Harpreet: [00:00:06] What's up, everybody, welcome, welcome to the comet, M.L. Office officers powered by the artist at Data Science. I am your host Harpreet Sahota. I'm super excited to be here today. Super excited to have all these guys here shout out to Harpreet: [00:00:20] Everyone still in Harpreet: [00:00:21] The room right now. I know some people got the cameras off right now. Some are eating. Some people are shy. Some people are just connecting. But special shout out to Marion Lukas in the building haven't seen Elka in quite some time. Hopefully you're doing well. Austin's sitting here Harpreet: [00:00:34] Chillin. Harpreet: [00:00:35] There's Elka and charms in the building as well. Super excited to have all you guys here. Hopefully, AIs got an opportunity to to tune into my podcast episode that I released just a couple of days ago with Max Frenzel, who is an A.I. researcher all around. Cool Guy, author of the Harpreet: [00:00:53] Book Will, Harpreet: [00:00:54] Coauthor of the book Harpreet: [00:00:55] Time Off. Harpreet: [00:00:57] It's all about cultivating a rest ethic, so it was cool. Man is really, really good conversation talking to him, so hopefully get a chance to tune into that. Harpreet: [00:01:03] A lot Harpreet: [00:01:04] Of other awesome episodes coming up in the next few weeks that I'm super excited Harpreet: [00:01:10] About started to Harpreet: [00:01:12] Book up some guests for the podcast as well. For, you know, these episodes that I'm going to be recording in the next couple of months, they won't be released until Harpreet: [00:01:22] Like March Harpreet: [00:01:24] Or April of next year, but they're all going to be Harpreet: [00:01:26] Live so you can get like Harpreet: [00:01:28] A sneak peek. So I got a lot of good friends coming around. The show, got a lot of awesome authors and thought leaders and stuff, so that should be a good, good time. So for those of you tuning in live on LinkedIn, tuning in, live on YouTube, I want you guys to know that we are most definitely taking all of your questions. So feel free to drop your questions into the comment section, into the chat section. Wherever it is that you are joining us from Harpreet: [00:01:55] Will be Harpreet: [00:01:56] More than happy to get to your questions. So on Friday, we kicked [00:02:00] off this question. We kick off the session on Friday with just advice on how to. Set yourself up for success Harpreet: [00:02:10] When you start a new role. Harpreet: [00:02:13] A lot of good advice and a lot of good tips during that that session, so tune into Friday's episode to to hear a lot of good advice there. But I wonder, like, does anybody have advice just in terms of how to set yourself up for success, not even in the first 30 days, but just continuously write because you know, you hope to be in a job for much longer than 30 days? What tips do Harpreet: [00:02:36] People have, you know, out there in the audience Harpreet: [00:02:40] To just keep that momentum going and keep that excitement going when you start a new job, right? Because you start a new job, it's like you're so excited about everything, but inevitably you might lose that excitement. How do you keep yourself excited about things? What do you what do you tell yourself? How do you how do you stay connected to that initial burst of excitement you had when you first started a role? I guess as we go to Elka, then let's go to Marion and we'll start there. And if you guys have a question, go ahead and let me know. You can either type out your question in the chat or you just say you have a question Harpreet: [00:03:13] And I'll Harpreet: [00:03:15] Add you to the queue Harpreet: [00:03:17] And we could take it Harpreet: [00:03:18] From there. But let's let's go to Elka and then then Marion. Elka: [00:03:22] The shooter, so. It's kind of an open question, but I would say to stay curious, maybe because I feel like there's so much interest in things in the world and in the domain of Data science especially. And like whatever you end up doing in your job, there's interesting aspects of it. There's interesting ways to look at it. So just look for these ways to look at it, try to find them and like approach problems in this way. And as long as you are learning and finding new interesting things, then you will keep getting excited about it like that? Harpreet: [00:03:54] Stay curious, stay open minded Harpreet: [00:03:56] And just willing Harpreet: [00:03:57] To explore. Marianne, what about [00:04:00] you? Marianne: [00:04:02] Basically, I hold a bit of doubt that NLP Harpreet: [00:04:05] Said, Marianne: [00:04:07] Unfortunately, in my previous experience, because I was sort of in a field that is very immature. I couldn't keep my excitement for long, because once you sort of London, you feel that you don't know, then there is nothing new to learn, and that's, I think, the most important thing. If you don't. A basically done in a case when there is nothing, there is nothing Harpreet: [00:04:33] Useful, then Marianne: [00:04:34] You'll be excited always. So don't box herself into something like a niche thing that is very limited in opportunities to learn. And two pass new questions over the new problems and to define. Harpreet: [00:04:52] So that's great advice from everyone, I love that. Let's hear from Toshi Mantashe. Good to see you again. She's been an O.G. man. He's he's Harpreet: [00:04:59] Been a Harpreet: [00:05:00] Supporter for four hours a day since from day one. Men haven't seen you in a while, hopefully doing well. Toshi, if you got some tips. Definitely would love to hear from you. Elka: [00:05:09] Hello, Harpreet. Yeah, it's been a while Harpreet: [00:05:11] And I've been learning a lot from from Harpreet: [00:05:15] You and all the folks Harpreet: [00:05:16] In the happy hour. I actually have a good news. I, you know, you know, I've been here ever since I was a student, right? And I just recently graduated and Elka: [00:05:27] I accepted an offer from Bloomberg for the Data endless straw. Harpreet: [00:05:31] Nice. That's huge, man. Yep. So I'm starting in 10 days. Harpreet: [00:05:35] It's like the Harpreet: [00:05:36] Entire process are really fast. And the question asked today was is very Elka: [00:05:41] Relevant to me because now I'm starting a new job and I was like, Wow, perfect for me. Harpreet: [00:05:47] In terms of tips, I don't really know. Elka: [00:05:49] I'm a young professional, so I think I'll just like, wait for you guys and Harpreet: [00:05:53] Listen to what you guys have to say and take notes right now. Yeah, absolutely. Well, congratulations, Toshi. That's huge, man. Mike Bloomberg. That's that's [00:06:00] no joke. It's a huge company. I definitely tune into the episode that of the Happy Hour session from Friday. It was such good advice in that episode. I think you can benefit a lot from that, even though I'm a big fan of like books and stuff like that. One of my favorite books is that sitting on the shelf over there, there's the first 90 days, and then there's the follow up to the first 90 Harpreet: [00:06:21] Days called Master Harpreet: [00:06:23] Your Next Move. They're both by the same author, but I can't read the author's name from here. Anthony, congratulations. Man, that's that's a that's that's huge. I'm very, very, very happy for you. Let's go to let's go to Austin Austin. How do you keep that? How do you keep that excitement up, man? Austin: [00:06:42] I think kind of connecting to what I said was, I think, saying skeptical both of myself and what I'm bringing into a role. So opening myself up to learning new things, learning the new structure of a thing, but also to like. The things I'm hearing from people who've been at the company for a long time because that skepticism is like the first step towards challenging, presenting, challenging ideas or showing. Something to the leadership that they might not have seen before questioning their biases, so being skeptical of like, you know, it doesn't have to mean you're negative about about the vision or what the product is or what you're working on, but just being skeptical of the way sort of everything is canonically being presented to you so that you can start to really question and identify what the biases may be and then where you can kind of come in and present a different paradigm or it's within your team or organizationally at large. It depends on how big things are, obviously and where you fit in. But I think skepticism is something that is maybe overlooked and and because you want Harpreet: [00:07:43] To like, I think the tendency Austin: [00:07:44] Is you want to dove in and accept the new things and feel excited about them and feel like, yeah, there's really something happening here. But if you can force yourself to be skeptical about what you've learned and what Harpreet: [00:07:56] About what the sort of the company Austin: [00:07:57] Or the organization your team is presenting you, I think that [00:08:00] gives you avenues then to think divergent. Whereas if you sort of sort of accept what's being presented to you and accept even your own experience, I think you kind of can shut yourself off to new avenues of thinking about solving a problem. Or, you Harpreet: [00:08:12] Know, if it's like, Austin: [00:08:13] Oh, I think the main problem is actually this, and it's a slight alteration of what I thought when I was hired. So I think like, I'm experiencing that right now kind of a little bit at comment, which is an interesting phenomenon and I think is something that's really opening some doors for us to figure out what we want to do next and where we want to go with community and content and all these different things. But yeah, I don't mean negativity. I just mean questioning in a skeptical way and sort of is that, you know, really how this looks to me and that kind of thing. Harpreet: [00:08:42] Great tips. Very, very good tips, I think I'll be employing that as well. Mostly because, Harpreet: [00:08:47] You know, also my my Harpreet: [00:08:49] Yeah, you don't drink the Kool-Aid, don't drink. Austin: [00:08:51] That's what I'm telling you. Harpreet: [00:08:54] Shout out to everybody is doing the room. By all means everybody that's joining. Please feel free to turn the videos on. Would love to see everybody's wonderful faces on this beautiful Sunday morning. I'm ready to jump into questions if anybody has questions. Shout out to Oshkosh is in the building. Good to see you again, my friend. As usual, I'm going to put that one put me on the spot charm. You hit me up on Harpreet: [00:09:14] Linkedin Harpreet: [00:09:16] A couple of days ago, maybe, and I directed her to come to the to the office hours. So by all means, now let's get to your questions, Toshi. I'm not Toshi. Charm me. Super excited to Elka: [00:09:27] To to help you. Everyone is everyone doing so. Harpreet: [00:09:32] I am a Elka: [00:09:34] I'm a third year student in Toronto. I'm studying software development and since the beginning of the year, I've been very interested in data analytics and data science, basically. I also did a couple of courses such as the Google Data certification course and a few more. And I have a good hold of languages and what actually goes into the work. [00:10:00] And right now, I Harpreet: [00:10:03] Am looking Elka: [00:10:05] For a mentor or someone who can guide me to do some projects or maybe like an internship so that I can get hands on experience in the real world. Hmm. So yeah, right now as thought, she said. I'm I'm actually taking notes. So what everyone's saying? Harpreet: [00:10:25] Yeah, definitely. Well, I encourage you to come to these. Like, you know, every week I told you, we got this one every Sunday. We've got another one that happens every Friday in terms of, you know, I mean, just so happens that like this is interesting. The problem that you're describing because I'm like, I've created a course that deals with this exact situation that you're in that is going to be launching sometime in October. So but I won't use the comment my office hours to promote my own course, but that's how that's happening. Definitely, you know, put you into the loop for that. But I think the first thing to do. Like so it's it's interesting when you are at this stage where like, hey, I know Python a know some data analytics. But if I get the actual question placed in front of me, like, like if I actually have to do the work, I might just freeze up and not know where to go with it, right? Like he might just like fruits. What do I do? How do I do this? How do I approach this? Where do I go from here? And I think the way that you overcome that is just by doing more and more like small problems, right? So I call these just little discrete problems, right? Little miniature discrete projects that you don't necessarily share with the entire world, but you do it just for you to understand a concept or understand how something works, right? As an example, Harpreet: [00:11:43] You know you're saying Harpreet: [00:11:44] That, OK, I'm good with with the Python. I'm good with Harpreet: [00:11:46] Some data Harpreet: [00:11:47] Analytics, but like, you know, if I get a brand new data source. What do I do with that, right? You can just do a little project for yourself. It doesn't have to be any more than one Harpreet: [00:11:58] Day and then you [00:12:00] Harpreet: [00:12:00] Project could simply be, I'm going to take this raw data and I'm just going to practice Harpreet: [00:12:04] Cleaning it right? Harpreet: [00:12:06] You know, what are some some things that I could do to clean the Data? Right? So begin by exploring get begin by getting a feel like not exploring it, but understanding it. I call it Data understanding where you just kind of inspect each column, inspect the Data, see what's going on and see how you can clean it and just make that a project, right? And you don't necessarily have to put it on a portfolio, but you just practice it. Alternatively, let's say you're trying to learn how to work out of like a API or something like that, and you don't know how to interact with Data that's in an API that's coming in from like, I don't know, like a JSON blob. How do you normalize that? How do you add structure to it? Well, you can do a little mini project might spend no more Harpreet: [00:12:48] Than a day on it, like I'm going to go to the Harpreet: [00:12:50] Wet Weather Channel API, I'm going to pull data from like weather data from all my favorite cities around the world. It's going to come as a JSON blob and I'm going to do this. I'm going to make a pipeline that pulls the data structures it cleans. It does some feature engineering and then outputs a little graph, right? And that little graph, just like visualizes temperature by day, right, something like that, so that's how you just build momentum, like the first project you do doesn't need to be like this. All our big grand, massive masterpiece project you have to like build up to that iteratively. And the way you do that is just like I said, these small, discrete projects that help you build this much larger body of knowledge. I'll pause there and see if you have any questions. Elka: [00:13:40] Right, I get it. I have also worked on many projects for over Data that I found it was probably in July that I did it. I went over the wall Data and. Made analysis on I found out where are the hotspots and which [00:14:00] regions are being affected, the more. The most and everything that I have it on Tableau, I use SQL for that and then used the ability to visualize it. So that's how I'm going about. But the real challenge that I faced in doing personal projects, it's coming up with the questions. What questions should I answer in the process? That's what confuses me a lot because there's a bunch I can do, but I don't know exactly where to start. Harpreet: [00:14:30] Yeah. So out of the like that that's the most challenging part of any project, whether that's personal project or a project on the job, it's like, Okay, what is the question I'm actually trying to answer, right? Like, like clarifying a question, clarifying a problem statement is actually the most, I think, important part of any project, because otherwise, if you don't clarify that, you're going to be heading in the wrong direction. So to clarify a project or problem statement in this case, if you're just doing a personal project, just make sure the the question you're asking is interesting to you. And then just if you've got a bunch of questions, but you don't know which question to start with, Harpreet: [00:15:08] I would say, right? All the questions down and see if you Harpreet: [00:15:11] Can order the questions with the numbers one, two three. So in such a way, that one question follows the other kind of like a chain of questions. Harpreet: [00:15:20] Does that make sense? Elka: [00:15:22] Yeah. And again, Harpreet: [00:15:24] To make sure I want to make sure I understand your question. See, I got a yes. Asha: [00:15:28] Yes, you got it right. Harpreet: [00:15:30] Yes, I Elka: [00:15:31] Will try doing that. Maybe it helps. Harpreet: [00:15:34] Yeah, yeah, definitely. And like, just write about your project when you do it right, like, it's not enough just to do it on Tableau and just leave it there. Like, write a quick, you know, LinkedIn post three thousand characters talking about what you did. Share it. You could even tag me tag, you know, whoever is in this Harpreet: [00:15:53] Room and ask them to Harpreet: [00:15:55] Look at it and get just get feedback from it right? And then you just be open to the feedback, right? Like, don't take [00:16:00] it personally. Just take it as. Constructive. Right. That's another element of it, Harpreet: [00:16:08] Because if you Harpreet: [00:16:09] Just have the project sitting there like you have to hope that people stumble upon you and find it, and you know, this isn't that. It doesn't doesn't work that well. Yeah, positive again. Questions, comments, anybody else got any advice definitely would love to to hear from everyone here? Elka, what about you? You got any tips here? Elka: [00:16:31] Maybe I can jump in. Harpreet: [00:16:33] I think Elka: [00:16:34] If you don't have any inspiration for questions, another way to go can be to like, go to websites like Google and look what they have there, because Harpreet: [00:16:43] There's often Elka: [00:16:44] The problem statement is in the challenge that they post, so it's something to start from. And then as you're doing that challenge, you are looking at the data and other questions will pop into your head. And it may help because for personal projects especially, it's hard to come up with these questions because there's no other sites, right? There's no business stakeholder customer who is asking you for something so you cannot ask them questions to understand what they want. You just have to figure something out. So maybe that's a way Harpreet: [00:17:13] Of finding Elka: [00:17:14] A question that's doable and not like the greatest mathematical questions that are not achievable for, well, most of us, at least somewhere in the beginning of their Harpreet: [00:17:23] Career and still Elka: [00:17:25] Like a source of inspiration that can help. Harpreet: [00:17:29] And by all means, if you want to like, share your project Harpreet: [00:17:32] Like if you want to just even Harpreet: [00:17:33] Like right now, if you want to pull it up or maybe later on in the call, if you want to pull it up Harpreet: [00:17:37] And just show us right Harpreet: [00:17:39] Here right now, we can provide you some feedback as well or come into one of the Friday sessions or any session. Really, if you got like whatever project you're working on, you need feedback or input. We definitely be happy to look it over. All right. Me, if you're talking. Elka: [00:17:57] Yeah, I will not. I will join [00:18:00] the next fighting session for that because I'm not at all right now, so I don't have hold of my project. Harpreet: [00:18:07] No worries. And yeah, we're here Harpreet: [00:18:10] Every week, so Harpreet: [00:18:11] Definitely feel free to Elka: [00:18:12] Ask questions. So I wanted to know that as a beginner, should I focus on two or three skill sets or dig into everything like I'm good at school excel, Harpreet: [00:18:29] But Elka: [00:18:29] I'm not very accustomed to Python in terms of data analytics. So should I focus more on that or try to sharpen my skills on SQL and the skills that I already know or the tools that I already will? Harpreet: [00:18:44] Yeah, I think I would, first of all, the caveat that Data science is like a meta skill, there's not like one skill I can point to and say, OK, you are missing that skill without that particular skill. You are not. That's not how it is, right? Data science is really a matter of skill. It's a combination of many, many, many different skill sets. That being said, we're focusing just on the technical abilities of a data scientist. Having skill is definitely a must, right? That's something you should know how to do. But at the end of the day, there's only like 10 things from school that you really need to know how to do right. At least land your first job. Excel, people sleep on it, but it's very important, so definitely have that skill. So naturally, the next thing is, OK, if you don't have that python ability up yet, it is key. It's a critical component. So definitely learn how to code. You have to. You have to know how to code and you know, Python or whatever it doesn't matter, it depends on. And then where you painted to go, where you're trying to work, what kind of work you're trying to do, but I it's Python, so I'd recommend that Harpreet: [00:19:52] My favorite, my Harpreet: [00:19:53] Favorite website for for like my favorite resource for learning Python for beginners who know nothing about programing is [00:20:00] Python principles. So as a website, I believe it is still free to sign up for Python principles. And it'll just teach you and it's all web browser based, so you write, you know, do all the exercises in your web browser and then from there, just move on to another book called Python for data analysis. That's the book by Wes McKinney, Wes McKinney is the guy who invented pandas. So it's a really, really good book. Yeah, I'd say definitely complete that technical skill set with. With Python and the other stuff he just learned on the job, right? You know. Right. Working out of the terminal like you should be comfortable at least just a little bit, you Harpreet: [00:20:45] Know, working out of Harpreet: [00:20:47] Command line. Not like you need to be able to write bash scripts or anything like that, but just be comfortable with it. And stuff like Dr. Phil, all the other stuff you can learn on the job, cloud stuff, even. But if you get to a point where like, OK, I feel comfortable with everything else, then move on to the harder and harder topics. Elka: [00:21:10] And definitely look into the website, Harpreet: [00:21:15] If anybody else has tips, please do go in Asia, go for it. Asha: [00:21:20] I'd also like to add Harpreet: [00:21:22] That you also Asha: [00:21:23] Have to be very patient to yourself, you want to know everything. A lot of these things you learn while doing a project, you will need to to do something, then you'll have to learn it on the fly. You look it up, learn it then. So a lot of these things, you just learn on the job. A lot of the things you pick up on the job, on your own, you'll know like the basics, but a lot of these things will pick up while doing an actual project. Harpreet: [00:21:46] Yeah. Data science is one of those things that you learn on the job. The thing is like on the job doesn't necessarily mean working at a company Harpreet: [00:21:53] And like Harpreet: [00:21:54] Working in a like environment like professional environment on the job is Harpreet: [00:21:58] Literally in this [00:22:00] field, at least can be done Harpreet: [00:22:02] Without having an actual job. Does that make sense, right? Because everything is freely available, right? Harpreet: [00:22:08] Anybody can Harpreet: [00:22:09] Download. Anybody can get Harpreet: [00:22:10] Set up with a Harpreet: [00:22:11] Sql database on Harpreet: [00:22:12] Their local machine or in the cloud if they want. Anybody can Harpreet: [00:22:16] Get Python for free Harpreet: [00:22:18] Or anybody can download Harpreet: [00:22:19] An I.D. Anybody can get Jupyter notebook. Anybody can find Data anywhere and actually learn on the job by doing. And then Toshi here sent a great. A resource for. Are you solving the right problem from HP? I'll share that as a link on LinkedIn as well. Thank you very much. Harpreet: [00:22:42] Any other tips here for me on that topic? Harpreet: [00:22:51] That's not look like it's so. Charming, hopefully, that was a. Some good tips for you if you got any follow up questions. Please do, please do, let us know we are here to help. Elka: [00:23:02] Yes, definitely. Thank you, Marcia. I will go and thank you Harp for. Yeah, absolutely, giving all the advice, I'll definitely look into it and keep you guys updated on my progress. Harpreet: [00:23:15] Yeah, looking forward Harpreet: [00:23:16] To seeing what you come up with. Shout out to everybody joining us on LinkedIn. There's like 20 people at one point. You guys have questions. Please do let us know anybody else in the room. Got questions? I see hurrying out this back in the building looks like he combed his hair a bit. We're turning on the camera. Thank you very much for doing that, actually. Harpreet: [00:23:37] Hi. I had created I just wanted to share my update as we all interacted with that in to and all, how should I approach that truth? I probably maybe I could share it here or here. Like right now, like I got placed as an intern and I'm into middle of all patients. Well, so I'm an intern. And yes, so [00:24:00] it's definitely like I said this, and she gave like and especially Mark like she said, like how to present yourself in the LinkedIn and showcase like show that your skills at all. I couldn't do that, but I thought, like, it helps me to change my résumé, actually. So instead of adding all unnecessary stuff, I didn't do it in the graduate, and I only focus on my whatever knowledge I had with the projects I did. But the applications I try to explain the recruiter, and even during the interviews, the first thing was my résumé was shortlisted. And I know like I'm into a company and into a project like I do animal operations and my goal. So I'm learning through stuff like democratization and establishing like data pipelines. And I also see like, there's a great future for a lot of options for like because, you know, like the Data changes, like there's a concept draft and the Data and all. Yeah, I'm very excited and I'm also glad that I made Harpreet: [00:25:09] Sure that Harpreet: [00:25:11] A congratulations man, that's awesome news. Like, definitely Ml Ops is the future. Everything you're talking about is super important. I mean, like if you get a chance to play around with comments, platform Harpreet: [00:25:20] Definitely do so because Harpreet: [00:25:22] Like, there's all this stuff you're talking about, like the concept of Data drift stuff like that, Harpreet: [00:25:27] Like, you know, Harpreet: [00:25:28] We got this, it's called artifacts. So it's it's version control and data sets. All that stuff is super important. Experimentation management and all that stuff is highly applicable to the realm of Ml Ops and getting a job as an intern in Ml Ops. That's huge Madison and set you up for a lot of success in the future. Harpreet: [00:25:48] If you Harpreet: [00:25:49] Allow man. Harpreet: [00:25:50] Congratulations, dude, that's that's awesome news. Thank you. And you said it was a mark that better that provides you with some, some good resumé tips. Harpreet: [00:25:58] Yes. Yes, yes. Like [00:26:00] a like actually like I had. I had lots of things on my résumé because I did most of some development, etc., etc. and all. But when I talk like, I think, like I said previously, I was into software and that software engineer. Then I quit that and like, immediately like that resume selection was also it was just a few changes. Like whatever I did, I did the one. I mentioned the three projects. But the key is that what I did and what the concepts I use and to test really like that, and I had a best interaction. Like that interview was my best interview. I would say. Like those straightforward like he asked McCaffery. They support all things you could do at all. And he said, like, I think you are capable of learning like you can deal with them at all patients. And I guess like you got the knowledge of else's for less software engineer skills also. And he said, like, we would like you to take you as an intern for this reason that I'm happy for that. Yeah, I was pretty excited when I forgot who I actually dig deep into it and I was like, Yeah, well, I took these things as like a student. But when are graduates? We don't care. Like, always think of putting a best model to Data, but in the production is what actually match? Harpreet: [00:27:22] Yeah. Harpreet: [00:27:23] So that's the thing. Harpreet: [00:27:25] Like, the modeling is just a very small piece of the entire thing, right? Because they're still like, what happens after the model is in deployment and what happens when everything starts breaking down and melting? Like, What do you do? How do you how do you fix that? How do you Harpreet: [00:27:41] Usually like right now? Like, I was assigned to learn actually so. So I was dealing with some metrics to evaluate my model components like not just the small recall and also like we mainly focus like how exactly if there is like shift, we have some own metrics as well [00:28:00] business metrics like whether the model is performing performance of the business needs and other technical metrics based on our mission and model. How would. If they feel this is unacceptable. But I got to know, like other like usually like my college Harpreet: [00:28:17] When I Harpreet: [00:28:17] Was exploding national model, all I used was actually, I recall a French call. Those are the only three SUVs which are used for evaluating the model. But later on, I found like lots of things like automatic evaluations to occur when there's a Data imbalance, which kind of medical evaluation I need to take care of. And so I actually I cannot disclose a few things right now. Yeah, yeah. No worries. So, yeah, I want it to be accurate as well. So I cannot Harpreet: [00:28:51] Say random stuff. Yeah, know Harpreet: [00:28:54] There's a lot more than just our precision and recall. When you're working on classification, don't forget about log loss, cross entropy and all those things. Thank you very much for sharing that. Very happy for you. That's excellent news. Mark, if you're listening, you helped her not hear with that with that resume, man. So thank you very much. Look at that. That's that's that's awesome, man. Harpreet: [00:29:17] I love hearing and I'm regularly like, I always say your LinkedIn update yours marks. And I just feel like what you guys are up to. You share that feedback on LinkedIn. I always follow that product because those create some kind of ideas and insights. And so I know like I think I recommend people because when it's a beginner like previous channel was also asking like whether you need to have other skills like like, I'm good at school and all. But my journey as a beginner, like some SQL was never really Data Data. I never worked one, but I had it still knowledge. But I think like Python, like if you nail python like, OK, like as a beginner, if you are able to do good stuff with Python, [00:30:00] I think it would pretty much help you in the Data. Thanks very much, because with program, you could automate stuff. Yeah. So that's more important right now, like automating stuff. So and business clients actually want that stuff. They don't want to manual things more. Harpreet: [00:30:17] Yep, yep. Absolutely, man. Well, congratulations again. That's Super Super Super. Happy for you, man. Harpreet: [00:30:25] Thank you. Thank you very much and also thank you, Mark. Harpreet: [00:30:29] Yeah, I know where you're at. He's probably listening. He's probably listening. He's not listening. I'll let him know. I mean, I'm in contact with him pretty frequently, so I'll definitely let Harpreet: [00:30:37] Him know Asha. Harpreet: [00:30:39] Go for it. Asha: [00:30:43] First of all, congratulations. Secondly, I have a question. In regards to how do you deal with when you get into a company, the DBE is so huge, right, and you have a project that learning the whole. I feel like I've asked this over and over, but it's still giving me a headache. None of the there is no dictionary to begin with to any of the columnists is no piece. So and waiting on like writing all the requirements you have, you give it to Data Engineering that's just takes like three or four days to come back on a critical of just knocked it out. How do you do you just sit down initially and just try to go through the whole day, be yourself and try to create your own key because it's too huge? Harpreet: [00:31:32] Yeah, I know exactly what you're talking about because I have been there before, and this is a common situation. Harpreet: [00:31:38] It's like, you know, any Harpreet: [00:31:40] Time you need a Data dictionary, it's really not that this is why Data governance data management is so, so critical and so important. And when people talk about like Data maturity analytics maturity, this is what they mean. It's like, OK, you're not really mature unless you have like something in place where somebody can come in and just see. Without having to bother tons [00:32:00] of people like what the Data is all about with the fields mean what this represents, so on, so forth. So. My approach has been this right, typically, like when you're on the job, you can be assigned to a project, right? And for that project, you're not necessarily going to need every single data set in the entire Harpreet: [00:32:20] Database in order for you to get Harpreet: [00:32:22] Work done. So just kind of focus on one. Like. One corner of the database at a time and try to figure out, OK, if I'm like, let's say I'm a stakeholder, right? Let's just say, Harpreet: [00:32:33] For example, you're working at a Harpreet: [00:32:35] Company and in your company, you're working with people who are, you Harpreet: [00:32:40] Know, you need access to sales data. Right? Harpreet: [00:32:43] The first person to talk to is the salespeople. Right? When you guys have questions, who is it that you go to to ask questions? We need to dig up reports and stuff. Do you have like a go to person and then go find that go to person and then talk to them and kind of pick their brains on? What tables do you seem to work with the most? Harpreet: [00:33:01] You know, you think you can walk Harpreet: [00:33:03] Through some of these tables with me and just kind of help me understand it. And then you will likely have to make the dictionary yourself. And then from there, it's just a chain of questions like, Oh, do you know, like they're going to be like, Oh, well, Harpreet: [00:33:14] You know, somebody else is in Harpreet: [00:33:16] Charge of that table. I know salespeople need that table, but I don't know too much of what's in it. My little thing is just over here. Then you go talk to other people and you start stitching together some of this knowledge yourself. So yeah, it's going to be a lot of. This is where the communication skills Harpreet: [00:33:31] Come in, right? That's what I said, Data Sciences. Harpreet: [00:33:34] There's going to be a lot of communication and a lot of talking to people and a lot of taking Harpreet: [00:33:38] Notes and Harpreet: [00:33:40] Stitching things together yourself. You're going to have to figure out, okay, like what keys connect these two tables if I need to aggregate them? Yeah. So manual detective work is my answer for you. I love to see what anybody else says. Like any any tips Elke Harpreet: [00:33:56] Or Marion, Harpreet: [00:33:57] Or how do you not? Anybody got tips on how to navigate [00:34:00] this? Elka: [00:34:02] Yeah, I would say I would take the same approach as what you said, but I would also say as Harpreet: [00:34:07] You go along documented Elka: [00:34:09] And as soon as it's of a decent amount, like publish it somewhere, even internally so people can start collaborating on it and it it scales, right? So the database is huge, but other people are probably also figuring things out. Maybe not even on the same parts of the database as you. And if they contribute to your documentation, then you can start from what they already figured out in the next project. So then that helps you out in the long run. And also, if other people join after you, they can use this documentation and they Harpreet: [00:34:36] Don't have to start Elka: [00:34:37] From a blank page as you have to. Harpreet: [00:34:40] Yeah. Documenting and putting it in like an internal company wiki or something like that. Super, super important. I'll pause there, actually, just to make sure that we're understanding your question correctly or if Harpreet: [00:34:52] You have any follow ups Harpreet: [00:34:53] On the right. Asha: [00:34:54] Yes, yes. That's definitely I understand it, but I don't know. It's just been a lot of back and forth, and I have these deadlines like immediately I got it and I have to create like right now what I'm working on. I'm creating a customer lifetime value model, a predictable CLV. And after that, the deadline is like the end of the month. After that, I'm creating a recommender system. At four different products with like 10 million transactions a day, so the database is huge. The customers are many and the debate itself is just taking hours to just figure out the working from home. Harpreet: [00:35:29] It doesn't help at all, Asha: [00:35:31] Especially when you're new in a company and it's working from home for a while. It's just not helping. Harpreet: [00:35:38] Yeah, I mean, to get acquainted with the Data Harpreet: [00:35:40] Man I like, especially if they have no Harpreet: [00:35:43] Dictionaries and stuff, you're going to have to Harpreet: [00:35:46] Reach out to people and then talk to them. That's the best Harpreet: [00:35:50] Way. Biggest thing, you. Harpreet: [00:35:52] Biggest thing Harpreet: [00:35:53] I would. I would say that you should not do is just assume that a particular column means something without getting verification [00:36:00] or validation validation because I've done that and. Didn't work out very well. Asha: [00:36:08] Thank you, thank you. Got a headache Harpreet: [00:36:10] Shed? Oh, it is, Harpreet: [00:36:12] It's definitely I mean. Even when I was in clinical Harpreet: [00:36:18] Trials, like it was Harpreet: [00:36:20] Bad for me to say this because clinical trials are supposed to be highly regulated and highly, you know, governance and stuff like that, we still have these issues like what does this column in this Data set mean? Like, how do we figure this out? Harpreet: [00:36:31] So it's, you know, a lot of companies Harpreet: [00:36:33] Don't have that Data maturity yet at that level to make lives Harpreet: [00:36:36] Easier for data scientists. Harpreet: [00:36:40] The sad truth. I wish I had more tips for you, for anybody listening. I know there's a bunch of people on LinkedIn, so if you guys have any tips for us, let us know. Harpreet: [00:36:49] I will relay them Harpreet: [00:36:50] Back to her. Yeah. Just to just be patient and ask questions and grind through it, I mean. And you push back on these deadlines at all, like like, where are these deadlines coming from? Who's like, you know, saying that it has to be done at this point? More background than that. Asha: [00:37:14] So a lot of the times it would be like people from random departments like coming in or you're telling me, I need this already done like yesterday. Like, Yeah, OK, then someone else just. Harpreet: [00:37:26] Yeah, so this is going to have to be like you have to kind of push back. All right, cool. I understand you needed it done yesterday. Maybe you should have came to me like two weeks from yesterday and we could have talked about this. But this is where you are right now. Let me tell you what I need to get done in order for me to make this happen to you. It's not like I just push a button and stuff magically appears. First off, I need to figure out where the Data is, right? Can you tell me where it's at? Can you tell me what all these columns mean? Because without me knowing all this stuff like it's going to be hard for Harpreet: [00:37:52] Me to do actual work for Harpreet: [00:37:53] You, right? There probably be like, Oh no, I have no clue. Or like, All right, well, that's going to take at least a week right there [00:38:00] for me Harpreet: [00:38:00] Just to figure out what, what Harpreet: [00:38:01] Data I need with the columns I need are. And then on top of that, then I've got to do exploration of the Data. I got to understand it, right? And that's in Harpreet: [00:38:11] Taking another Harpreet: [00:38:13] Three or four days, almost Harpreet: [00:38:14] A week. And then from there, Harpreet: [00:38:15] It's conceptualizing what type of model I'm Harpreet: [00:38:17] Going to use to do Harpreet: [00:38:19] See, you know, customer lifetime value, Harpreet: [00:38:20] Right? Harpreet: [00:38:21] Because here are the different ways we could do it. I need to first understand the business model like, is it? You know? You know, there's four there's different types of interactions that customers have with their business, they could just, for example, a furniture store. People don't typically go to the furniture store Harpreet: [00:38:40] Every week, right? Harpreet: [00:38:41] They buy it an incremental time periods across long delays, right? Whereas somebody had a grocery store typically coming every week, right? So I've got to understand the business model because once I understand the business model, then I could figure out which lifetime valuation model is going to Harpreet: [00:38:56] Make sense for us here. Harpreet: [00:38:57] Right? That's going to require some research. Harpreet: [00:39:00] So right now, you know, Harpreet: [00:39:02] With you coming to me with this request, I'm already in need at Harpreet: [00:39:06] Least two and a Harpreet: [00:39:07] Half weeks just to get oriented. And then from there, the modeling work, you know, depending on how large the Data Harpreet: [00:39:13] Is, I've got to figure out if Harpreet: [00:39:15] I need to write optimize code, if I need cloud compute resources, things like that, right? So this is where you start Harpreet: [00:39:20] Pushing back as a data Harpreet: [00:39:22] Scientist and be like, All right, cool. Hold on, man. It's not like a magic button. I push and shit Harpreet: [00:39:25] Just happens like I need Harpreet: [00:39:27] To figure stuff out first, right? And yeah, just. The stake in the ground be like, yeah, hold Harpreet: [00:39:34] On, man. You don't come Harpreet: [00:39:35] And tell me how long it takes me to get the stuff done. That's not how this works. So. Man, go for it. Marianne: [00:39:47] That's sort of welcome just from previous experience, not Data science, but the high tech wireless communication students. I have a question for you when you start [00:40:00] working on a project, typically from my experience to have all the stakeholders coming together in a meeting and sort of identifying what we're trying to achieve. What is our goal? What is the problem? What are the critical points? What are the choke? I mean, Harpreet: [00:40:17] What do we need to Marianne: [00:40:18] Basically solve this problem? Do we have any sort of idea which also what kind of things we need to do that if we don't have that in front, then it's very difficult. I mean, you just shouldn't accept anybody coming to you and ask you, Hey, when can you give me this? That this stuff is ridiculous. If you go into that cycle, you'll burn out very quickly and. Nothing is basically planning when everybody is involved in the everybody that is relevant for the Harpreet: [00:40:55] Project or the problem Marianne: [00:40:57] Should be involved. And also another tip. You're not alone. Do you have a supervisor share with him what is going on and how you can handle the issues? That's my advice. Not from that standpoint, but from experience in different areas. Harpreet: [00:41:17] Thank you very much, Mary, great, great tips there. And on the other end of the more technical aspect of it for, you know, if you need tips on customer lifetime valuation stuff, there's a couple of packages in Python. Harpreet: [00:41:30] One is for Harpreet: [00:41:31] Churn and that's called Harpreet: [00:41:32] Lifeline's. Harpreet: [00:41:35] There's two of them lifelines and lifetimes. Harpreet: [00:41:38] Both of them by Harpreet: [00:41:39] The same author, both Harpreet: [00:41:40] Of them written by Harpreet: [00:41:42] Shopify. Harpreet: [00:41:44] So Lifeline's Harpreet: [00:41:46] Lifetime. And then there's different behavior models you can use for customer lifetime valuation. Harpreet: [00:41:51] One is called Harpreet: [00:41:53] By till you die. So you can look into research done by Bruce [00:42:00] Fader and something hearty so fator and hardy that are professors out of Wharton School of Business. And they they're like the gurus when it comes to lifetime valuation and customer behavior, things like that. So definitely check those resources out. The, I think, just a lifetime to Lifeline's package, either either one of those two packages has a tremendous amount of documentation that is going to be extremely helpful to look into. And then like in terms of like statistical models, there's like the Harpreet: [00:42:30] Perito negative binomial Harpreet: [00:42:33] Distribution. Harpreet: [00:42:34] Then there's the like beta gamma negative binomial. Harpreet: [00:42:38] So these types of statistical models are useful or the scenario Harpreet: [00:42:44] Beta Harpreet: [00:42:44] Geometric or something like that as well. Austen, go for it. Austin: [00:42:48] Yeah. I was thinking about as Harpreet was talking, there's a sort of social component of what you're talking about. Like, those are difficult conversations to have. It's difficult to like, especially when you're new to be like forceful about that. And I think one of the things you're trying to do early in a role this is not just you, but generally is like build some sort of credibility or ethos around yourself and your expertize and like, prove that you have that expertize. And I think what Harp three is sort of getting out there. It's like the more specific you can lay out that process Harpreet: [00:43:14] And what each part of that Austin: [00:43:15] Process looks Harpreet: [00:43:16] Like. Even if someone doesn't Austin: [00:43:17] Understand it, you're explaining it in a way to them that's like, Oh, this person actually has a theory of this and has a sort of paradigm through which they view this and understands the work. And that sort of feels like the first step to it. Also like establishing that credibility that will help you in the longer term as well Harpreet: [00:43:35] And sort of Austin: [00:43:36] Eliminating that sort of feedback loop where you're just like sort of accepting that deadline that's been given to you and then struggling internally or maybe with one other, your supervisor, whoever else, the more specific you can be in laying out the the process and then identifying the roadblocks within that Harpreet: [00:43:51] Process, even if someone doesn't Austin: [00:43:53] Understand it from a technical standpoint, I think that that's the best one of the best ways you can build that ethos and credibility so that you can start [00:44:00] to break through that. And it might not for this first project. It might not get you all the way to your like your goal of having more time and this and that. But it sort of just lay that groundwork so that you avoid that cycle of that's going to burn you out over time or make you just like, I can't do this. So I'm just going to do a shitty job of this. And then that leads to, you know, like that's I think that's the danger. If you if you're not specific and you don't like work to build that credibility. So I think that's sort of what's going on socially underneath the surface of all that stuff. Harpreet: [00:44:27] Yeah, absolutely. Very, very well. Put that exactly what it was that I was trying to communicate. Thank you for that. Austin, I'd love to hear what what look like. You got any tips when it comes to dealing with aggressive stakeholders and they're like tight deadlines. Elka: [00:44:44] Yeah, I think I agree with what you and Austin just said. I get that it's hard, especially if you're new in a in a new project or in a new company, but being assertive about it as long as you know what you're talking about, of course, is it's just a must because everybody is in their own stress, in their own projects, with their own deadlines. So they, as long as they depend on other people, they are going to push that stress forward. And at one point, it's just not doable anymore. And it's basically a choice between being realistic right now and maybe pushing their deadlines back a bit too or disappointing people at the end line. And that's even worse. So if something is not feasible or you feel like you're not sure about a deadline, it's better to communicate. Harpreet: [00:45:32] At least in my opinion. Elka: [00:45:33] It's better to communicate this as early on Harpreet: [00:45:36] As you can with the Elka: [00:45:37] Right arguments. Of course, like this is going to take specifically. This part of the task is going to take more Harpreet: [00:45:42] Time or I don't even Elka: [00:45:44] Know how to start on that one. So I need to figure it out and get out before I Harpreet: [00:45:47] Can so Elka: [00:45:49] That everybody is clear on it. And especially if you stay with the company for a while and they see that, OK, we're giving that person deadlines. Sometimes she pushes back, but then [00:46:00] at the end, she always delivers by the time she says she's going to deliver it, and you also get a sense of they start to trust you. And so it will be easier to get that data that you're asking for. So, yeah, it's a process, but it's important to start pushing or working at it from the start, I think. Austin: [00:46:21] And like and like the outcome isn't like for this project, you get every thing fulfilled that you need. It might be that like if you document like what it is you need, you communicate what it is you need to say. No dice. This is still the deadline. Then you have something to point back to a conversation, a document, a plan, whatever it is to point back to and say, Hey, this is why we're not getting the result out of this project that you wanted. Like I had this clearly stated so like, you might still have to go through it, like you might still have to go through that. Put something in front of someone that's not your like. What would be you would consider as your best work or the thing that solves the problem like that might just be the nature of how this has to work. But like, I think it's like building this case and then documenting it and being clear about that process so that even if it doesn't work out in this project, you're building a foundation to to make those social interactions serve you better and serve the company better and what you're trying to get done. So I think like it's not focusing on like, well, I tried that in the outcome still remain. It's like you're you're building that process and it might take some time and depending on the size of the organization and the different stakeholders and all that kind of stuff, like some of those things are going to be out of your control. But it's like, what is the best thing I can do to sort of build this moat around my expertize, my ethos, my ability to contribute in the medium to longer term with this company? Elka: [00:47:34] Yeah. And sometimes also there's options, right? You can, for example, say, OK, you want these three things I can deliver them like at that date or if you want an earlier date, I can either do and B or C, but not all of them. And then they know they can tell you what their priorities are. So you also know what to start on. Harpreet: [00:47:55] Excellent tips. Thank you very much. Austin Elk, Asha. Go [00:48:00] for it, if you have a follow up question. Asha: [00:48:03] No, thank you, thank you so much. Oh yeah, Harpreet: [00:48:06] I do see a follow up question here. Are there any software that you use for documentation that's easy to add a lot of details to you? Does your company have like an internal company wiki of any sorts? Asha: [00:48:20] I mean, the normal documentation, a lot of it we've been doing has been there's a Microsoft Harpreet: [00:48:26] Sharepoint in SharePoint. Harpreet: [00:48:28] Yeah, it is Harpreet: [00:48:29] Sharepoint on Microsoft, isn't it? Sharepoint? Asha: [00:48:32] The name is Gordon is one, but it doesn't Harpreet: [00:48:34] Capture Asha: [00:48:34] Everything that I'm trying to capture, like the steps and the process is how long each step takes. Like, I'm just trying to capture the whole lifetime of the whole beginning to the end. Yeah. Harpreet: [00:48:44] So if you're using Microsoft, that means you probably use Teams as well. And then inside Teams, Harpreet: [00:48:51] You can Harpreet: [00:48:52] Set up like a workspace and teams like for your data science workspace, whatever. And then each project has the project wiki associated with it. I usually just put everything in that in that project wiki. Yeah, I wonder if I. Oddly enough, I still have access to teams from my old Harpreet: [00:49:12] Job, like Harpreet: [00:49:14] What if I pull it up and show you? But that's probably not ethical, I won't do that. Harpreet: [00:49:18] But yeah, Harpreet: [00:49:19] You can just the wiki wiki inside teams and then you could just tag people or just send them a link to that wiki inside teams. But yeah, Microsoft uses SharePoint as well. A comment would we use these notion notion is a good one. Confluence as well. Yet Austin says your notion of zero wikis, maybe a Gantt board, something like asana asanas. Nice as well. Now, everybody here is talking about Jeera as well, so those are some good solutions. Thank you. Awesome. If anybody has questions, go ahead, let me know. I see a question coming in from a LinkedIn. I'll go to that question on LinkedIn. But in the meantime, anybody in the Zoom room here, [00:50:00] if you had a question, let me know I will kill you up. Question coming in from Harpreet: [00:50:05] Abdul on Harpreet: [00:50:06] Linkedin. I wish to work physics with artificial intelligence. Is there any opportunity for me Harpreet: [00:50:11] If by Harpreet: [00:50:12] That Harpreet: [00:50:12] Question, Harpreet: [00:50:13] You mean, do I know of any jobs that you can apply to right now? Not off the top of my head. But if that question means that, is there an intersection of those two? I say yes, very, very much so. There's so much, so much machine learning Harpreet: [00:50:30] Used in the world of physics. Harpreet: [00:50:33] If you're still listening to Abdoul, check out the conversations I had with Shantha Natalie she. We talked a lot about how she uses machine learning as a physicist. Harpreet: [00:50:45] Lawrence Marini. Harpreet: [00:50:47] He's a podcast as well. I forget the name of it, but he he also is a PhD in physics that uses a lot of machine learning. Mayor knows you have a Ph.D. in physics. Marianne: [00:51:01] Yeah, I have is in physics, but unfortunately, we learned about machine learning much later, so it could not. Yeah. Harpreet: [00:51:10] Like, I was watching a video yesterday on YouTube, and it was about how recently they were able to see the energy or information emitted from the back side of a black hole and the way they're able to stitch together the image of the black hole was through machine learning. So I thought that was really fascinating. Harpreet: [00:51:32] So, you know, Harpreet: [00:51:33] Even in particle physics, there's a lot of use for machine learning, like Shantanu was talking to me about it in particle physics, where they're trying to identify a particular phenomena that happens like Harpreet: [00:51:43] Typically in normal Harpreet: [00:51:46] Machine learning. Let's say we're doing credit Harpreet: [00:51:48] Card fraud Harpreet: [00:51:49] Detection. We have to do upsampling of Data Harpreet: [00:51:51] Because the class that we're Harpreet: [00:51:52] Interested in is so few and far between. So we need to do, you know, Harpreet: [00:51:56] Have a smote in physics. Harpreet: [00:51:57] They have the opposite problem [00:52:00] where they have to downsampled data because there's so much noise in the data. Harpreet: [00:52:04] And all of that Harpreet: [00:52:04] Is just the machine learning. Say, I think the intersection of physics Harpreet: [00:52:09] And and machine Harpreet: [00:52:10] Learning Harpreet: [00:52:10] Ai is huge, Harpreet: [00:52:12] It is massive. The opportunities are abound. So I'd Harpreet: [00:52:18] Start there like the website Harpreet: [00:52:19] I just released the other day with Max Frenzel. He's a PhD in physics. He's doing some crazy like quantum Harpreet: [00:52:26] Information theory, Harpreet: [00:52:28] And he was trying to figure out the smallest Harpreet: [00:52:31] Machine Harpreet: [00:52:32] That you can use to do computations on. And we talked about the intersection of physics and AI in that conversation as well. Harpreet: [00:52:40] And coming up in the Harpreet: [00:52:41] Future with some of the authors I'll be bringing on, we'll be talking about this intersection as well. Harpreet: [00:52:44] So stay tuned to the artist Data Science. Harpreet: [00:52:49] Ok, here has a comment, then we'll go into a Caroona question by Elke saysthat about wikis for documentation. The same holds true for Azure DevOps as far as Microsoft Teams. And since it's both, Microsoft integrates nicely together, but the wikis are yet to repos under the hood, so you can work on them from your ID, for example, in markdown. Yes, that's true. I also used Azure DevOps as well when I was that price, and I love that wiki functionality. Harpreet: [00:53:17] Markdowns, awesome. Get comfortable Harpreet: [00:53:19] With it. It's nice. Are you still here? Yes. Asha: [00:53:26] Thank you. Thank you for giving me this opportunity. So basically, I have I'm working as a product analyst currently, so my job is to get the data and all the insights and tell about the requirements of future development next feature development that we would be doing. So in that process, I mostly I get the Data and I. I pull out the data for the people who are in the operations team or I just get [00:54:00] normal Data and I look for some questions that I'm having and I try to pull data out of that and try to answer those particular questions. But I have never been able to understand the theoretical statistical techniques that you learn of, like hypothesis testing and so on and so forth. Descriptive analysis and inferential statistics. How do you actually use that or is that actually being used at workplace? Can that make my insights better? And how do I actually use that? I have no clue. Harpreet: [00:54:38] Yeah, so in terms of statistics for Harpreet: [00:54:40] Products, right, you can have, for example, let's say you are working on. It's the easiest example, let's say, website design, or maybe you're changing the color of a button or something on a change in the color of a button for call to action on a Harpreet: [00:54:56] Particular Harpreet: [00:54:58] Device or website, right? So you can test if two different colors have a different click through rate, right? That's like a B testing. So a b testing would be kind of the biggest the use case of doing this type of hypothesis, testing and inference. And I think the product world. Right. Right. Harpreet: [00:55:20] So you likely heard of that. Asha: [00:55:24] Not, I mean, not by the particular heading that AB testing, but yes, I do this. I compare things that if this was there, then what was the result? How did the user react to that and so on? Harpreet: [00:55:37] Yeah, yeah. So the same thing now it's like, OK, well, how Harpreet: [00:55:40] Do you like, you know, how do you tell that they're Harpreet: [00:55:42] Actually different? Well, that's where the statistics comes in, right? That's where you can use some type of hypothesis test using some statistical testing procedure. I don't know, just listening to the normal T test right now, right? And then you can make a claim and say that, you know, I'm inferring from this sample that [00:56:00] the larger population will have a higher click through rate based on this test that Harpreet: [00:56:04] I ran, right? Harpreet: [00:56:06] You know, 10 people like, you know, Harpreet: [00:56:07] One hundred and ten thousand Harpreet: [00:56:08] People got Harpreet: [00:56:10] Option red and maybe Harpreet: [00:56:12] 8000 people got Harpreet: [00:56:13] Option blue. Harpreet: [00:56:15] And then you run a test and then you say, OK, well, the results for, you know, this blue button was statistically significant. Therefore, I can make a conclusion. Harpreet: [00:56:27] I'm like greatly Harpreet: [00:56:28] Simplifying it, but therefore I can make a conclusion that, you know, the general population will Harpreet: [00:56:34] Like blue better, like Harpreet: [00:56:36] Greatly simplifying it. But yeah, a b testing is where you'd use this, you know, hypothesis test and things like that in the real world. Asha: [00:56:43] Right, right. Got your point. This works, I mean, I also have another question. Can I say it here or should I wait for another chance? Harpreet: [00:56:53] No, you have to come back next week. I'm sorry. No, I'm joking. Asha: [00:56:57] Go ahead. So again, but can anyone please suggest me? How should I start learning model deployment? I have worked on scraping data on creating data from the very scratch, creating data from real time data sets. I mean, collecting data, primary data research. So then now I'm working on data analysis. I have done internally for my learning. I have done projects using machine learning techniques. I haven't deep dove into neural networks yet. But yeah, for machine learning projects that I have, how do I start learning model deployment? What would be the best way to start learning? Harpreet: [00:57:42] I think the best way is just to forget about model deployment for now, because it doesn't seem like that's part of the value proposition of where you're like, you're doing a model deployment. I would kick that over mostly to machine learning engineer type role, right? It's important for a data scientist to understand for sure conceptually [00:58:00] how it models get deployed, but like for product analytics or, you know, product management type roles, I think that it wouldn't be in your realm of responsibility. I'll pause there, maybe I'm not understanding that. Asha: [00:58:13] No, you're right, it is not my responsibility, but I want to take my career ahead in that, so I would definitely want to have my career in Data. So does a data scientist require model deployment? Obviously, they need to know a good understanding of that, right? Harpreet: [00:58:29] So, yeah, so you have a resource at the top of my head. That I could point you to for model deployment. Do you have any specific questions, maybe I could try to answer that. But if anybody has like a high level review of the model Harpreet: [00:58:45] Deployment process resource that you can send, but please do Harpreet: [00:58:49] Share it here. But if you have like a specific question about model deployment, I can Asha: [00:58:54] Not just dipping my toes in so I don't really know what. Harpreet: [00:58:59] Yes. Yeah. Like deployments like a huge spectrum, right? There's deployment in the sense that it's deployed on my machine, like on my phone, so that when I take a picture of an object, it says it's a hot dog or not a hot dog, right? Like that, that's one type of deployment. But then there's also a type of deployment where all you're doing is just Harpreet: [00:59:18] Pushing your predictions Harpreet: [00:59:19] To a database so somebody else can get it. Or all you're doing is pushing your predictions to a CSV and just sending that out every week. So I guess something that comes to mind it might be a Harpreet: [00:59:34] Little bit too might be overkill, Harpreet: [00:59:35] But like full stack deep Harpreet: [00:59:37] Learning, Harpreet: [00:59:39] That's I mean, that might be a lot of overkill. I'll pull it up right here just to show you. Sure. My fear, though, that it could. Me too much, and this is all about, you know, what it means to deploy, so it's all about shipping projects, right? Ok, so no. Yeah, it's [01:00:00] all about what model deployment means. Asha: [01:00:04] Oh, sure, I'll check it out, Harpreet: [01:00:05] Definitely check this out. Asha here says PML, I don't know what that is, but again. Iml, that's probably not what she's talking about. Yeah, actually, if you have a link. Asha: [01:00:24] Let me find the link and attach it. Harpreet: [01:00:26] Yeah, yeah. Model deployments a huge. A huge spectrum, right? There's a huge spectrum of shipping Harpreet: [01:00:33] Machine learning projects, it just Harpreet: [01:00:35] Depends on what your particular. Asha: [01:00:37] How important it is to have a user interface to deploy a particular model. Harpreet: [01:00:45] It's important in the cases where your end user will need a interface. So you could like you could deploy something like on stream lit, right? And then all the all a screen lit app that let's say you're trying to predict housing prices, right? So you can you can deploy a simple linear regression on screen lit and somebody comes in just interest. The features of their house hits calculate, and then it spits Harpreet: [01:01:12] Out a, Harpreet: [01:01:14] You know, a value for the house, right? So that's something you could do. And you know, that's useful for those situations where your end user is going to need an interface. That's not always the case, right? Sometimes your end user is surfing the web on Amazon and you're pushing recommendations to them without them having to interact with anything right or your end user is watching Netflix and movie recommendations are popping up right? So that's a different type of deployment. Harpreet: [01:01:42] So deployment Harpreet: [01:01:43] Is that definitely spectrum Austin here, says a stream lit Harpreet: [01:01:47] Gallery. Harpreet: [01:01:48] That's that's an excellent resource to look at kind of the spectrum of deployment. Asha: [01:01:54] Oh, sure. That is my question. Thank you so much. Thank you, everyone who pitched in to the place. [01:02:00] Thank you. Harpreet: [01:02:02] Sorry, it couldn't be more specific with that question because as you can tell that it's a huge, huge right. Asha: [01:02:10] If we just talk learning, maybe I can come up with better questions then. Yeah. Harpreet: [01:02:16] Like, I deployed something locally where it was like classifying, you Harpreet: [01:02:20] Know, tweet sentiment, right? Harpreet: [01:02:22] And that was kind of it required a user interface because then I could just, you know, open my tweet and it'll give me a score for the sentiment, right? Yeah, hopefully. Hopefully, that's helpful for you. A bunch of great resources here. Pml is an attempt at standardizing model formats for deployment across environments. I definitely would love to learn more about that. I'm surprised I have not. But yeah, extremely it. I think is probably one of the the easiest way stream. That radio is nice as well. So those are two great places that you can play around with model deployment. Harpreet: [01:02:59] Sure. Thanks. Yeah. Harpreet: [01:03:04] All right. Any other questions coming in, I don't see anything on LinkedIn, I don't see anything in the chat here. Last call for questions. I should go for it. Asha: [01:03:15] Better question, it's a congratulations to you and your new role. Harpreet: [01:03:18] Oh, thank you. Thank you. Thank you. A lot, a lot of good stuff happening for for everyone, man. Harpreet: [01:03:23] A lot of good. Good news with the entire community. Harpreet: [01:03:27] You know, a bunch of my mentees also landed jobs, a bunch of people on Friday. We're talking about how many new jobs people got. So I think it's Harpreet: [01:03:34] It's a good, Harpreet: [01:03:36] Good season for everyone. I'm totally excited about this, that new role, that comment. It's definitely. That is the best type of rule for me, I would say this to say that I guess that's exactly what I've been wanting to do. And I'm excited for it. Harpreet: [01:03:53] A lot of learning to do, but Harpreet: [01:03:54] We've got a lot of awesome things planned on the horizon for all of you guys, [01:04:00] so I'm excited for that.