly defined. The problem is that you want to work on that. OK, great. This is the problem I'm working on. This is how I'm going to establish a baseline line in the sand that is the simplest model that will give me good results. And then from there, I'm going to assess whether machine learning is going to be the right solution by applying a suite of techniques. And if this suite of techniques don't meet the simple baseline model, then hands up. This is not a good application for machine learning. But the thing is, you have to you you can you make a choice, right? You make a choice and start a project. Do I want to be stunned into inaction by all the nuances of the real world, or do I want to just put it in paper and make a contract with myself that this is what it is I'm going to do and I'm going to execute on it? Right. So the choice is up to you. Like, do you want to be standing to an action or just move the needle? Right. Because the fact is there's a multiplicity of good models. I, Leo Brockman talked about this. He invented the reenforced algorithm. And essentially he says that for any any set of input in any target, that there is going to be a whole host of complicated, complex machine algorithms that will machine learning algorithms that will give you as good results as any of the other ones. Right.

[00:53:31] So I don't know.

[00:53:33] Not yet. There's lots and yeah, the study does is helpful. It's not the question that I'd really put on the chart was basically it's it's the buzz out there that for a Data science profile, you have to have a minimum at least three to five years of experience. What at least I have seen up till now that employers do what someone even posted this on the LinkedIn that Tensorflow was reaching them out so that they could join their company. And they said that you should be having that at least 15 years of experience in applying Tensorflow. And the person commented that Tensorflow was Belder or it was in place in 2015 and it has been only five to six years that Tensorflow had been there. Then how can you even ask for a ten years of experience? So I get a hard time on actually getting the employers to basically at least go to the profile and then you can decide whether I'm fit for the role or not or not, just deciding by the number of experience or having a graduation from McIlvaine Audio to institute.

[00:54:52] Yeah. So I mean, at that I'll give you my answer. You might not like it, but the first the first part of that question is, OK, I see these job postings, I have these ridiculous requirements and how am I even going to break in like I'm not even going to bother playing? Well, first of all, like nobody is going to reach their hand through your screen and slap you for trying to apply for a job which has these crazy kind of requirements. Right. So don't worry about that. If you see a job requirement and this is a question that I talked about with a lot of people that come on to my podcast, ignored the job descriptions completely and apply for it anyways. Right. The fact is, when you have a career that is certified as sexy and everybody wants to become a data scientist, now you're going to have a ton of people applying for the job. And sometimes it's not even the Data scientists who write these job requirements that these job descriptions. Right. Is done by somebody who doesn't have any clue about their science. And they put these really crazy requirements just to deter people from applying for the role. Right. That's definitely a possibility. I don't know if that's answering your question at all, but my message here is ignore the job descriptions, apply for roles anyways.

[00:56:04] Right. But here's the thing. Like your job isn't over just by submitting your resume. There's still work that you need to do on your end. And that work is getting people inside the company to notice your profile. Right. And the way you get people to notice your profile is by trying to message recruiters and hiring managers through LinkedIn or however it is that you need to get in touch with them and. Building relationships and trying to get them to look at your your profile and you do this enough, right, because it's a numbers game, right? You have to keep applying for jobs. You have to keep attacking before one of these applications converts to a actual job offer. I think for any given job that you apply, just just just assign yourself a prior probability of less than one percent chance of you landing this role. Right. And then as you progress through the process, you can update that probability. But I don't even know if I answer your question at all. I'm going to turn this over to Odelia and then we can see if we answer your question.

[00:57:04] Yeah, I think there's a valid point to made, though, about not being genuinely considered for roles, especially so many companies using ATSI systems that are looking specifically for certain keywords, as well as the user experience being one or just a number of prior roles. I would say that you kind of hit it on the head Harp it was really just making relationships as these organizations. So not just kind of adding someone on LinkedIn to say, hey, you know, can you give me a referral for this role and make sure it gets seen, but really trying to build a relationship with people. And I know that it's incredibly difficult because I've been there and I've done it. And you'll probably get silence more than you will ever get, like a genuine response. However it is it is that numbers game and that if you aren't making these connections and you aren't reaching out, otherwise, there is little reason or there's little other ways for you to update or increase your chance of really being considered. So sometimes what I'll do is before I even apply to a job, see if I have connections to these people, if we just happen to be following each other on Twitter or something, especially if it's a large organization, it's easy to narrow down on Twitter and find those with specific job titles and data science or email and then try to build relationships there. I've done really just short informational interviews. So you want to work in this organization, so you spend a couple of minutes talking to someone who does work there and maybe, you know, you can build a good relationship from there and they may offer you referrals. So it's it's definitely taking some of the creative and pretty bold route to just start kind of cold messaging people out of nowhere in order to get seen, because it is really difficult when you don't have a certain number of years of experience on your resume

[00:59:14] And like there's hundreds or thousands of people applying for the same role. Right. So they put up these crazy requirements in a sense to deter a fraction of people from applying. Right. So that's definitely out there. And I firmly believe that you can replicate Data science experience for yourself through a well done project. Right. If your project if your portfolio or project is nothing but a Python notebook hosted on GitHub. And that's the only thing that's not a great project that doesn't really showcase that. You are a professional. Right. So do what your competition is not doing right. And separate yourself from them. If all you're doing is looking at, you know, the playing field. Right. And not wondering how you can separate yourself from them, then you're not really going to separate yourself. I'm talking in circles right now. I apologize for that. But I'm going to stop talking because I didn't see Corona with that. Answer your question first. Let me know if I'm saying your name right now.

[01:00:11] Sure it does. I, I really get do I know what you're talking about? And yeah, I do not really apply apply on the jobs that I do not even fulfill the number of experience, but I'll just work on my projects more and I'll get back to you if it doesn't work anymore for me. Sure. But yeah it was very helpful.

[01:00:34] Yeah definitely. You know it's more than just the project. Got to make sure your resume, your game is on point. Make sure that if you do have a project that you're recording yourself and you're putting yourself on YouTube talking about your project. Right. Just give people a good sense of your candidacy holistically, I guess. Um, yeah. Good luck in the search. So I see that. The next question I got mine here from Depeche Sender. If you have a question, do you want to just hold the place online or if you have a question that's related to what we're talking about right now, you can take the floor. But if it's not answering,

[01:01:09] My question is related to actually, OK, I'm a master in the science student and my work experience is related to my masters, unfortunately. And I'm trying. I started applying for jobs in December, late December. Because I'm getting my master's soon in the summer, so I want to start with practicing interviewing, especially technical interviewing, and I had a couple interviews, but is being hard to take your job offer because the there's a lot of competition right now, as you mentioned. And the I need to I realized after having a couple of interviews that I need to my step up my game on the technical part, not so much in the behavioral because I already have more experience. So it's more about showing that I can do Data science. So so that's why I related to you, to the question that Kuruma made. And so my question is, how can I better position myself? I already know everything that you talk about, like reaching out to people that work at these organizations and asking them about the house work there, the the culture and also being getting a referral. So I I'm like I still have time. Like I read the summer, I still have time to get a job and but I just want to get better at this time. I can get offered.

[01:02:50] I think I very much relate what I was trying to do when I was interviewing, trying to get rules of the jobs on my resume actually related whatsoever. The first piece of advice would be trying to focus in a little bit on the industry of your pastoral. So if there's any overlap there or if your previous work was all in one industry trying to target Data science roles in that industry, the other piece of advice I would have, I guess, would be to there's a couple of routes you could take. So I've seen a lot of people go the freelance route and find a lot of success, even if you're not doing full time freelance work. But having a couple projects to speak on is really helpful when you do get to the interview stage. And I've also seen some, like, amended things you can do to your resume. So even before, like, you have the heading for work experience and under that kind of put what most people would consider their objective. So I'm a career transition er going from all these kinds of rules targeting Data science and normal rules, because I have done my master's in this and X, Y and Z. So I think that kind of primes who the recruiter or the interviewer for. I know that I should be looking at the bullet under this work experience to try and match to the job that we have open. So those are a couple of things that you might be able to try.

[01:04:30] Yes, I tried about looking for companies that do fintech because my background is accounting, my work experience, and that's what I'm trying to focus more like. And so I if you give

[01:04:49] Me a link to get Harp, let's look at your portfolio and let me dissect it for you right now and tell you what you could do immediately to improve on that so you can start getting more callbacks and also applying since December. That's not not too too long. So I think you'll you'll notice that there's going to be a delay, you

[01:05:08] Know, but I've been able to get interviews already. So it's just I have to work on my interviewing on the technical part since I don't work in the field. But I have the I'm doing my academic studies in in Data science. I just have to practice more like do mock interviews. That's what I realized I have to do so I can and are doing them. I'm not doing them. So that's why I kind of put a post on applying to jobs right now while I work. Oh my oh my interview. My technical interviewing.

[01:05:50] So yeah, they have to do mock interviews and then, you know, do research into the most commonly asked science questions are for interviews. There's literally hundreds of blog calls out there and you can or even FOLOTYN versus the online. He's got some great on LinkedIn rather he's got some great resources as well for these interviews. So I highly recommend that. But if you want me to look at your portfolio project and just tell you some immediate improvements, either, I mean, you could come to the Friday after our session. We can come talk about it right now and we can start making some serious improvements either way. Let me know.

[01:06:25] Thank you. I appreciate it.

[01:06:27] You'll be happy to do that. But you got to keep applying to remember. It'll be a time delay, like if you apply to jobs in December. You might start getting calls, callbacks, like towards the end of this month or early March for those jobs, because there is a window of of incoming applications. Most companies will probably accept applications after a certain point cutoff and start looking at them and reviewing and then start narrowing down the final list and things like that.

[01:06:55] Yeah, and I must say I must say for other people in the call to that, that referrals do work, that you can increase your chances by by getting a referral.

[01:07:07] Anybody wants to give Sandra a mock interview, send her a private message right here. You guys should be helping each other out as much as possible. So, Sandra, I would even put a status update.

[01:07:17] Thank you so much.

[01:07:19] And just tell people to give you a mock interview, because here's the thing. If you help people with mock interviews, you're helping yourself as well. Yes. So definitely help each other with mock interviews. I'd love to see that happen. You know, even post it up on LinkedIn as a as well status, whatever it was called, and ask somebody to help you out. So. All right. So let's go on to the question that after the first question, I think we'll wrap it up for you. Celebrity.

[01:07:48] So. So actually, my question was around. When we start off with a project which we are working on with our clients. So like from a business business perspective, we are all you know, as a data scientist, we are already at a disadvantage that we don't have any domain expertize as much as other clients do. So in many cases, what happens is that the senior management doesn't need us to produce any novel insight. But what they do want is that they have a very clear decision in their mind. But what they do want is an analysis or a story or a narrative to support their decision. So what does happen is that if we stick to our own correct methods, approaches to bring up to bring about a solution, it may or may not align with what their thoughts are. And if it doesn't, then the senior management directly questions you that, hey, dude, maybe your approach is wrong or the data that you're taking is wrong or there's something wrong with what you are doing. So how do we go about navigating this kind of a situation where, you know, just because of a lack of Data expertize the business leaders or, you know, the business side of folks don't take you seriously as they should have?

[01:08:59] That's a good question.

[01:09:02] I think at first it's difficult because there are other people and their expertize in question is, well, I think part of almost every data scientist job is a little bit of education. So trying to approach the situation with some tact and being able to educate them. So I have found it important to let them in a little bit on some of the statistical stuff that's behind what we're doing. So talking to them about the kinds of metrics that are important and the kind of bridging some of those gaps in their Data knowledge so that you can do your work either convincingly or help them to understand that the vast majority of what we're doing is predictions. And we don't really have the real truth. And we I think it's part expectation setting. So I know so many people think that we are kind of wizards and can kind of create a model that just works well or create have an analysis that backs them up. And I think that trying to get them away from that mindset and say, you know, sometimes it's about understanding motivations. Why are they motivated to have data that supports a decision? Sometimes it's a direction they want to move the company in and they want analytics to back that. But I think working with them to understand a lot of the limitations in what we can do, and that may be the approach, the best approach is not to just look for data that backs them up or backs a specific position, but data that is more close to the ground truth and an analysis that can give them options. But I would I would urge you to suggest that they have a hypothesis or something that they want to verify that's not just coming from their position, that they want Data to verify, if that makes sense.

[01:11:29] Are saying that you could take all the boot camps and classes that you want, but they don't actually teach you the fact that you work with people. Right. And there's a whole element of soft skills that I think a huge swath of Data size candidates just ignore. Right. And they just think, oh, I know models, I know Algorithm's, I know Cotting. And the best data scientist. Naaman, you're going to run into people consistently throughout your career that are going to question everything that you do. Right. So you need to be able to work with people. And if you're coming out from the perspective that I know the Data, I'm doing it right. You finally got this wrong. Not the intuition is worth something. Right. And it's up to you to influence people. Right. So I think one thing that you should do is work on your interpersonal skills, work on your influence and persuasion skills. Right. There's a whole host of non obvious skills that you need to have as a data scientist to succeed. Good thing my friend Keith McCormick has a class on LinkedIn Learning that is titled The Non Obvious Skills for Data Science. So sit and watch that course. There is no clear answer to a question. His answer was spot on really good. But I mean, dealing with people who, you know, once once people have a theory of how something should be, everybody wants the next thing to be just like the first, um, which it's not the case.

[01:12:53] Not reality. Right. Um, so I have a framework called the Epic Framework epic. Right. And it's empathy perspective. Taking influence and concurrence. Right. So if we're working with the stakeholder and the stakeholder doesn't buy into your model, first put yourself in their shoes. What is it that they are feeling right when confronted with this evidence? What do you think that they're feeling? Right. Once you can put yourself in that place? Right. They're probably feeling this way, then take their perspective, the cognitive processes that are going on in their mind. All right. What is this person thinking about the situation? Right. And how can I understand that once you can apply empathy and perspective taking, then you can move towards influence, which is right now that I understand how this person feels and what they're thinking. Now, I can talk to them about how we are actually working towards the same goal. Right. And influence their decision or their perspective on what it is that I'm doing and hopefully achieve some concurrence. Right. And this framework has helped me greatly dealing with people who think they know how something should work and are so set in their ways. Um, it just helps to to nudge them into the direction to see things your way. So hopefully that's helpful. If any follow ups that let me know.

[01:14:10] With an thanks for sharing your perspectives, I would just like to add one example of a situation that I have been through, that it takes a certain amount of time to understand the why behind every project. So as and when we start over with something. So maybe the client would give you some vague description of what we want to do. And as we progress, we go about trying to structure their problem, trying to figure out how to get these other steps that we can take and come up with a solution. But as we progress and, you know, encounter hurdles in a process that why is this not acceptable? So that actually has an example or in a project that I could work on. So it gave me an idea through these iterations, through these hurdles that, OK, why is this direction being taken by them or why is this approach, you know, very well received or this one was not, you know, well accepted by them. So I think that's the approach which you mentioned that that epic that pretty much, you know, that was very thoughtful. And I would certainly try to apply this approach going forward.

[01:15:17] Yeah, definitely. In, uh, take a look at the link. I hope you guys have LinkedIn learning up. It's worth it. It's definitely been a huge benefit for me. But of course, I linked to by Keith McCormick the non technical skills of effective data science, uh, scientists with the non obvious skills that you need to succeed as a data scientist. I definitely check that out. Um, and so we do have a couple questions in the chat. Then we'll wrap it up. Question here from Quinten. What methodology do you use for framing the problem statement and valuable resources about that? Uh, the scientific method applied to Data science. And there's been really a couple of really good articles online about that. I can't remember the author's name, but if you do like a search for like, quote, data science and, quote, scientific method, I think it's like the first or second search result that comes up. Uh, definitely check that out. Ideally, what do you use to frame.

[01:16:13] Oh, my. I actually posted a link in here for problem framings. So this is a method that I've used in in in the past. And it goes from scoping to acknowledging some of the. Ethical requirements or things that aren't to say non obvious, but things that are usually not part of our project are still being planned. So we tend to use a lot of scientific methodology without trying to include some of the potential impacts of these modeling of this modeling, especially when we're using like health care Data or anything that else that is high impact, higher risk. So I would start off there and trying to get comfortable with asking these similar questions for pretty much every kind of modeling project. You face

[01:17:09] It. Yeah, I think that's probably the hardest part is framing the problem correctly, which which is why I think it takes so much, so much time. But once you do that, I feel like it makes the rest of the work so much easier to see clearly defined the problem and what it is you're trying to do. And you just things are so much smoother. So last question. I think the challenge will be last one for a four hour session today is from a look. A look if you want to go.

[01:17:34] Yeah, yeah. I have had so many questions. Like, we see a lot of algorithms that if I see from a beginner point of we see a lot of those in front of us. So I will do exactly the approach to get a good start and a good hands on them so that at least we can start approaching the interviews and then keep on and bring partly because the dashboard thing and then the cosmetics part, we can manage it anytime. So that's not the difficult part for the algorithms is. So how do we up

[01:18:08] The dashboard in a visual? That's the hard part for me. I don't. How do you handle this?

[01:18:14] Yeah, I would say trying to get as much hands on and then hands on practice, but trying to do things incrementally. So if you are just starting off, you can, let's say, take a go down the rabbit hole of like logistic regression. Really feel like you can understand this, put it into practice and then answer the hard questions about it. So you'll see, I guess logistics is a bad example, but let's say you're doing linear regression. Things that you'll see in interviews are down the line will be tell me about specific like linear models or tell me specifically what kinds of situations you would want to use us for. So I see each kind of algorithm as having its own really long, in-depth kind of learning plan, making your way down that without feeling like, OK, you're supposed to know every single thing about a specific algorithm within a week. But even if it's you are spending two weeks or whatever that kind of sprint looks like, going deep and making sure that you can answer the whys. So the why you choose a specific kind of boosting model over another model. Why a scenario where you're trying to predict like illnesses is not a good use case for some types of models. So those are the questions that aren't really included in these tutorials. But with practice putting that into applying it with real data, you'll be able to build an intuition for what are good scenarios and what aren't bad scenarios for different kinds of algorithms.

[01:20:08] Yeah, exactly, because currently I have the same job I'm going through with my studies. Right. So I just started using data analytics, Postgraduation and Canada. So I have some subjects upcoming for those concepts. And so no, I'm just trying to be prepared well for the project as well so I could perform the best of yet. So that's what my aim is. So thanks for the advice. So I'll just let them awesome.

[01:20:34] Yeah. So any other questions based on that.

[01:20:39] I would find out long but better I to begin at some point.

[01:20:42] The Corbitt well doesn't look like there's any of the questions in the chat. So I just came in at the very last minute you joined. Right as we're about to end it. Thanks for uh thanks for coming by the way. Yeah. Doesn't look like any more questions with you guys. If you do have a question, ask me the question. Now's the time to get it out there. If not, guys, thank you so much for taking time out of your schedule to be here today. I appreciate you guys swinging by. Um, take care. Have a good rest of your Sunday. And, you know, definitely if you have questions, feel free to to reach out and shoot me a message or email. If you have a question, but you can't make it to the next session, we'd be happy to tack it on. So guys, take. Have a good rest of the weekend, remember, he got one life on this planet. Why not try to do some big cheers, everybody.

[01:21:34] Thank you. Have a good one.