25OH-08-08-21_mixdown

[00:00:05] What's up, everybody, welcome, welcome to the Comet ML Open Office Hours, powered by The Artists of Data Science. It is Sunday, August 8th Super excited to have all the guys here. Hopefully AIs got a chance to tune into the podcast

[00:00:17] That was released earlier

[00:00:18] This weekend on

[00:00:20] Friday, had an episode with the one and

[00:00:22] Only Jonathan Tesser. I think you guys might recognize him from LinkedIn. He's got a lot of awesome posting. That was really enjoyable conversation I had with him. So definitely go check that out. Some other big news. I had my last day at Price Industries on Friday,

[00:00:36] And I'll be embarking on a

[00:00:38] New journey here with my friends, that comet Miles. I'm super excited about that. Also, thank you guys so much for helping us cross that seventy five thousand download mark for the podcast. We were able to to crack that that we couldn't have done without you guys. Thank you so much for your support. Yeah, super excited superexcited. Have everybody here shout out to Merrion Mohammed Mark. Awesome. What's going on, man? So we are live. We're taking your questions. Have you guys got any questions whether you are joining us here on Zoom Room or on LinkedIn or YouTube or Twitch? I'm keeping an eye out for all the questions, so go ahead and let us know what your questions are. And actually in the link for the description of the video, if you want to join in on the live session, you can just click on that link and I'll take you

[00:01:23] Right into the room.

[00:01:24] So definitely excited to have you guys here, has everybody doing man. What's that? What's everybody up to, man? Mark, how you been, man? How do we combine?

[00:01:33] The busy been working on a case study, just trying to figure out various Data problems, got to play around some Montecarlo simulations, it's just super fun. I've been wanting to do it for a while, so it's having a good time playing with Data.

[00:01:50] I haven't done Montecarlo simulations in such a long time. And those are those are a lot of fun to do those in the actual days when they're trying [00:02:00] to simulations and whatnot. Obviously it's what we used to call for. What's the what's the case? What are you doing this for? Like a for work or like a personal project or just the

[00:02:09] First personal project, which is basically, you know, my friend a couple of years ago, he's a Bayesian thinker. And he started showing me Montecarlo simulations and blew my mind because he basically did the typical thing like, hey, what's the probability of flipping heads? Right. It's 50 percent. But then you flip it 10 times and it's like seven to three. Like what's happening here. Right? I think it's like a really cool way to visualize. I'm a visual visual learner. So I think it's a really cool way to visualize distributions and think about probability. Yeah. And so right now I did a simulation for like a craps game and then ran a Monte Carlo on that. And then depending on the various biased, I do like where the potential outcomes and also like potential

[00:02:58] Money you can

[00:02:59] Make from it. So I'm getting inspired for like a cool tutorial to put out later.

[00:03:04] Yeah, my whole thing when I go to the craps tables is it's always six and eight or potentially nine because they have the highest probability of turning up the heat, the die one hundred times, then you'd expect to see those numbers come up far more frequently. Oh, that's interesting thing that you're talking about. Bayesian thinking, Bayesian epistemology. I've been thinking a lot about that as well. I came across this podcast called Increments, and it's with Vaden Masrani, who is a PhD student at University of British Columbia, and I think his name is Brian Shaggs or something like that, who's he's a computer scientist now going into law and had this series of episodes. And that was all about the philosophy of probability. And so there's three episodes back to back. And I've been listening to, like this one episode, um, like multiple times just because it's like mind blowing my mind that the stuff they're talking about. So it's really, really. Interesting [00:04:00] stuff, man, I can send you link that podcast if if you're interested.

[00:04:04] Oh, that would be great because I don't think very basic background is like a foreign language to me because, like, I think a researcher will allow its frequencies that they are making that switch for some people for that base thinking. But like I think statistics, I think your frequencies perspective.

[00:04:21] Yeah, I've kind of adopted this motto for myself

[00:04:23] Where I'm Bayesian

[00:04:24] In my own personal decision making, but I frequent when it comes to experimentations. And I was reading this interesting paper. Uh, I'll send you a link to the paper as well, but it's called Bayesian Frequent Tests. And it's these people who are like me who think bagian like by default, but then apply frequenters tendencies to the experiment, as I think you will really, really, uh, enjoy that. But speaking of being a visual learner, man, like. Like, that's huge for me as well, like I used to think that I wasn't a visual learner until I started seeing stuff presented visually and like, it makes so much more sense. Uh, so part of the you know, as you guys know, I'm starting the twenty one days of a deep learning starting tomorrow, uh, which is going to be a fun and exciting challenge.

[00:05:09] Um, and I'm heavily relying

[00:05:11] On a lot of visual type of material. So there's deep learning illustrated, which has been phenomenal. Uh, grokking, deep learning. And this is all just it's awesome because they have like all these like cool like. Sketches and and. Little cartoons in there and there's this one, dude, this book is massive, it's like literally,

[00:05:32] I think about six pounds,

[00:05:33] But it's about seven, 700 something pages. Deep Learning illustrated by Andrew Glassner, who's super, super entertaining. I saw some. Deep learning crash courses with him online. This guy's funny man, and he told me to get his book and it's been super helpful so far. Seven hundred twenty three figures and no math in it at all. Um, just beyond basic. Arithmetic and [00:06:00] stuff like that, he really makes it really simplified. There's something there, but not too much, but it goes deep from it starts with just the basic fundamentals that moves all the way into deep learning. So I'll be leaning on these three resources for the 21 days of deep learning. None of the content I'm creating is pre crafted. Everything will be done the day of and I'm really time boxing myself to three hours to create the content, two hours to kind of study the material, take notes, and then one hour to create some type of content, whether that's a write up or infographics or something along. Those lines, I think that will just help get me prepped for this, and you like Comet will just be can I put content

[00:06:40] On a on a deadline

[00:06:42] And and just getting into that mode, hopefully AIs will be joining me on that journey starting tomorrow. Two days of deep learning. We've got a whole. Also, we will have that. That being said, man, let's go let's take some questions for the audience that I'm excited to I'm excited to take some questions about Muhammad saying he's a newborn in the field of data science and programing. Just here to learn new stuff. Awesome, man. Hey, well, if you have questions whatsoever, please do let me know. Be happy to take on your questions.

[00:07:11] Mohammed, we want you want

[00:07:13] You to join us for this conversation. Tell us a little bit about where you are coming from in your journey like you new in the field of Data science programing. Bill, what have you been doing up to until now?

[00:07:26] Well, it's basically been like a week now, I'm quite new to this, it's like I'm really excited about this. I've just started a 14 day challenge on my part. And basically I know nothing about program.

[00:07:42] If so. So it's quite

[00:07:45] Exciting.

[00:07:46] Yeah, awesome. And that's that's good. I'm glad you're excited about it. Just keep the momentum going, keep that enthusiasm up. And I would say like my absolute favorite resource, but I'd like to send people to when it comes to

[00:07:57] Learning programing, it's a

[00:07:58] Free resource too, in [00:08:00] this entirely Web based. I'll pull it up in a second here. It's entirely Web based. So you don't have to worry about downloading the stuff and getting the packages and all that stuff. You can just focus on learning the syntax. And it's called Python Principles, and I'll pull it up right here. Um. It's the pipeline principles and

[00:08:19] They usually Haji's, but they're given

[00:08:23] Away. Um. Free subscriptions, and I think even just like the charge is nominal, it's like 30 bucks, whatever, but, um, have. Password here for this or not, we'll see. Um, yeah, I do. And it takes you through the entire basics of, um. Python syntax, though, is definitely a useful

[00:08:46] Resource, and just to kind of

[00:08:48] Show you how it works, let's go to this introduction. It'll have this and then

[00:08:51] You have your code

[00:08:53] Here and the output here. So everything is entirely Web based. So I highly, highly recommend this. And it gets cool because after you go through all the modules, you can go to the challenges here. And some of the challenges are really fun and interesting to do the, um, your. Ability to write interesting functions, and this will kind of get you prepared for those coding challenges that come up in Data science interviews, I mean. This will help you build the confidence to do those type of problems. So definitely check that out. Pipeline principles, dotcom hands down one of my. Most recommended resources. There it is right there. There's a link for that.

[00:09:32] Um, what are you

[00:09:33] Doing up until, uh, up until now, what's, uh, what's kind of your background, where you come from? In terms of like education, work experience, stuff like that.

[00:09:43] Mohammed. Oh, yeah,

[00:09:46] Well, basically, I just recently

[00:09:49] Graduated with an engineering degree, but I realized

[00:09:52] That everything is just Data centric these days. So I thought, why not? I have a little bit of free time, so why not learn a [00:10:00] little bit about machine learning and programing and

[00:10:04] Exploring it on. What kind of engineering in particular were you studying? Civil engineer, OK, so I would recommend doing this, too, as well, right? If you have that undergraduate studies, graduate studies, how

[00:10:17] Far along are you in that

[00:10:20] Undergraduate? OK, so I would I would recommend doing this man like looking at case studies that involve Data science machine learning with civil engineering. So you could do this, right, civil. Civil engineering and then machine learning and then look up. Uh PDF AIs if they're white people, I don't know what happened there, um, and just check out some of this stuff like look like these guys got a book Machine Learning Techniques for civil engineering, uh, machine learning and data analytics for civil engineering. So this is looks like a class syllabus, but still

[00:10:57] It will make

[00:10:58] You feel like you're not starting completely from scratch if you're able to find the intersection of what it is that you already know how to do with Data science and then proceed that way. And then I think that might even help with, um, keeping up with the enthusiasm, that motivation. If you're like, oh, OK, look, this is how this thing, um, works in my field.

[00:11:18] Right. And you can start to view things from that

[00:11:21] Perspective if that makes sense. So you can remove the PDF part and just look for, you know, probabilistic machinery for civil engineers. That's the book. But, yeah, that's that's what I recommend as well, looking at that as much as possible.

[00:11:37] Well, company want to see, like, oh, sorry. I was saying a cool company, if you want to see, like civil engineering and practice or like machine learning, is this company called one concern? Essentially, they use Emelle and various like sensor signals to understand a city's infrastructure [00:12:00] and kind of predict the impact in the hot the the high

[00:12:06] Risk areas

[00:12:07] For natural disasters. And when natural disasters happen, where to where to prioritize resources. It's really cool. I think a few years ago, the founders came and talked, talk to our class. And it was just really amazing how they they approach their PhDs from from Stanford, who just kind of stumbled on is like a thesis kind of class project. And then it's turned into a company and they start getting government contracts.

[00:12:35] That's pretty cool, man. What's it called, one concern.

[00:12:39] One concern. I got put in the chart real quick, but I thought it was really cool because it takes all that sensor data from civil engineering and this like does the city's landscape and all the intricacies of that analyze it and is able to simulate like a like a flood happened. Right. What areas do you prioritize first?

[00:13:00] Yeah, I was actually I was interviewing for a company that that kind of did something similar to this one concern, but also as an intersection with civil engineering,

[00:13:08] As they're called.

[00:13:08] Urbino, you are being or something like that.

[00:13:12] Um, so check

[00:13:13] Check them out as well. Mahamad, see if you can look through their blogs or postings to see what kind of work they're doing. And, you know, just to to. Get good at that intersection of machine learning and what you already know. It just makes it much more approachable. Um, but yeah, I'm looking forward to, uh, to taking everybody's questions about somebody, ask questions, go for it. I'm monitoring all these streams here. I don't see any questions coming in. Just some, uh, some nice comments. Uh, Ben saying that we're troupers. Research the Data community. Thank you very much, Ben. Or is saying he's roughing it out in twenty eight degrees Celsius and the sun in France is suffering for all of us. Think you for poor suffering from operagoers, patron saint python principles is awesome. Uh, and says that challenges are great support. Gauging [00:14:00] learning pathways, progress. That is also, um. And and so. If anybody has questions about school for questions, comments, that's the. A.M.. Isn't the same tour working for the auto company? That is the same exact or haven't seen him in quite some time.

[00:14:20] Why is he here this year?

[00:14:22] Because, man, he's out there enjoying the sun. He's out there and join the sun and listening in on the live, live stream. Um. I shall soon see again, nice background.

[00:14:34] Thanks. Nice to see you again.

[00:14:36] Yeah, is there some chemistry going on in the background? Is that with what's happening now in New York trying to.

[00:14:44] Yeah, that's my background. That's my initial background. I did chemistry. Wow.

[00:14:51] It was pretty interesting. Do you find it odd because that's how you.

[00:14:57] Not yet, but that's the blunt. Yeah, that's the plan, eventually, the circle around to it, but not yet

[00:15:05] Nice, it's cool. Um, let me know if you ask a question about it. Mohammed Merian crosschecking bogus questions.

[00:15:12] Let me know.

[00:15:13] We can't we cannot have an office hours without a, uh, an exchange of questions. And I should. Yeah. Go for it.

[00:15:22] Sorry. Sorry. I feel like I interrupted you.

[00:15:25] I'm just killing time and filling that space until people ask questions. So yeah. Go for

[00:15:30] It. If I freeze up, please tell me I'm just get you into the stable. OK, tell the robots to stop. So my question is when you go into a rule. A lot of people like you given one off project, I need you to do this and this and that, you're not even done understanding how basically, OK, very green. I am very green. I don't understand how the business works. I don't understand. Exactly how [00:16:00] everything runs from even to do the analysis, how do you approach this? Do you go sit with the customer Tim fluxing? He went to see to the customer of a team to understand the whole business. Do you just wing it? How do you do it?

[00:16:14] Questions asked a lot of questions with a lot of people, right, if you put on a project like for example, like our price. They had me work on a project where I had to essentially create it was a regression problem. And I had to provide a prediction for what was called a suggested multiply rate and, um, me attempting to do that on my own. Who would have been very, very bad, because I had zero context about the business, had no idea how these people were making their decisions. Right. So I was essentially trying to create a system that would model a human decision makers decisions, but without actually sitting with them and talking to them, understanding what they do, I wouldn't be able to model that system. Right. So for me, they flew me out to Atlanta and I was in Atlanta for like a week just sitting around with people who are doing this this work and just picking their brains like, hey, just let me just watch you how you work and then just just give me color commentary as much as possible. Like, what are you thinking when you are looking at this particular request? What stands out to you and just try to understand what it is that they are. Looking at as a human so I can encode that in the form of features for my system, my machine learning system for a lot of a lot of just interaction in terms of questions and trying to understand the process. Right.

[00:17:36] And it might be different

[00:17:37] For another use case. Mark, do you remember this example for the customer? Um, it says that she's referring.

[00:17:48] Yeah. I mean, I think I think this this is kind of how I think about things for when because this so quickly understanding the business is I think what separates a data scientist from [00:18:00] a great data scientist is just like your ability to create amazing assumptions and keywords, assumptions there. And so when I first start with a project, I basically have a list of business assumptions that I have and need to validate each one or prioritize which ones I want to validate. Right. And so I think Harpreet Sahota example is really great where you actually got to fly out and see exactly what was doing. So you can quickly see if assumptions are correct or right. Sometimes you don't have that option. And so I would say like, all right, this is a type of analysis. Is this hub framing it? These are the assumptions that need to happen for this to be true. And so then I start going through, I prioritize like what's the most pressing assumptions from there? And then I start reaching out to people.

[00:18:46] Some of it's like a small company.

[00:18:48] I like the person in charge of it, or I go try to find documentation within the company to validate that. If not, then I start trying to schedule more, more and more meetings and really try to balance that out between kind of like the project timeline itself and and try and get closer and closer. And what's useful about how those assumptions is that now when you share your results, you can lay out what those assumptions are. So there is something wrong and it's most likely going to be something wrong every single time because you're not going to get everything right, because you laid out your assumptions when you shared more broadly the one person you miss and be like, hey, actually, that's sometimes a little wrong. We just change this last month. Right. And you can update over and over again. And so I think for me, our prerelease. Correct, because it's kind of more so from a startup perspective. But I don't really try to focus on gaining understanding the business one hundred percent, because that's just that's just a challenging thing to do. My job is to do the Data not be the business professional. I try to create a framework to be wrong quicker, if that makes sense. So I always give high quality work and I feel confident. So I give but I have a framework where if it is wrong in some certain area, we [00:20:00] can always constantly improve it because I listed out those assumptions for the business.

[00:20:05] Yeah, I like that approach a lot, and I mean it it comes down to just the problem that you're working on, the problem solving and just understanding how solving this problem is going to impact the business. And why is this even a problem for the business that

[00:20:20] Is necessary

[00:20:22] To be solved? Right. Like what? What pain points is this problem that I'm working on causing the business? Right. So in my situation, it was we had a lot of really high level executives who are spending a lot of time approving these requests. Right. And we need to minimize the time they spend approving these requests so they can focus

[00:20:43] On other more important

[00:20:44] Stuff. Right. So that was kind of the pain point. And, you know, it was like, OK, if we can have something where we can just have a suggestion, pop up or says, you know what, based on similar things that looked like this, this is what we think the suggestion or the prediction

[00:21:00] Should be, it

[00:21:00] Just helps streamline the process. So it's understanding the problem that you're working on in the context of like, you know, how it impacts the business I think would be. Or rather interesting, I see some questions coming in here, Ostern says, actually, you could just read that comment.

[00:21:19] Good one. Yeah, I liked what you were saying there. Marcus sort of reminds me of and this is been stuck in my head because I had this conversation with we did an industry meet up with the a couple of folks, Raposo with the CEO of Radio and the CEO of Deep Note, sort of on collaboration email. And there's just a lot of focus on domain expertize and sort of how data science doesn't like exist within the realm of like leaning on and relying on domain expertize. And it's sort of what you were saying is almost like a mental model version of the way you might want to validate like a domain specific model, not just on a validation set of data that's sort of like set in stone, but actually by putting it in the hands of those domain experts, letting [00:22:00] them formulate the challenge to the model, the adversarial challenge to the model, and then sort of give you feedback on like, oh, this is performing like we wanted it to, or it's not performing on this sort of case that you, as the data scientist in a case, this analogy is like to the business perspective, might not be aware of. So you create those quick assumptions. You create a quick version of a model, you validate it with domain expertize in the real world, and then you can just sort of iterate more quickly. So in this case, it's on the model, but in your case, it's like your mental model iterates more quickly. So you can actually, like, incorporate that feedback without spending too much time, you know, getting into the into the weeds you don't belong in. I guess what would be how I would word that

[00:22:38] I like that that

[00:22:40] Kind of echoes

[00:22:41] Marcosi like to be wrong as fast as possible by just creating a solution. Does it work or not? Absolutely. Can we do what can we do to improve that? Awesome.

[00:22:52] Real quick, a key thing I want to say is I think something that a lot of early Data professionals, I'm still early, but like early, early, like the first mistake I made was I didn't communicate this fact that we're working through uncertainty many times. And so when I presented results, people thought that that was the result, especially when I work with sales individuals, because, like, this one gets them out to the customer to get the client really quick. And so being able to communicate early on, like, hey, here's the uncertainty around this and like how we've quantified that in a way that that gives you more leeway to have those that change management. So when you do update your assumptions, being able to communicate like these obsessions where change is, how it impacted, that's where it was before. And I think that communication piece is super important because I totally missed that when I first my first term, my first Data size job, and I definitely got some bruises through that. A big part of that, too, is the communication aspect.

[00:23:50] Some good questions, I mean, sorry, good comments coming in from the chat here, PDF says, understand the business problem, the outcome and the success criteria, and you need to spend time [00:24:00] with the customer or end user. And he's talking about this thing that I'm going to butcher

[00:24:06] The pronunciation of the kanji

[00:24:08] Can busto principal, can she can busto principle go see for yourself and understand. I like that. I will copy and paste that here and she can busto I've seen that right. Or is saying that assumption validation is key, assumptions can be can be changed. The key is to properly evaluate and support them. Peter says that an assumption is a lack of enough information to make them. It then validates them the road to nowhere is paved with the

[00:24:41] Best intentions

[00:24:42] And incorrect assumptions. There's just the whole like discussion about boxes and the LinkedIn chat. I'm awesome with. Great question. I thank you for asking. Yes, sir. Please go for it.

[00:24:53] I just want to shout give a shout out to your to the podcast. I like it a lot. The guys talked about how they approach business problems from my dad point, essentially. And it was really very, very. Not eyeopener because of the things

[00:25:15] They knew, but the guy

[00:25:17] Has been through this and also is talking about it, like I said, this is a very good thing. What I liked about his point of. I don't know what this is about, but you are a smart guy. You'll figure this out.

[00:25:35] Yeah, so that's that. That was really, really good chat.

[00:25:38] Well, that's actually

[00:25:39] Wasn't it wasn't podcast. I thought it was a presentation at the Science

[00:25:42] Show Data Science got.

[00:25:44] Yeah. So for anybody listening post, uh, you know, Post Live, it's the future of Data Science with Adam Votaw, the CEO at Data Diligence from the Science Go Virtual Conference in April of twenty twenty one. That was hands down my favorite [00:26:00] presentation of

[00:26:01] Of the day. And it is

[00:26:03] Right. I think this will definitely cue

[00:26:05] You up

[00:26:06] Quite perfectly for your question. Yeah. Thank you for reminding me of that man. I mean, I do so much stuff that I sometimes forget the stuff that I do. But yeah, that was really, really good chat. Thank you for reminding me. Um. Some questions coming in. There's questions in the chat

[00:26:21] Here and then there's questions

[00:26:23] In, um, in LinkedIn, let's let's tackle the ones on LinkedIn, if you guys don't mind. Let's tackle those first and then

[00:26:31] We will move on to

[00:26:33] The questions. And Zoomlion, hope you guys don't mind, though, uh, questions coming in from Patro, uh, on LinkedIn. He think

[00:26:40] He is introducing

[00:26:42] Machine learning as a new competency and learning pathway for Data engineers at his company.

[00:26:47] What do you suggest in resources

[00:26:50] To start with? Um. That is a good question. They got Data engineers who are trying to increase the competency from machine learning. So I'm assuming as they're engineers, they're probably already

[00:27:01] Good with coding

[00:27:02] And all that stuff. So that's not going to be an issue. They probably need to focus more on, um,

[00:27:07] Foundational concepts and things like that.

[00:27:10] Um, one of my favorite

[00:27:11] Books, and I

[00:27:12] Think it's a community favorite book, is already

[00:27:15] Undrawn or really

[00:27:16] On Iran's book. I don't know if I'm saying his name right. And that's, um, hands on machine learning with so I could learn and tensorflow. I think that's a great introduction to Machine Learning book. Another book that, you know, like disclaimer I've only ever had this book for like a week. This is a really good book, a deep learning illustrated. And I know this has deep learning, but the first three hundred pages of this book is dedicated to just the intuition of machine learning, and it's provided with so many visuals and visual visual examples of very little math makes it really, really intuitive to understand. Just to give you a sense, like the part one is, uh, six chapters [00:28:00] that covers everything from,

[00:28:01] Um, you know, an overview of

[00:28:03] Machine learning, the essential statistics that you need, uh, how to measure performance based rules, curves and services, information theory. Then it goes into classification regression, so and so forth. Um, really, really good books. I highly recommend this one. Very minimal coding examples. Um, so it's just purely just for intuition. So I recommend this one that was helpful. A question also coming in from. Around 80, um. In LinkedIn, so around Dotty's question is. What can what can be Data model or machine learning algorithm used to predict Data when you have Data

[00:28:45] For last six months,

[00:28:47] It's a big, big question. I would need you to clarify that a lot further for me. Um, around 80. So, I mean, that's I cannot answer that question, as it is stated, without making many, many assumptions, many of which will likely be incorrect.

[00:29:03] So if you provide

[00:29:04] More information, I will, uh. Well, let's turn to the questions into the chat, Mark, you have a question. Yes, you do, right.

[00:29:16] I did from our from the office hours on Friday, so it's and in all transparency of not knowing things, even though I potentially should, I have like research design and I struggle with the sample size calculation we're talking about. It's just, again, being a visual learner, completely messing up. I just overthink it. And the main challenge I have is it was to my understanding that we create a sample size. You need to have a known treatment effect, like what's the factor you're trying to have? And that's typically how you're able to get the sample size. So you'll typically look at other research to see what's expected. But many [00:30:00] times when I'm working on Data problems, just to complete new use case, and so the treatment aspect is completely unknown. And so how do you determine the sample size when you have these levels of No.

[00:30:12] I think for a sample size doesn't necessarily take into consideration treatment, in fact, it takes into consideration power. But in order for you to estimate power, so when you're doing a sample size calculation, um, you kind of can proclaim that this is the power that I would wish to achieve for this particular statistical test. And you can just proclaim that up front. Right. So, um, so that that you wouldn't need to have an estimate in effect, size to calculate sample size, sample size is typically going to be just you need or it depends on which distribution you are sampling from. Each distribution will have its own calculation, but you'll need, you know, standard deviation, uh, level of significance, power.

[00:30:55] And, um, something else depends on.

[00:31:00] The distribution, again, I'm blinking, um. But to to. If you have power analysis, that's when you would need the treatment effects, as if you're trying to find the right. Now, I know that question,

[00:31:17] I think that makes sense, I think that's where I'm being tripped up. Is that like so I guess I'm doing a sample size. Just the assumption is this is the power we want. This is sample size. And so I'm trying to do too much right from the sample size and the power. And so that's where I'm getting tripped up at. And so I guess, like when you're making a sample size, Netezza is OK for you to make up a power by is like a general guide like this. The power center moved forward with.

[00:31:42] Yeah, yeah. OK, yeah. That this is the power that I wish to back.

[00:31:48] That's the key part of that I missed in all my classes that you just proclaim the power. I just, I've been trying to solve the for the power and I'm like how I don't have the treatment effect. So I guess the next question then is, all right, so you're [00:32:00] trying to determine power, but you're working on completely novel thing, which is typically happens in industry. Right. How do you determine power if you don't know if there's no other reference point? Is it just through experimentation?

[00:32:14] So I kind of think back to like what the definition of power is, right? So when we talk about, um, power is essentially one minus Data, we're bidets the probability

[00:32:26] Of committing a

[00:32:28] Type two error rate. So if we want to minimize the probability of a Type two error, that means that we want that beta to be not small as possible. So you kind of think of it from that point of view, like, OK, if I if I want my power to be, um, I want a highly powered test, that means, you know, I want to minimize those type two. I'm talking in circles right now. I want to pause there. Let me know if that's if that's helpful and then we can move forward from there.

[00:32:55] This is extremely helpful. So thank you. Also ask others around here, like, if they want more clarification, as long as I know we're going to charge a little bit more people.

[00:33:05] Yeah. So what I'm going to

[00:33:06] Put up another resource

[00:33:08] For you here as well. Um, I got a few here, but the Penn State is the university that I wish I went to, but I didn't get accepted. But I still refer to all of their stuff quite often. They've got some awesome stuff here. I'll send you a link to this. So they've got, um, how to calculate sample size and they'll give you the different formulae formula for I think in this case they're just looking at normal distribution. But then also, um, power analysis. Um. This might be able to help you a little bit.

[00:33:41] Amazing, I've taken multiple classes on this in grad school and this didn't stick well. So, yeah, this is great. Thank you.

[00:33:51] Yeah, no problem. Yeah, let me know. I mean, I'm happy to, uh. They Rodney's in the chat, so maybe Rodney might have a little bit of an [00:34:00] insight into that as well, because he's quite well versed in statistics, right. I don't know if you heard any of the previous question or not.

[00:34:06] Imposible has done this in grad school and this didn't stick well, so.

[00:34:12] I don't know what that was. That was an echo óglaigh.

[00:34:17] Hi, sorry, I was following you for LinkedIn and then this came up, so I thought I'd better look into the question. Yeah, I mean, what you do. Is. You do have to identify and effect size, but this comes a bit from the problem setting that you have. Can you hear me? Yeah, I can't perfectly, yeah, so it comes a bit from the particular statistical problem you have. So how you

[00:34:49] Doing a test

[00:34:50] Or are you are you doing that

[00:34:54] Or what?

[00:34:55] What is the particular method you're using? And then what you basically do is, I mean, typically you set

[00:35:04] Power at

[00:35:05] The wall if you want power to be outputted, you want to know the power, then you would you would put in a particular sort of effect size and sample size. And if you want to know if you specify the power, then it will tell you the

[00:35:21] The particular

[00:35:22] Sample size you need when you're doing it that way. So there are a number of actually a number of packages. So so what I use and I'll have to look this up because I have a right to hand almost all

[00:35:37] There are a number of packages

[00:35:38] That you can use for this that will calculate it directly. So so I'm mostly using off the power calculations. And let me just tell you the package that might help you. He had got. So I'm using two packages, [00:36:00] one called PWI and one called power mediation. So if you look close up, power, mediation is all lower case except for the M in mediation. So if you look close up and play around with them, that will give you a bit of an idea on how to do this. So so there's different there's different sort of ones, depending on which sort of effect you're looking at.

[00:36:32] I think your idea as your

[00:36:34] Comments coming into a LinkedIn as well, um, they're hopefully a combination of, uh. Everything that was said here, plus the links are helpful to you, Mark.

[00:36:44] So the got a running start now is I can play around.

[00:36:47] Ah, thank you. Just just I mean, it's pretty simple. Once you once you once you sort of get it and there's some good tutorials on it and ah I've look for

[00:36:56] Python but I haven't really

[00:36:58] Found much on how to do this in Python yet.

[00:37:02] I mean you can always code. Yeah. That's like putting yourself. I think that's the frustrating thing, because I know it's simple and I'm just overthinking it and I just need to break through that cycle and just get it down once it clicks. I think I'll

[00:37:14] Stick it simple

[00:37:15] With sort of simple, simple hypothesis testing. But let's say you want to calculate the power of a particular coefficients on a regression. Then then then it's it's not at all so simple anymore. Things get more complicated as you move up to more sort of sophisticated methods. So and that was that, in fact, the problem that I had. And so mostly by telling you to use R squared or F the the the effect of that, and then then you're calculating it on that, which is the overall regression relationship when you might be just interested in a particular coefficient. So [00:38:00] that's when I switched from, I think, using the power library to using the power mediation library, which gives you a bit more functionality.

[00:38:09] Yeah, thank you for having me. Yeah, hopefully, hopefully I didn't confuse anyone work, and hopefully if I did confuse you a bit. Perfect stuff.

[00:38:18] Um, both are really great. Thank you.

[00:38:22] Let's say there's a question I missed. There's two questions I missed in the Zoome chat here. Um, so let's, uh, let's go to those real quick. So, uh, question one is, uh, did I get Jim Craig on the podcast yet?

[00:38:35] No, man, I'm still working on that.

[00:38:37] Uh, trying to trying

[00:38:38] To make that happen soon.

[00:38:40] Jim, quick, if you're listening. Come on the podcast someday, I know it, um, question coming in from Mohammed, um. How important do you guys think formal education

[00:38:52] Will be

[00:38:54] In the coming years, formal education just in general? That's a good

[00:38:59] Question.

[00:39:00] Um, actually, I did an episode on this with Djinnit Iqbal. Uh, he's the founder of No Degree and host of the No Degree podcast. And we're talking about the future of education. Um, so definitely tune in to that podcast, as it's called. The Future is No Degrees with that. And I recall on my podcast and I think you'll find that important. I mean, I don't want to entice Ariete by sharing my views here on this. I mean, like I think. I think formal education will always be important for. Some fields, right, like I would always want. Any medical professional

[00:39:42] To go through formal education like I want

[00:39:45] I don't want a dentist who took some looks and practiced on, you know, like a skull and or douchy with my teeth, like I don't want that to happen. Right. Uh, so definitely those people should go through formal education, uh, software [00:40:00] engineers, data scientists. Is that important? I mean, sorry, I don't think formal education is that important for those types of fields, mostly because a lot of the stuff that you can learn. Is freely available online. Right, and we've got communities like this where you can bounce ideas and develop and hone your understanding like this, like I doubt dentists get together and talk about it. I don't know. Maybe they do no good. I only know one dentist. Um. But I mean, that's just, you know, just saying that I'm going to pause there for a day, some inflammatory comments and get people riled up. But what do you think? What you think, Mark?

[00:40:41] I have kind of a follow up question. Maybe that that is I've been thinking about a little bit. I've kind of heard this rumbling

[00:40:47] About people who so

[00:40:49] If you think about advanced degrees, are going to get a Ph.D. in machine learning or Data science in these fields. Kind of the sense I'm getting is that like to especially for the more reputable, like in person, like traditional education, you kind of already need to be a data scientist or a machine learning engineer. Like, you can't just like be an aspiring one and get into those programs. So it almost to me feels like there's an antecedent to this or it's like you kind of have to figure out some sort of more informal education as a starting point to get to the point where you could pursue like the formal graduate education. I don't know if anyone else sees that trend in the field or not, but that's just something I've sort of started to pick up on from folks who are kind of approaching it that way.

[00:41:26] That's interesting, um, Mark, what do you think?

[00:41:31] I love to come, I think it really depends on the type of Data science you want to be. So I think a great example is just my my my opinion on this. But like, if you're like a data scientist who's like heavy you like experimental design, I totally think a graduate degree is extremely helpful for that. I can imagine learning experimental design without doing research in academia. That's where it kind of clicked for me. Maybe it's just I just kind of went on my own and maybe some other people can. [00:42:00] But I think going through that that process in academia at least once or at least seeing others do it was extremely helpful. But on broader things like like analytics in like high level, easy stuff, many times lots of Data science world, like especially in the company's early and mature cycle. You don't need advanced stuff, like you can do a lot that you can do self-learning. And then if you give them something like that and they grow like you can learn all the way as well. So like a lot of the coding, the algorithms, you can learn that on your own. I think when they start to like apply more advanced methods, that's when undergraduate degrees need it. But you have a team of data scientists, so you don't need to be that person. You just work with someone who is like that.

[00:42:45] Yeah, I kind of like to set the threshold, like up to I need high level graduate education or not is OK. Does the work that I'm doing involve people's well-being, like, you know, my medical well-being or safety or their freedom or something like that? Um, then probably should make sure you guys are well educated man that I'd love to hear what Rodney thinks about this. Um, Rodney is the owner of a PhD.

[00:43:16] It's a difficult question. I think the education landscape is changing, so, so and I think there is is huge demand there for for education, which is why you're seeing groups like this springing up. It's an example of people sort of searching for. Searching for knowledge. That traditional education is not necessarily offering them and and so so we have to recognize that there are big changes going on. Some of this predates [00:44:00] covid-19 pandemic, and the pandemic has probably reinforced in nontraditional forms of education, whereas universities have not necessarily adapted particularly well to it. I mean, that's typically what they're doing. And teaching is they're using zone, but they're just sort of doing the same old thing that they've always done in their style of teaching. So there's a question as to what extent higher education is going to change going forward

[00:44:42] And whether

[00:44:44] People in higher education are

[00:44:46] Going to look at things like

[00:44:48] This and and begin to think about incorporating some new approaches into their teaching. So I'm not sure we're going to see a blend coming out or we're going to see people switching more and more into a whole range of different ways of learning that they don't have an awful lot to do with higher education. So an example here is a few years ago when we started putting higher education, people thought, OK, there's not a lot of future in and they're not going to work. And then what happened was eventually the courses people were taking on things like Coursera and ADEX and these sorts of places started being accredited by the universities. And they then introduced a couple of years after that, they started introducing these micro master's programs where you can do sort of the first stage of a master's degree through an online platform. And then you would switch to sort of the

[00:45:55] Residential platform,

[00:45:57] Residential approach a few [00:46:00] years later, a bit later. So so that's an example of universities blending the two approaches. And. And I think I think there is potentially a future in that, but it's going to depend what happens with the pandemic and whether they just go back to business as usual or whether the lessons are drawn from it. And they begin to retain quite a bit more of online learning in what they're doing. So. So that's that's probably my view. Thank you, Rodney Margueritte.

[00:46:38] I think one thing, though, I want to highlight is something that said that you really can't get through because the relationship building that you get when going to universities and you really can't understate that the power of being in a cohort and going through something intense, a school, you start to build really strong relationships with people. And like I'm still friends, a lot of people from from my undergrad university. And now that we're growing in our careers now we're thinking about where business moves make together. From grad school, I had a couple of friends and we became co-founders for a business we try to pursue, didn't work out. But again, those are the relationships that that forge. And I think this is more so the Akir side of it. But there's a level of nepotism as well where essentially I'm just trying to call out kind of like the imbalanced ness of it is really clicked. I mean, my first job at a startup is one of the startups that, like, hired from top tier universities. And I remember being at the lunch table and being around these other new new grads and they were essentially like, oh, do you know so-and-so? Right. And they all just met today. They all had connections from people at these top tier universities where they went to school together. And I'm kind of for someone we're meeting my parents with the college. Right. So this is all new to me. And I was like, oh, [00:48:00] my God, this is how they network like it is know each other from being in these universities. Right. And so just being aware, like there's some non educational things that are very tied to the price of going to college. Is it fair? I don't think it is. Is it a reality? One hundred percent. And being aware of those connections can can potentially

[00:48:23] Help or put you at a

[00:48:24] Disadvantage.

[00:48:27] How can I respond to that? Absolutely, yeah, so, yeah, this is this is often raised as a

[00:48:34] Point,

[00:48:35] But there are a couple of sides to this. One is a group project where you can begin to form an online group project work. You can you can begin to form those sorts of networks when you're working together on a project. And and that does tend to build more lasting relationships. I've run those projects in our own forces myself, and I've noticed that. Another thing is when you look at the networking that you see through the universities, a lot of that. Does come about because you studied with someone, but also a lot of that comes about through the seminar system, through research seminars, and a lot of those have recently moved online. So you and they're open, right? What they've done is they've opened the up so almost anybody can attend

[00:49:33] The

[00:49:33] Research

[00:49:34] Seminars at top

[00:49:35] Universities, at least to observe, and you can generally ask questions for a chat and that sort of thing. So so we now have a situation where those networks are almost completely broken, wide open. And I attended a conference online conference last year, and we had a social thing at the [00:50:00] end of prison where we're just all sort of chatting with each other. And exactly what you're describing, what's going on with saying, how do you know such and such and such and such? Now I know some of these people, so

[00:50:12] And I didn't know

[00:50:13] Anyone in the group because I joined helping someone I knew would stand up and they didn't. So they didn't know me and they were wondering, will you? And I said, well, I used to work here and sort of I know this guy and this guy and and and responded that way. And the more you attend those things, the more

[00:50:32] You you are

[00:50:34] Accepted into into the networks. Right, so so that's that's your way in, you attend. You attend seminars and you attend conferences and, you know, to to sort of get noticed at a conference, you have to present a paper. So you're going to have to put some extra effort into

[00:50:57] Sort of potentially

[00:50:58] Present a paper. And there's potential conflicts then with your

[00:51:01] Employer who may not like you

[00:51:04] Presenting a paper. So you have to worry about that. But but that's that's basically a way into those networks, right?

[00:51:14] As when I put in a check, consider my opinion change. I appreciate your perspective. Arsalan, go for it. Yeah, I also would just add to that that I think if you actually look at the mechanisms, by the way this happens, I think I kind of take my own experiences as a signal to like I'm the head of community at comment. You see what's happening in industry, right? Like these high growth industries that are demanding particular kinds of talent and a particular growth of particular labor sector or labor set of sort of labor sort of skills. You know, industries are going to jump in and start

[00:51:47] Creating contributor

[00:51:48] Communities and ways for people to network and to organize because they they need to be a part of training that labor workforce or like providing those opportunities outside of a university system. So you can also look, [00:52:00] there's just like different ways that this gets facilitated outside of universities, I think now. And it's this just feels so much more distributed than when even when I went to college back in like twenty eight or nine or whatever, it was just all this stuff is so much more distributed and like I've mentioned this before on these sessions, but like my job as head of community in the tech world did not exist 10 years ago. Like there were maybe things that people did that were sort of pointed in this direction. But now you see all these companies now, community, community, community. And that's that I think is underneath. It's like, yeah, of course you're branding, but underneath it's like this realization that that's people are hungry for these kinds of networking opportunities,

[00:52:38] Ways to put their name out

[00:52:40] There. And that's partly why I'm doing some of the things I'm doing a comment in

[00:52:43] Terms of our contributor program

[00:52:44] And education and connective stuff like this. I mean, I've already spoken to numerous of you folks offline in different conversations about what we can work on together. And so I think if you look at industry as well, that can be a signal as to where this is sort of headed or where it has headed.

[00:53:03] And technology just makes it so much more easier, right? It makes it so easy for people to get together from all over the world, like here we've got people from Australia and, you know, the U.S. and Canada and Kenya and all over. So it's it's these barriers are breaking

[00:53:22] Down at this

[00:53:23] Point. And also, man, like, there's just so much education out there that's. Rita to consume and I mean, one of my favorite sources, I mean, this probably going off on a tangent based on everything everybody else said, but one of my favorite sources is the great courses. And they've got the package inside of. It was on prime and they've got courses that that go from like elementary

[00:53:48] Math to eye level calculus,

[00:53:50] And they are courses that touch on history and philosophy and all that stuff. It's it's you know, that if you're curious and if you're interested, you will find [00:54:00] places to learn and grow. I think that's something that cannot be taught. Um. Great discussion, a great question, who asked that thing might have been Mohammed, a lot of questions or comments, rather, coming into through LinkedIn. I won't read them all just because there's so many of them. But if you guys are interested, go ahead and check out LinkedIn.

[00:54:23] Right for this. A lot of great stuff there.

[00:54:25] Um, let's continue on. There is a question here. Um, I think the next question goes to Maryanne.

[00:54:39] And sort of

[00:54:42] Tapping. Yeah, this came up actually into. Working with Austin basically to start

[00:54:52] On the computer, but the

[00:54:54] Permit and there was the question of the things developed above Tensorflow and Terra's, but in diplomatique I

[00:55:05] Know mostly

[00:55:07] Of I thought if I want to switch to catastrophising, good stuff to start from.

[00:55:15] Yeah, I like pipework

[00:55:17] As

[00:55:17] Well, but it's just it's very, I guess, intuitive to

[00:55:22] It's more subject kind of thinking. Yeah. Yeah, I like it though.

[00:55:28] Yeah. So I've been learning Carras through. I mean so again, just like in John's book, deeply illustrated. So all of his examples here are done in tensorflow and karez and characters just like an essentially an API

[00:55:41] Around Tensorflow,

[00:55:42] Uh, that just makes it easier to use. So this is a good resource if you want to get started with Carus and Deep

[00:55:48] Learning, um,

[00:55:49] Combined. But in terms of just an actual tutorial strictly for Carus, I don't know, one off the top of my head,

[00:55:56] But I can I've got a

[00:55:58] Couple here I can link you to. One [00:56:00] of them

[00:56:00] Is from three code camp

[00:56:02] And it say carries with Tensorflow course and it's about three hours in length. Uh, so the link is right there in the chat. The check that out. I think that'll be a good, good resource. And in the text version is right here as well. I can link you to these two. I think this will be a good introduction. You know, if you only get this book at the book, it's great you learn. A lot about modern architecture with cars here,

[00:56:30] Um, so check check that out.

[00:56:32] Oh, thanks. Yeah, no problem. Um, let's continue on. There is the question here from. From a.. And his question is about Data strategy. Um, what I was doing at my former role at the opportunity that,

[00:56:54] Again, the future

[00:56:56] Is not what I'm interested in, but did I end up being a Data, mature management maturity assessment? Yes, we did. I was there a particular

[00:57:03] Framework we ended up

[00:57:04] Using? Yeah. So I just used the framework that was laid out in my DMA, mostly because I didn't want to do with the framework that was tied to a particular

[00:57:12] Consulting organization

[00:57:14] Just because I just did not want to deal with that stuff. So I went with just the Daymo framework and it goes through level zero, through level five. Barring that, the other framework I really like was the one that was laid out from. In International Institute of Advanced Analytics, that's Davenports thing, and he had the analytic maturity, um. Yes. Framework for fair assessment. So those two are the ones I leverage mostly, um, but I emphasize the Daymo and just because it was free, open source, not connected to any particular consulting or anything like that, I will say, though, my friend George Furkan has a great course on the [00:58:00] various maturity models. Um, I highly recommend that course he goes through

[00:58:05] Like 10

[00:58:05] Different maturity models and just talks about the pros and cons of each one of those. So I highly recommend that. Mark, the only experience working with these Data management maturity assessments or things like that.

[00:58:18] I don't have any particular assessments, but I put in a link Data can't really came out of the white paper that I thought was really good. The quick bias is that they're trying to convince you to purchase their services for training to make that happen. But if you if you get past that point, they have some really great arguments and great frameworks within there. In addition, there's a recent blog post. I'm going to try to find it and share it in the chat. But essentially, they came up the scenario of a manager coming into a company and making them more Data mature. So it's not necessarily framework's, but some great resources to learn more about that. And I would love to learn. I know. I know you're part of the slack, the slack community when you in those please share that. I'd be interested in myself. I, I it's kind of silly. I didn't realize like there are frameworks to measure this which is interesting. I just thought it was like a state of being where you are. But now that I think about like totally some some consulting paper product around. So it makes a lot of sense because we talk a lot about it. I'm just curious. I think I think I'm more maybe potentially more fruitful conversation is like if you can't talk about it, where do you currently think your your Data maturity is at today and where you're trying to go? And if you can't talk about that more than happy to talk about the company, I'm.

[00:59:45] Big A.. Might be, uh. Unable to chat right now, it seems like he's out on the road. He says he's really modern Data strategy at the moment and think about which one to try to work out modern day strategy. Good book. I unfortunately had to return [01:00:00] all my Data strategy and Data management books back to price on my last day. They bought me all those books. Thank you. But also thank you for coming up. That's based on my bookshelf. But if you want to talk about your particular situation or use case

[01:00:15] Marked, I mean, if you're

[01:00:16] Open to chatting about that, I'd be happy to hear.

[01:00:19] Definitely saw a mass start up, and the key thing is that being a startup, the big thing is product market fit and trying to understand what is a when you build features, does it align with the market and what they need and what what they want to pay for it. And so thinking back to like a great person all over Data strategy's been my ticket business strategy. Of course, I think Harp you're there as well. It was really great. And really the key thing that I remind you of is what's your business model and how does Data fit within that? And that's going to be very true

[01:00:56] For when you

[01:00:58] Are in a startup because there is so much ruthless prioritization happening. A startup, everything is on fire, everything needs to be built. So like what you built today and put out today. And so where where Data fits within, that informs how much kind of like resources can be put towards infrastructure. And so for for us, like we we have like a Data warehouse. We're able to do a lot of SQL. And so then now we're building Data features and products so that we can capture Data to actually start doing a lot more like advanced analytics and cool things with that, even more so than we're currently doing. And so the past six months, I've actually had one main kind of OK, and that is to increase Data access for the entire company. And so my role was interviewing a lot of people within our company. Like what your use cases, what's working, what's not working right. Suckering, seeing those pain points, consolidating, that's a main problem [01:02:00] point. And then creating proof of concepts of like, all right, this is the framework for our business that I want us to go forward with getting by. And for that, something's not getting by it because, again, prioritization and working within those constraints and then delivering on that.

[01:02:17] So my projects actually wrapping up this week of improving access. And what I look like for me was creating data marts of all of our data that's like really curated training people throughout the company on SQL. And in addition, for people who are not going to law school, I created dashboards to better understand their data and in a curated fashion and ability from as a download it directly into Google Drive. So that way they have access to data to reduce the amount of data requests. Right. And so that's shifting people in the data maturity cycle where there before we are in a state of like people constantly had to go to data science to understand their data needs to. Now, people now have resources to either pull the data themselves with SQL or pull it through dashboards and only give requests for data science like payday analytics or really advanced things. And so that's where we're Data maturity cycle is like, how can you find the main point and move the needle slightly forward? That was a long explanation.

[01:03:14] That was beautiful.

[01:03:15] I absolutely love that. And I mean and just you guys are already pretty mature. I would say, again, you're a tech company, so you kind of take that in

[01:03:24] The get go. But we're not Facebook have definitely.

[01:03:27] Definitely. I mean, yeah, they're they're on next level. But everything you're describing, like we put that in the context of, like, the Daymo framework, that would be like a level two and a half level three ish type of maturity. And we had to do the exact same thing. At my last job is I spent well over a month interviewing,

[01:03:47] I think twenty five different

[01:03:49] Stakeholders. And then taking those interviews we recorded, every one of them, transcribed them, cleaned up the transcripts and then took the transcripts. And we scored the conversation [01:04:00] based on some several different dimensions to figure out what this particular individual was prioritizing for the department and aggregated that for the entire company and started to develop some type of. As Mark was saying, some type of initiative to move the needle forward to a lot of work, and he says at the very beginning of the journey, it's a long uphill struggle,

[01:04:23] Trying to get Buy-In from the

[01:04:24] Top as quickly as you possibly can. Otherwise, man, life is going to suck.

[01:04:30] Um, actually, I was just going to give a go ahead, Mark, a follow up question, but it can definitely ahead as well as, say, talk about buying in and being a startup. I got buy in for this major project, an infrastructure piece, and the week before I started it got cut because we had another priority jump in. So that's another thing. It's just so much just timing and it's still going to happen. But like rightfully so, we've got these priorities. I completely agree with it. So the point of it being a long journey, it's it's it's long and rough to move that maturity over, but it's so worthwhile once you get the pieces that you do have a place in. Yeah, cool, I think this sort of feeds into because I was just thinking about what you were talking about earlier and I wanted to circle this back around because it's like taking the approach of put something, put a bunch of assumptions up quickly. How does that sort of approach? And I mean, it could just be different. And this is a different task entirely and a more long term thing. But do you see any any parallels there in terms of like using those interviews to formulate quick assumptions, test them, circle back and over this longer term? I just kind of curious, like tying that sort of or abstract perspective about the work into this particular example, because it's really interesting to.

[01:05:51] One hundred percent, so kind of my bias, and I talked about this earlier in previous talk about how design thinking really shapes our approach problems and [01:06:00] more importantly, like I got trained in entrepreneurship and I'm like certified in that from from business school. And I did this to my training, like, really shape how I approach all my problems. And so I approach every little problem as if I'm a startup startup of one delivering our product. And so whether it's a Data science one off project, I'm doing analysis. That's a product I'm serving to it end user, even though it's a one off analysis to ask those questions that surface, what's the real need? What's the product market fit. Right. Same thing when I'm trying to develop with the goal. My service as a startup within my company is to increase Data access and move maturity. And so one of the first steps, as if you have a startup idea, right, is that you need to understand who's your customer and where their needs of that customer and what product or service can can deliver on those needs and capture about. Right. And through that, Steve, blankies like you got to do one hundred user interviews. There's not one hundred people. You have my company. So I prioritize like, all right, here's like the twenty five or the twenty people I can talk to and prioritize that work.

[01:07:08] My manager, my advisor, my board of my startup. Right. And say like, hey, here's a what you want to talk to is the most impactful. And from there these are like my business assumptions about the market where, you know, I believe these are the valuable things. But with these hypotheses, I'm going to test those business hypotheses out and go talk to people. And I'm like, oh, actually, people don't care about X, Y, Z. They actually really care about this one component. Or when I talk to a collection of people, this one person was very vocal about this, but other people are very vocal about these other things as a collection of the market. And my market is internally my company. And so essentially from there, that's how I'm testing my assumptions and iterating quickly on what a potential solution will look like. In addition, [01:08:00] similar to like a startup is like I'm trying to build, buy in and both buy in awareness of my product. Right. So our company, we have these things called drinking demos. So every two weeks you demo things you've worked on. Right. And be very raw. And so for each iteration, I demo at my company to build, buy in and get the messaging across Data access, Data access. This is how we think about Data access. Hey, two weeks have gone by. This how iterate on this Data access.

[01:08:31] Right. And so now the whole company is aware, the market is aware of all of the all of the work and all the direction. Try going to build that buy in. And then you build the product, get feedback, and then finally, after you get this Pulecio, your your MVP right now, you go out to to the market, try to sell it, you know, and those are the decision makers. So like the head of such and such. Right. And then from there, you negotiate on what's the part of you going to sell. And so it's a different perspective. But this how I approach everything. It's like as a startup owner. That's awesome, and that's exactly that's so much of what I'm struggling with or struggling with or working through a comment because I was like the first hire on the gross side of the business and first full time hire and engineering sales. And I have to build I have to figure out a way to do all of that and get the buy in internally. And that's like new to me. So that's a really good perspective. I really, really appreciate that, because there is a lot of like do a thing experiment iterates, it goes build a VP of a community program versus a software or engineering solution. So it's just like a great perspective. And I think, like, that's something I can definitely take into my work because sometimes it feels in collection like a daunting, overwhelming thing.

[01:09:48] But to hear from you and sort of get that perspective, like there is this sort of way to approach this, that's more like more kind of thoughtful from that direction is like super, super helpful. So thanks for sharing that work. [01:10:00] And and something I'm slowly building. I'm trying create a workshop on this and actually sell this through my own state peak and start my own website and stuff like that, working on it, slowly working on it. But one of the frameworks I tried to break this down is called the try framework. And so tribe where TI is talk, you talk to your stakeholders to determine what's a need or they come to you with the need and you talk to your hours requirements, so you build out and hypothesize what those requirements are. So that's why I'm crying like a table show. I'm doing a a paint of like what exactly what exactly the output should be. And I get those requirements really down really fast and then AIs iterate. So I go back to the stakeholders and talk again and go through the requirements until we get aligned on something because I actually build it. And then finally AIs evangelize. So I try to like take this idea that we built and evangelize, get everybody to it. So that's the.

[01:11:00] Yeah, that is firemen,

[01:11:02] I like that, so I'm working on it. I'm like, I'm trying to do some Toxi and some on my podcast, I mentioned a little bit, but hopefully I'm aiming in a few months to create a whole workshop where I'll teach you how to go through that with different use cases, whether you're doing Data science or someone in Data science trying to crush your projects or you're a manager trying to figure out a framework to describe to your your your direct reports.

[01:11:28] That's awesome.

[01:11:28] And I love that. I think we should and

[01:11:30] We should do something there. Yeah, we should. I think there's so many that sounds like something that applies not just to Data science, but to sort of like nascent parts of any organization or things that are maybe have less visibility and are trying to build sort of like internal credibility, an internal narrative around that you're doing. And especially for me, like the thing I struggle with is like things that are not typically data driven. Like, a lot of the stuff I do is qualitative. It's relational. So like a way to a framework that's unique, that sort of presents it as you were going through these steps in this process, some parts [01:12:00] of it might be qualitative, but the aggregate as a sort of

[01:12:02] More

[01:12:03] Quantitative way of scoring that approach. I mean, I can see that appealing to not just Data careers, but also a

[01:12:09] Lot of other sordidness

[01:12:11] Spaces as well. So that's definitely definitely something to talk about that on it. Yes. So I guess this comes from like that training from that business partner my did and also all the mistakes I made. And so this is like the game plan I would give to my myself when I first became a data scientist and not make all the mistakes that

[01:12:30] Nice, absolute love that man. Yeah. And like, you want me to put you in touch with 80 percent the dedicated men, I think they'll be awesome to have there.

[01:12:38] Oh, yeah. Well, let me let me get that blog out and and get this get get a read it out to other people, see. And I'll sign get some iteration of feedback on it. I would love to I love the present because I think it'd be really helpful for for people who are new just to think about, like how do I just like take a request and actually deliver and get by in to really accelerate your career.

[01:12:58] That's so good, man. Mark, thank you so much for sharing that. I would be excited to see a that is of course when it comes out. I'm there, man. Yeah. And that's a good question. Coming in from Asia. It actually is still there. Go for it ushe.

[01:13:21] Ok, yeah, no, I was

[01:13:25] Actually asking if anyone has actually had the

[01:13:27] Chance to use

[01:13:28] Neural networks at work. I haven't had the chance, but that's because I'm very fixed in my ways also sometimes. Does anyone have the chance to use this at work and how to get help?

[01:13:38] Yeah, I've never had the opportunity to work and a lot of the scientists that I know that work on my actual business problems and stuff who are not at big tech companies, also don't get a chance to implement these type of methodologies in their day to day work, which is why I'm so excited about this opportunity, I call it, because now I get to get [01:14:00] good at deep learning and neural networks and stuff like that. Um, that I believe that's like the future of, you know, a AIs is that, um, I don't know, like Mark or Rodney or you guys use these methodologies at work.

[01:14:16] I know I know people who have been very crazy, smart, and they have like a very specific use case, I can try to reach out to them and try to connect. Just interested. But neural networks are so complicated to put into production and maintain that. I feel like that's for the Data maturity conversation. You need to be far in the Data maturity side to be doing stuff like that.

[01:14:45] And like most of the problems I was working on the last few years

[01:14:49] Weren't so complex

[01:14:51] That they required these advanced methodologies. It would be overkill to use those. But Rodney, would we?

[01:14:57] Well, a few years ago, I had a student and we were doing a neural network project related to finance. So so that's that's in in an academic role where we used it, which is probably more likely. And then with the stuff I'm

[01:15:15] Doing now,

[01:15:16] I had a problem, a price prediction problem I was working on a couple of years ago. They wanted us to use more sophisticated methods than than Arama was, was what I was told to do it, which is

[01:15:33] Sort of an odd

[01:15:34] Task request. So what I did there is I was evaluating a number of different methods. And so I looked at things like random forests, and then I looked at

[01:15:46] Support vector regression

[01:15:48] As an option. And I ended up going with support vector regression, but. I also began looking at using Keris to [01:16:00] implement a neural network for that, but basically it was becoming too complicated. So so I sort of stepped back from it, but I need to return to that. So so so one of the things to upgrade that particular model is, is to add a neural network component to some of what we're doing. So so I

[01:16:26] Will be doing it,

[01:16:27] But I haven't haven't had the time to actually get it working. I mean, it's difficult because, you know, it's really time consuming to do the neural

[01:16:37] Network type type stuff

[01:16:39] So it can be really time consuming.

[01:16:43] Yeah, I'm, uh.

[01:16:46] I understand my different types of networks. Well, then for a time series type of problem, you're probably looking at like a recurrent neural network or LSM type of

[01:16:55] Methodology to make that work.

[01:16:58] Yeah, that's that's that's pretty much right. I mean, there is another option, which if you're doing it in an hour, is Rob Heineman is forecasting package, actually has a neural network set up built into it.

[01:17:10] It's not deep learning

[01:17:12] Stuff, but but you

[01:17:13] Can basically do a switch

[01:17:15] On it and and and set up a simple neural network in that. So so that would be the quickest way to sort of do it. But.

[01:17:28] This isn't the only thing

[01:17:29] I have to do. So if it was the only thing I have to do, I would have done it by now. But we we've just not had the time because because other things have been a priority.

[01:17:43] Or go for it.

[01:17:45] As a result of a rather interesting use case of deep learning in your networks, for the long as I was like, what's the point of doing this? Like, why are we doing this? Much simpler things work better. And my opinion was finally change when I [01:18:00] talked to this head of analytics at Pepsi. And basically he was saying this for like the online presence. And basically what he was saying was essentially like the reason why we do this, like, super complicated approach is because when you're working at the scale and you have competitors like Coca-Cola or these other big names, just that extra one percent can seem like a huge competitive advantage. So that's the right business. Use cases like every trying. I get the first 80 percent right. Neural networks overkill. But once you start competing at the point where, like, the difference between like ninety and ninety one percent accuracy means like millions of dollars and like a competitive advantage in a market, that's when the business case starts to override the hurdles of implementing these problems. And when when that person described that, I was like, oh my God, yeah. That's why I like the Googles in the face of these big tech companies are doing it because they're competing at a stage where those percents, they've already figured out the product market fit. Right. And so now they're trying to expand their market, not trying to find a market. And so the deep learning networks give them the ability to expand their market, just such where they can outcompete their competitors. And so I think that was an interesting kind of business use case. And so, like for me, a startup, we're not we're not to extend the pipe trying to find the pie. And our bread and butter isn't like our product is a neural networks itself.

[01:19:28] Right. So I think that was

[01:19:29] Just the interesting thing I want to add that convinced me with deep learning so important,

[01:19:35] I mean, just the

[01:19:36] Part from the business use cases and stuff like that, like, oh, really go for it

[01:19:41] To finish first and then I'll come on

[01:19:43] In and say, apart from all the business cases, deep learning is fascinating in terms of what it can do. I think there's so much potential for deep learning to help augment and enhance human creativity. Like you look at these generative models,

[01:19:58] Like there's this you know, there's [01:20:00]

[01:20:00] There's like the speculation that Spotify is using deep learning to create fake artists to pump out music. And I'm convinced that one of these artists is lemon pie because nothing exists about lemon pie. But the music is so good. I love it so much. But those are interesting use cases like how can we use deep learning to help generate interesting music? Or there's another platform called Jarvis Ehi, which is like a a copywriter where you upload some text and they'll help you write a blog post. Right. So that's my. That's the area that I'm really fascinated in, where deep learning is those generative models, and then then also applications to.

[01:20:38] For financial

[01:20:39] Trading, stuff

[01:20:40] Like that, that's fascinating as well, um,

[01:20:43] But yeah, run.

[01:20:45] So I should add that in another one of the divisions where I work are what they're doing is that they've got a couple of pilot projects where they're outsourcing

[01:20:58] All of the

[01:20:59] Machine learning work. So what what is what is happening there is they're either outsourcing to government research organizations. NGOs are increasingly using machine learning in some of their work and also that in talks with a company to do to outsource another aspect of it. So there is a neural network going on. But but a lot of it is is simply being outsourced to specialists. So, yeah.

[01:21:36] And just to push on a creativity thing, I was trying to find this link. I mean, it's something similar to what Drew was working on at at Comit with that type of a text representation of what you want and it'll create something. I forgot the person who did this. There's a Google CoLab notebook. If I can find it real quick, I'll share it. But essentially what he did was that he

[01:21:58] Like he he created this

[01:22:00] This [01:22:00] generative network where you just type in like you type in something like a library in a forest in the style of Thomas Kinkade. And then the generative model created this beautiful painting. And so there's a book that I'm currently reading. Uh, it's somewhere on my desk here and listening to

[01:22:18] The Creativity Code by

[01:22:20] Marcus do so or something like that. Can I say his last name? But that book is so good and it talks about all these different use cases of deep learning,

[01:22:28] Um, or

[01:22:30] Helping augment.

[01:22:32] Human creativity and

[01:22:34] Fascinating.

[01:22:35] I have a quick question. Oh, go ahead, Austin. Yeah, I was just going to comment on what you were saying about our our project. I'm very partial of a little obsessed or that it's just kind of a fun thing.

[01:22:46] Yeah, it's clipped.

[01:22:48] Clipped draw that comment. And so I think generally the general point is what you're seeing is like these sort of large transformer models is I think what we're talking about, like the AIs and the sort of language models and then these sort of literally transforming in the sense of like text to image or text to audio or whatever it is. And so I think what you're seeing is like these implementations and what we're really excited about a comment is like the ways you can implement an instrument, those models, and then create sort of like what we're trying to do is create a public gallery of things people have submitted and then also keeping track of the like the sort of input parameters. So like what they actually end up resulting in to. So it's like it's not just a thing you share on Twitter, but it's actually like a public repository of how these generative models can be used in different four different prompts and different sort of input compared like hyper parameter comparisons and then sort of also watching the evolution of how the neural network actually draws what you're creating in sort of across steps of training on this individual sample. And so then you see this. I think there's a hugging face. There's spaces now where it's just like live demos of models sort of using some of these technologies. And they have like a VQ gane one, which is another sort of generative model that does a slightly different style of art. You're seeing different styles of art, different mediums, different all this kind of stuff. And [01:24:00] then I think the next step is to sort of create this middle layer on top of that, where you use other tools and platforms to sort of help present those to the world in different ways. So it's not just the output. It's also the way it's packaged. And the way it's presented is sort of like a gallery or a crypto art sort of thing as well. Like there's all these different avenues now

[01:24:18] And we'll see which

[01:24:19] Ones kind of went out. But I think that's the next level of this is like, OK, these generative models actually work and they're pretty badass. And now it's like they'll keep improving. But it's also like how what's the medium they get expressed through? So that's why I'm like

[01:24:31] Kind of obsessed with this project we did last

[01:24:33] Week. Just to plug that,

[01:24:35] It's super fascinating.

[01:24:36] Like, I'll pull this up on screen.

[01:24:40] So guys, check it out. Clip draw comment, Amelle. This is probably need to get some more instructions, like what do we do? Like to see like 40 dead.

[01:24:52] There's like a blog post on this. But basically the only thing you really need to enter is a prompt and the rest of it is sort of optional parameters. And we have a little glossary on a blog post that I could share as well.

[01:25:03] And, of course, with a milkshake. Nice to

[01:25:04] See what happens,

[01:25:07] But then while this is

[01:25:08] A while, this is running here. Oh.

[01:25:13] Queued up on the at the post I was looking at, so this right here, Tyler, Soire, Swades, what I can say his name. Um, so this was a generative painting just done through what's called poly.

[01:25:28] Um, and this is literally

[01:25:31] He just had a forest in a library in the south of Thomas Kincaid and.

[01:25:35] The generative model made this, which is super

[01:25:37] Fascinating, um. Yeah, it's so interesting when deep learning, so fascinating

[01:25:42] When it comes to

[01:25:44] How it can help human creativity. Mark, go for it.

[01:25:47] It is more so like a philosophical question, but do you see this replacing the art world as being a subgenre within the art world, especially when like a big piece of the art world [01:26:00] is scarcity? And so by the ability of like A.I. creating amazing piece of artwork, they're like tailored to people's like wants and desires that can happen quickly. The real scarcity goes away. Right. And so I guess I guess the overall question is like with art being so easy to create through these models, what you know, what's the value of art moving forward with this? I mean, I as someone who enjoys art as a dancer and you want to be an animator at one point. I love art. Right. Like, what is the industry when if that is the art of the business. Right. What does that mean for

[01:26:39] You to do that? That's something that markets do. What talks about in his book, The Creativity. It's great that I grab it if I can, but he talks about exactly what you're addressing and he's talking about a a painting that sold for like one hundred and fifty thousand dollars or something like that. Right. And there's these paintings are just so beautifully drawn

[01:27:00] And and

[01:27:02] And just so emotive that when people were told it was done by a guy, they felt cheated. They felt wrong for some reason. So he's also got a bunch of talks on the creativity code. One of them is like 15 minutes. And he has the audience look at two sets of artwork,

[01:27:18] One set created by a

[01:27:20] Human's once they created by A.I. and just the A.I. stuff is so much more evocative and so crazy looking and so, so interesting. Um, you know, definitely check check that out. I'll give you a link to that. And then there's also. He's talking about there's a model that was created which was trained on Rembrandt Data, right, and it was trained just on Rembrandt painting and it's generated it's a painting in the style of Rembrandt on its own. It was just stunning. Like it looked just like a Rembrandt style. A face looked kind of weird on the person, but. Yeah, I mean. I think if humans can interact [01:28:00] with I mean. If you think about it like the human created the model, you human create the algorithm, right? Human came up with the idea that is art in itself right there. Right. That creativity to say, OK, here's a here's something I want to try to do. Here's how I'm going to do it. Here's the methodology I'm going to use. And here's the final output like that entire pipeline is all human creativity. Right? So it might be generated by. An algorithm generated by a model, but that model is not inspired. It doesn't have a it's not a move to create something. It's not compelled to create something.

[01:28:37] Um, but the human behind it,

[01:28:39] You kind of get that that. You feel that in the code? I don't know if I'm making sense, man, but

[01:28:47] It makes it makes a lot of sense. I mean, I imagine like a service, like if you have like I lately been seeing, like where you pay for a subscription service to have, like, digital paintings in your and your house, it's like they're charging like thirty dollars a month. That's how I like access to digital paintings. But I imagine something similar where you can like some of my lights on, like I want pink lights. Right. You could do something along lines like artwork that looks like X, Y, Z. But this feeling because I have a dinner party with this theme, right? Yeah. I think this cool business applications of all this.

[01:29:18] I mean, think about like a music platform where you select your ten different favorite artists and now would create artificially generated music based on your ten different artists that you like and create music designed just for your taste. Right. You know, based on your content and so on.

[01:29:36] And now you're moving to Malibu. Let's start a business on this now.

[01:29:41] I definitely don't change, so I'm going to pull this up. Is my here's my random

[01:29:48] Forest with a milkshake

[01:29:50] Happening live in action. And that looks pretty like that's way better than I would be able to ever draw. That's for sure. It's a random forest with a milkshake. And you see there's trees, [01:30:00] there's forest. Here's like milkshake going on. That's super cool, man. I just see this happen live in action. Um, where I mean, there's so many different use cases with with deep learning that, you know, help augment human creativity. That just blows my mind. It is fascinating, especially with tech stuff. Right. Like I can imagine. I again, like I was just exploring this Java Sea ice after I was like, what if I start? Because one thing I was planning on doing was a model to help take my transcripts from all of my happy hours and office hours and just come up with, like, interesting articles from that. But then I found this company, Java, that does it. And, you know, it's like 100 hundred bucks a month or whatever, but. I'm going to test it out and just see what happens, because that's so fascinating, right? It just it helps me a lot because I feel like I'm the kind of person that's like a remixer.

[01:30:51] Like, I'm like it's hard for me

[01:30:53] To come up with something from scratch on a blank sheet, but

[01:30:57] I can look at several

[01:30:58] Different things and combine them in interesting ways and create something original from that. And so having an algorithm or a model where I just put in a bunch of text Data and spit something out and I'm like, OK, great, now I got something to work with. Now I can like play around with this and add my own twist about style to it. But it helps get started, right.

[01:31:19] Yeah, I, I did something very similar to this. I think those are the two. So I need to actually revisit this. I found like when it's like stream lit web apps or whatever. And so I wrote a book of poems in grad school and published a book of poems. And so what I did was what I thought would be really interesting was to was to take all like the first line or first two lines of those poems that usually have some sort of image in them or some sort of

[01:31:43] Narrative sort of structure,

[01:31:45] And then just seeing what the model spits out in the back and and see if I could like do version two of my like with the machine learning version of my of my book of poems and sort of like like the remix edition or something like that. That's kind of fun, stuff like that, where you can kind of take. Yeah. Take what's already here [01:32:00] or what's already been produced and sort of reinvasion it or just see what happens. It's just fun to even just see what happens and even just taking away the the quality assessment of it is just kind of fun to see. Something I've been really playing around my head is taking GPG to and taking all my posts contain and Scheveningen and that model and doing, I got 30 days of getting to where I just post on LinkedIn using the API. I generate posts and see what happens and is kind of like, hey, disclaimer is built by A.I. and then playing the GitHub link in there. But I feel like that would be super interesting. And also give me a break for 30 days to focus on some other things.

[01:32:42] That's cool. I like that. I love and I would do that to actually just just need a bunch of the three, a bunch of aphorisms and just have computer generated aphorisms happen like Bailly posted on my behalf on Twitter. I'll be interesting.

[01:32:58] Um, I mean, here's the

[01:32:59] Book, though, the creativity code. So this book, I'd listen to it on Audible. I liked it so

[01:33:05] Much that I you know,

[01:33:06] I bought a physical copy, as I tend to do, um, and some of the stuff that this book touches on. So one thing he talks about, there's like three different types of creativity and then he proposes what's called the

[01:33:18] Loveliest test, which is how can we tell

[01:33:21] If something is artistic? So it's very, very good. And he talks stuff about painting by numbers,

[01:33:27] Uh, music, the process of

[01:33:30] Sounding mathematics, a songwriting formula, uh, language games. I tell you a story. Um. Fascinating stuff.

[01:33:38] So, um, yeah, I'll be.

[01:33:43] So the last day, day 21 of 21 days of deep learning is all about A.I. and creativity. So I'll be talking a little bit about the stuff

[01:33:51] In this book, in that last post.

[01:33:54] Um, but, yeah, let's go ahead and start wrapping things up. Skin being a great, great, great session [01:34:00] man has already been out of forty five minutes. So guys remember, tune in. Twenty one days of deep learning that whole 21 days of content lined up again. Like I said, I'm, I'm creating content on the day of I haven't pre made content or anything like that. But that's not to say that I won't do that. I might have, you know, my critics a few days worth of stuff in advance, but definitely tag along and follow along. Looking forward to sharing that journey with that, with all of you guys. Yeah, I guess that's I see some questions trickling in on LinkedIn. One of them was checking out deep learning usage in the field of gists. I have not. But that's something interesting. If you find something interesting on that, please let us know. Please post a book titled The author. Book title is The Creativity, Code Art and Innovation in the Age of. I think if you have orrible prime or whatever it's called, orrible premium, it's included in in that can't remember, but it's really, really good. But definitely check it out. I recommend it. Yeah. So guys, take care, have a good rest of the weekend. Remember, my friends, you've got one life on this planet. Why not try to do something big or something like.