comet-ml-march31.mp3

Speaker1: [00:00:09] All right, what's up, everybody? Welcome, welcome to the comet and all of this hour's powered by the artistry of science. Thank you guys for being so patient as we work through some technical difficulties on our end. But I'm glad you guys are here and I'm so happy you guys were patient in there with us. Ayodele is out today, so I was just going to be us. Hopefully you guys don't mind just having me around how everybody doing this this Sunday afternoon, evening, morning, or whatever time of day it is for you guys. You guys are mostly AIs are in Europe. Some of you guys are other parts of the world and Ogiek in Kenya. Wow.

Speaker2: [00:00:47] Ok, we had this start saving tonight.

Speaker1: [00:00:50] Oh, you guys.

Speaker2: [00:00:52] All right. So for me, it's six p.m. again after two weeks.

Speaker1: [00:00:57] It's interesting times and stuff, man. Hey, well, happy to have you guys here. Let's go ahead and let's get started. Man Anybody got any questions on anything, any comments? Anybody want to want to tell me about what's going on with their lives?

Speaker3: [00:01:09] I have a question related to not to Data science. What was the laptop computer questions? Kind of. So I'm trying to get a laptop for to do Data science work deep learning work, training models and stuff. So is it good to have dual operating system and those two Ubuntu and Windows Ten, is that good? You have dual opportunities are always better. Just stick with windows.

Speaker1: [00:01:34] So my laptop, our work. So I mean personal life. I'm always like I'm a Mac guy, you know, that's that's my goal too, because I like working out of Linux type of operating environment, but I work. They didn't let me get him actually get me a PC instead. And that PC, it's just one operating system that builds up and that's windows. But with windows within windows, I all my work that I do is out of the Windows subsystem for Linux and that is Ubuntu twenty

Speaker3: [00:02:03] Twenty twenty point six or something like that.

Speaker1: [00:02:05] Yeah, I was twenty point four. I should incremented up a little bit, but it's not necessarily dual boot, but it's, it's just a it's not even like a virtual environment, but it's just it's biggity Linux environment running a kernel so to speak. So that's what I do. And it works perfectly fine just because I don't know. I'm not I'm not an expert in this either. But I do know that working python environments out of windows can get messy. Not everything will play nicely. But yeah, Windows subsystem is one way to go about doing that, which is pretty easy also.

Speaker3: [00:02:41] Okay, so Python looks better on Linux system you feel. Or Windows operating system.

Speaker1: [00:02:47] Yeah. Yeah. Oh I see. Yeah I assume so. And the thing is when you have Windows subsystem for Linux on your machine like you will primarily be operating out of the command line. There's not like a virtual like a graphical user interface that you will work out of. It's just a just command line. And you'll go in there, you'll open up whatever environment you need. If you're using Jupiter notebook, you open it from there, but then it'll the Jupiter notebook will open on your Windows browser. What are we doing? All the computation out of that window subsystem.

Speaker3: [00:03:17] I see, because I'm thinking about getting the the Tensorflow book, Lamda Tancer book and it has and yeah. And it has wanto in and Windows 10 operating system. So I have an option of taking bolts or I just choose one or the other. So I was going to do both. But then, you know, I had this, someone was telling me that, you know, when Windows 10 would have an up and have an update, it it messes up the other operating system or something like that. I don't know how true that is, but yeah, especially when there's no Windows update or something.

Speaker1: [00:03:49] Yeah. I mean, that could be possible. I see Cristoff has a question here. Yeah. Oh. He's just saying he's not qualified to answer this question with a smiley face. I don't know if that you are qualified or you just know but yeah. I mean I, I'd say just work straight up one one boot that's the Windows boot and then do all of your work, your designs work out of the Windows subsystem for Linnik because using the Windows subsystem for the next can use the Windows file explorer to explore your files in, in the subsystem.

Speaker3: [00:04:18] Ok, so it's just it's better to do Windows Ten and have Linux inside of it. Yeah, it's

Speaker1: [00:04:24] Completely seamless to do that. But I mean you might want to look at maybe I mean it's completely up to you. Maybe if you don't want to spend thousands and thousands of dollars on a laptop, you can always do something like Google CoLab and just rent computation time from them, which is pretty cheap and you know, depending on circumstances, might be more financially viable to do that, especially if you're not doing crazy computations all the time. So.

Speaker3: [00:04:46] Right. Yeah, I'm just thinking if I'm going to spend money on a laptop because I do need to get one pretty soon. So I should I should I just to spend money and get a good one and then just do my work, go there for a few years, you know. So I mean that's what I'm kind

Speaker1: [00:05:00] Of I always say get a good laptop. I mean, I'm a I'm a Mac fan boy. Now you to be all about the windows. But a few years ago I switched over to Mac and I'm like. Much better. I love it. I love working with Max. So, OK, yeah, but me with the window subsystem, it's just that's fine. You just got to be comfortable working out the command line, OK? All right. All right. So, yeah. What's up? Everybody else. Anybody got any questions? We can go for it. I see somebody new here. Masab. Misbah. Sorry. Masab. Misbah. Hi.

Speaker4: [00:05:31] Hi. How are you doing. Yes I am.

Speaker1: [00:05:33] How's it going. How could I how can I help you make any questions or anything.

Speaker4: [00:05:36] Yes, I want to answer I want to bring you in on you. I don't know there to continue my journey as a data scientist. I learned so much from Mutu from different retired and from YouTube, from you, me and I learned Python and different things. But I didn't complete the complete course of Data there. So I don't have a complete roadmap and complete a computer map there. So I don't

Speaker1: [00:06:08] Know. There's not a complete roadmap to become a data scientist. There's not like a step by step guide that says do this and do this and do this. Then all of a sudden you're a data scientist. You need to start doing something. So it's one thing just to go through and take all the courses and get exposed to all that stuff. That's great. You need to have the foundations, but ultimately you need to apply what you have learned. Right, because the doing encompasses a lot of things. The doing is the learning. That's the learning in action. So you could sit back and watch all the other courses and read whatever on YouTube. But ultimately, you need to fire up a Jupiter notebook. You need to get some data. You need to define a problem statement, and you need to go from raw data to decision and practice data science and action. And you do that enough. You need to start doing that over and over and over again. So do a bunch of small projects for yourself to take what you have learned and apply it, put it in motion and start doing stuff right. And like, there's not like a best problem. That's not a best project. What are you interested in? Define for yourself what it is that you are interested in doing.

Speaker1: [00:07:08] Right. And then around that, create a roadmap to say, OK, great, I'm interested in finding out, I don't know number of traffic accidents that happen in my city. Great. I don't know. I want to explore that phenomenon. OK, great. Where do I get the Data once I get the data, how am I going to clean it? How am I going to then do some exploration? Am I going to do any type of statistical analysis? Am I trying to do a predictive analysis? And I try to make an inference like what is it that I'm trying to do and define it for yourself each step of the way? And that's how you practice becoming a data scientist. So that's the roadmap is great. I all the stuff that's great, you need to apply it and do something with it because nobody's going to hire like I mean, think about it. Imagine if a textbook walked into a job application. The textbook has all the knowledge. Right. But nobody hires a textbook. Right, because a textbook doesn't do anything right. So like, you know, you just just imagine that. Just imagine like a textbook with legs, walks into the interview, answers every question. Absolutely great. Because it has all the information in it, but still can't do anything right.

Speaker4: [00:08:10] Ok, yes. Thank you so much. I want to read some, but I bought one course from an offshore platform that oil from Canadian Register teach the other language and I will. I already worked on Python language, some small projects. I work on small projects and claim the right to throw weight on and therefore I am a little bit confused about which language I NLP. I think the most

Speaker1: [00:08:39] Important question that you should be asking, it doesn't matter. Like it literally doesn't matter because you can learn one language, can you learn any language? So the problem like the python or that does not make a Data scientist. Right. That's not what defines your ultimate career path, because ten years from now the language may change. Held three, four years from now, language might change. So that's not the issue. Pick whatever you already know. If you've done Python and you're comfortable with Python, great. Take that, run with it and start doing stuff with that. It doesn't matter. Like the the point about which programing language I should learn is like the least important question that you should be asking right now. The most important question you should be asking right now is what is a interesting problem that I want to attempt to take my skill set and apply it towards and do something with. Right. With what I've know so far.

Speaker4: [00:09:27] I'm basically from to here. I'm making software for connecting doctors, the patients laboratories, information on the platform. Within my study, I start small software development company. I work on an and and with different developers. But, you know, I my I want to work on medical Data their side because we collected a medical letter from different doctors and they're in future. I want to take a little bit and work on it.

Speaker1: [00:09:59] If you got softer skills, you already halfway there, right? More than halfway there. So if you want to work with medical Data like I mean, I don't know how. You can get your hands on that. There's definitely open Data portals around there where you can just look for medical Data or whatever and start doing stuff. I'll leave that up to you. I don't know how you're going to source medical Data because that is kind of confidential. But at least you have an idea of what it is that you want to do. So you can look at some open open data portals. I think the World Health Organization might have an open Data portal where you can get data from and just start doing things with it. But it's just all a matter of just applying what you've learned so far. Hopefully that makes sense.

Speaker4: [00:10:40] Yeah, OK. Yeah, I understand.

Speaker1: [00:10:42] Yeah. And looks like there's a lot of medical data on Kaggle, so that's a great place to start. So yeah. I mean think about, just think about OK, what you need to sit down and think about for yourself. What problem am I interested in exploring. It all starts from there. Any good thing that happens starts from a fundamental question, an interesting question, good things and interesting things. Don't start with what tools do I have available to me that's part of it. But part of it is like, what can I do with everything I've got available to me to solve an interesting problem and then try to try to solve that problem or to inform your understanding of it? Because if you're interested in it, then that is going to carry you along the entire process and it'll show in your quality of work and you'll show in your portfolio so that when you go and start applying for jobs, you could be like, hey, look at this really interesting question that I can guarantee nobody else has thought about because I'm uniquely interested in it. Right. I've developed a certain set of skills of a quite some specific knowledge through completing this project. And I can guarantee you that nobody else has done this set me right. That's the kind of thing you want to start doing.

Speaker4: [00:11:44] Ok, if I want to start my own business consultancy. Yeah. If I. How many experience are required for it? If I start my own datacenter consultancy business,

Speaker1: [00:12:00] That would have the answer to that question. Um, you can start a consulting business right now, but are people going, how are you, can you deliver on work results. I do have a proven track record. If somebody asks you and says and you go up, somebody say, hey, I'm going to be a death sentence consultant, let me do work for you. The first thing they're typically going to ask is, OK, great. You have any past projects I can look at to get a sense of the type of work that you can do and you start delivering results. The more results these are delivering, the more you are likely to get hired. Right. So I don't know if I answer your question, but start delivering results, start doing things. The more things you do, the more tangible artifacts you have that demonstrate that you're able to do stuff, the easier it will be to to start whatever business you want to do.

Speaker4: [00:12:50] I don't know much about that. So I don't know how many businesses I can go from if I were an expert in that sense.

Speaker1: [00:12:59] I mean, Data sounds kind of universal. The process of taking raw data to decisions is agnostic of any industry. Just start doing stuff I think you need to start doing project. Do you have a portfolio? Like if I go to your GitHub and I look what's on there, is there going to be any indication to me that you are able to do anything with Data? Right. That's a question I'm asking right now. Is the yes or no answer. So if I was to go to your GitHub right now and look at your portfolios, is there anything on that portfolio that would indicate to me that I should have some level of trust in you to execute on a Data project that I need help with? The answer to that question is no. Then you need to start developing that portfolio, because otherwise no one is going to hire you not for a job or not for a consultancy or freelancing or anything like that. So the first thing is to start doing projects, a bunch of projects. You need to start doing that. So there's a roadmap to start doing some projects. OK, sure. And how to do a project like that's not easy. That's difficult. That I mean, yeah, that that's that's where the learning is Harp action. Right. So yeah, it is an iterative process is an iterative process.

Speaker1: [00:14:06] You start and you increment a little by little start with the Data start at the start with the problem statement. Right with that problem statement, find some Data that could help you make progress against that problem statement set out in analysis plan that says these are steps I'm going to follow because if you if you don't have a methodology in place and not really doing science, you're just doing random stuff. Right, to find a methodology. I define what you plan on looking at during the exploration phase. Define from that exploration phase what statistical test you might want to look at to examine any relationships that you see during that exploration phase. And then based on your problem statement, whether your problem statement indicates that you're looking to do some type of inference from a sample towards a population, or if you're trying to do some prediction, then proceed with building whatever type of model that you need to build, making sure that if you build out a model that you have at least some baseline of comparison that you can say, right, I'm going to build a model. I know that I need to have a line in the sand that indicates whether or not this model is good. So let me start with the simplest. Model, I'll give me a decent result and then get more and more complex from there.

Speaker1: [00:15:12] Then once you've built the model, then how are you going to serve it? How are you going to how are you going to provide value from your predictions if you're looking to do a prediction? Right. Some cases you might need to have a Web API or somebody comes and drops in information and you get a prediction out, or maybe it's good enough just to push it off to a database that people can query whenever they need the result, or it's putting it in an Excel format, rather CSFI format and shipping it to someone. These are all details that you need to figure out for yourself for whatever project are working on. Right. So just pretend like you're in a business situation and operate from that context and you do that several times. But then the two or three projects that are your best work, that goes on to GitHub, but you should be doing hundreds of projects or you should be doing hundreds of experiments, hundreds and hundreds of projects in order to get good small little projects and nothing big, nothing crazy, just enough free, more comfortable with the methodology. But then make sure that whatever is on your portfolio is the it's fully fleshed out, full, complete representation of your technical ability.

Speaker4: [00:16:14] Ok. OK. Thank you for that.

Speaker5: [00:16:16] Yeah. Go for it.

Speaker6: [00:16:18] I was also coming from a technical background by technical programing. The biggest mistake I did earlier on was ignoring statistics. I cannot insist and emphasize enough. You need to get the statistics down. I came in with the programing, like you said, you came in from programing. So you come in knowing all the code, trust me to be easy to pick up the libraries, but then you get to machine learning, then you have the math behind it. You don't know why you're getting certain things. So I the best course I started with was these are course on Udemy. It's called Complete Bootcamp, Complete Data Science Bootcamp. It's really long, but it has it begins from the beginning, from statistics to everything. You really learn the things you didn't learn. I would insist on the statistics, but I, I have a fundamental yeah.

Speaker1: [00:17:03] I definitely don't get your question about statistics is fundamental to machine learning. Machine learning is essentially statistics. When I was in graduate school, it was called statistical learning. I took a few classes in statistical learning which was rebranded as machine learning. So yeah. So you get a leg up coming from a programing background like me. On the other hand, I came from a statistics background. I had to learn all the coding and all the stuff like that. So opposite ends of the spectrum where actually I think we're talking about this last week you might have been. So take a look at last week's office hours that are up on the YouTube channel. We're having a similar discussion around this to add to the science. Well, there's multiple paths, but these two common bad. But go along with your question, Natia.

Speaker6: [00:17:43] So this is going to be a bit of a long question. Please bear with me. I so recently I volunteered with a company with let's call it an organization. It's called an international organization. They help at risk youth in Kenya. There's been a problem of people being recruited to and we call it extremist groups, but they target the lower class people who live below the poverty level because they offer a form of payment and they mostly target about fourteen to twenty three years old. That's the age they really go for. So I came in trying to see how Data analysis can walk around. If you were coming into this, how do you think that analysis can help in the issues of extremism issues regarding extremism in I don't know how to put it like that's the best way I'm thinking I can put it. But I'm trying to see how you can we can actively predict when the recruiting is going to happen, who they will be targeting. If you were doing this, what are the areas you would look at and what would be the things that would guide you around?

Speaker1: [00:18:51] First of all, you need to you need to essentially trying to classify or predict some probability that a particular youth is at risk of being targeted by some extreme extremist group. Is that essentially what we're trying to do?

Speaker6: [00:19:01] Yes. Yes, exactly.

Speaker1: [00:19:03] So if that's the question, then then why do we have a Data set out there? No, no.

Speaker6: [00:19:09] Currently they them themselves, they don't have any active Data said. But the first place I started to walk through was I went and collected all the data for the economic status of the different areas, how much people are earning because there was a census just the other day here and they collect information like this. How much each household is any? Then I was going to check on the list of the people who are being recruited. What are the common underlying what is the common thing they were looking at? But I'm still fighting a hard way, like I'm finding a very hard place to stand on it. Regardless of what to look at.

Speaker1: [00:19:45] You're going to run into all sorts of ethical dilemmas with this, right. Just from the nature of what it is that you're doing. First of all, like this is like it is machine learning the right way to look. It's probably not like, look, I don't think you machine learning should have any place in this type of situation for sure. So I would say I mean, it all sorts of. Do you have what Data do you have? Available to you if you're trying to make predictions. You obviously have to have some type of training data on which that says, OK, here's a particular individual that was recruited. Here are some factors or features about that person. And then here's a similar person that was not recruited. And here they're, you know, bits of data. I mean, I think I don't know, like this is this is a tough area because the ethics of it, I think.

Speaker6: [00:20:30] Or what do you do you think Data analysis of any sort can help in at least help in any way? Do you think this is it's applicable is applicable way to go around it? Because a lot like it. We've come to notice, like initially people were thinking, at least on the side, that people were thinking a lot of the recruitment was done on religion. So you could use religion as one of the things to look at. But then now it's no longer that you can't look at that because it's money. People are being paid to do it consistently and it's young kids. Do you think there's a way do you think it's technical? Do you think it will actually work?

Speaker1: [00:21:10] No, not with machine learning and this kind of

Speaker6: [00:21:13] Machine learning of any sort.

Speaker1: [00:21:14] I just I mean, what are you trying to do? Right. If it felt like. OK, great.

Speaker3: [00:21:18] Yeah. What's the end goal? I mean, are you trying to help the kids? I guess I'm trying to understand your question for the end goal. If you want to help the kids or or maybe alert some government or organizations, hey, this is happening to the kids between age 14 to twenty three, we need to put a stop to it. Is that the end goal so that these people don't like it or are they illegally recruiting this kid? Is that what it is? Yes, it

Speaker6: [00:21:50] Should be legal because it's extremist group, specifically the group we're doing with undecided's, al Shabaab because of the Somali border. It's very close to here. So what we were aiming to do is to target we cannot approach government by say because we're targeting the lawmakers, but the people who have a say in changing this and actually helping on the ground. So what I'm trying to target is can we reduce the number of kids who are recruited every year get into how do we actively predict the people they'll be going to they've been going to universities, but now we want to see what I guess

Speaker3: [00:22:24] You can say. OK, so so this group who are recruiting the fourteen to twenty three year olds, you say the extremists, are they recruiting from the universities, from home, from school. So, you know,

Speaker6: [00:22:35] They recruited from everywhere. They're recruiting kids from universities. They're recruiting kids who have not gone to school. The highest rate of unemployment has been really high, especially with. So that has given them the opportunity. So they are reaching out. The thing is, they're reaching out online. Is there a script? Should I write and run and check with things? How would you go about it? I'm trying to pick your brains and it's

Speaker3: [00:22:58] Reaching them online.

Speaker1: [00:23:00] Ok, this is this is an area that I don't think is going to be appropriate to talk about on this channel just because of the nature of the topic that you have. And just, you know, when I tread lightly here, what I can't say I'm sorry you should look into some research has been done. For example, you know, here's something that you might want to look at, harnessing big data to respond to violent extremism. Oh, nice. Yes. So there's research that's done just to see what other people have done and try to let that inform your process. But I think I don't think any of us here are qualified to talk about how to use Data science to combat recruitment to extreme groups. Just that's a touchy topic.

Speaker6: [00:23:38] Yes, it is.

Speaker1: [00:23:39] So, yeah, and I wouldn't want to talk about that on comment and also, uh, office hours. So that being said, it's an interesting question. Um, I mean, I wish you the best in that particular domain, but start with just research to see what other people have done and try to have that inform your decision. Right. So I guess what I can talk about is the process by which I'll go about trying to find answers to questions that I don't know how to proceed with. Right. And just being able to research stuff and use keywords on Google searches is a huge, um, huge I think is being able to to adequately search. So maybe something you might want to do. So looking at the thing I did here, put Data and listed analytics extremist groups, I could do data analytics, extremist recruitment. Right. And that we could limit it to educational sites. Right. And you can distill it down to PDF and see if there's any white papers that have been written. Right. And and try to see what other people have done. So that's what I would recommend, maybe play around with some combination of keywords to, uh, to try to find research that's been done. Um, so here's another thing I can link you to as well. Um, so Will will drop that right there for you, that link and the other one that I had a second ago. So, yeah, yeah, so Thóra, so you had your hand up this I mean, it's a huge thing you're working on now, but I can also point you to another couple of groups. I use Data for good and get you in touch with some people there or rather get you in touch with the website so you can find people to, uh, to talk to yakitori, go for it.

Speaker5: [00:25:27] I just technically a small follow up, because one of the challenges I find what we talk about finding Data online. Yeah. There are services where you can purchase Data. But what I'm looking at you and you're doing the searches on Google and you got the cable semicolons and a call. And I mean this. Where have you picked up on and the parameters used for searching for Data? That's one and two. Are there some location websites or other where you can actually go and find free Data?

Speaker1: [00:26:09] Yeah. So for for in terms of finding Data, most major cities will have open Data portals so you can find a any given cities open data portal and they'll have a wide variety of information for that city. For for example, I live in the city of Winnipeg in Canada and we have an open Data portal and they've got all sorts of interesting open data through this portal. For example, they've got they've got a data set that is of all the different trees that have been planted in the city of Winnipeg, that type of tree, the neighborhood, the latitude, longitude coordinates of it. It's really fascinating how granular this data set is. And you can do some really interesting data science projects for that. There's also a data set from the city of Winnipeg that's a distribution of all the parking tickets that have been given in the city. And they've got the type of violation, the time of day, the date, the geographic location. And you can also tie that in with socioeconomic data, for example, census data about the neighborhoods. And you can combine data in these different ways so that you could do some type of statistical analysis. Right. So, for example, naturally, one thing that you might want to do is, OK, if I have the city of Winnipeg's Data that talks about the distribution of traffic tickets that were given, and that distribution is down to the neighborhood level.

Speaker1: [00:27:34] And I also have data that is the neighborhood's socioeconomic stuff, such as, you know, the crime rate, the median household income, average age, average household size, things like this, like this type of data is also available and combine data together and maybe perform some type of statistical hypothesis test. Does the for example, a hypothesis could be, does the incidence of traffic tickets increase as the median household income decreases? Right. Will that help me inform whether or not police are adversely or disproportionately targeting poor neighborhoods and handing out more tickets there as they would in richer neighborhoods if the distribution like is there the same type of tickets being given in poorer neighborhoods versus richer neighborhoods? It just opens a whole space of interesting possible questions that you might want to ask. You are only limited by your own curiosity and your own creativity in a process like this. So definitely that that's one source. Open data portals. If you just look for your city name, open data portal, there's a whole host

Speaker5: [00:28:45] Of because for me that is the dilemma I have right now. OK, as an example, I've been trying to find data on the amount of work and time that goes into preparing for it for an audit. OK, in my industry and I think gas and I have not been able to find anything. I've been on Google searching up and down and back and forth. I ended up launching a small questionnaire just to see if I could get people in my network to start giving some feedback. I do have my own personal experience and also the time tracking that I've used on my own audit where I perform. So you know that there is a stark basically the problem I have is to actually say confirm or support my hypothesis about why.

Speaker1: [00:29:42] So you have to create your own data. And the service is something that you would have to do to to get some of that data that you need. But sorry, go for no

Speaker5: [00:29:52] So, so technical. And that's exactly why I started that survey, part of it. But on the other hand, I'm still having huge challenges searching for Data online. It's like that's I'm kind of wondering if there are sources out there that have. Should Winnipeg go Canada a lot? I lived in Canada for many years. I am Canadian, but for me it's I found it very, very difficult to find Data. And that's why I'm just wondering if you guys have certain sources you mentioned, like the city of Winnipeg. I'm sure try out and say, you know, New York, etc., but also in my mind, a little bit more demographical. But when you're looking more into industry specifics that was mentioned here, COGA, I think was mentioned. Yeah, Carol Kaggle,

Speaker1: [00:30:49] Kaggle, the iDesk, just the website that has a whole bunch of different competitions. And along with those competitions, there's typically Data sites that are provided

Speaker5: [00:30:58] That they made available to people attending, etc.. So I'm just wondering if there is a place where you could kind of do a search on where to find Data support Data. Is there a more a professional site and if not, shouldn't one be made?

Speaker1: [00:31:13] Yeah, so there's just a whole host, a whole number of places. I could show you something that that I've created that will be sent out as part of the artistic science newsletter at some point in the future. But just to give you an idea, um, so these are just different places that you can go for Data. That's right. I mean, this is not going to be released yet. I'm not going to be given this out just yet. But there's even open Data. There's the USA Io, which will give you government data, UK data service. That's an example. You can get data from a financial and economic status. That's by going to the World Bank IMF. So all of these places have all sorts of amazing open data portals and open source data portals. So Google data set search is one place that is available for you to search through as well. So definitely take a look at that. And so, like, let's run with an example and let's try to I'll show you how I would go about trying to find data for a specific use case. So what what is your use case and what is that?

Speaker5: [00:32:17] Basically, I'm looking for preparation time or time spent on preparing for an audit average.

Speaker1: [00:32:24] Ok, so something used. So I would try to start with just you know, I use the quotes and everything, so Data set and then I could say audit to completion. Right.

Speaker5: [00:32:34] All the preparation of it.

Speaker1: [00:32:36] And that's fine. Um, if I could spell preparation. Right. So for data sets and what do we have here, maybe we've got all sorts of different stuff here, but we can even go to Google device, etc.. This is trying to automatically download a PDF, which I don't want to do. But we can go to Google data set search, I think is called just Data to Google dot com or data sets that Google dot com and myself sharing. What screen can you. Yes, still work. So we got the Google data set search. So maybe something we can type in here is just audit completion or preparation.

Speaker5: [00:33:16] There's my challenge.

Speaker1: [00:33:19] I see you got audit risk data set for classifying fraudulent firms. OK, well, you know, you will have to kind of think that

Speaker5: [00:33:27] Looking for it. This is exactly what I like. I just need some good starting points. I didn't even know about this search in Google. Yeah. So, I

Speaker1: [00:33:37] Mean, just go to Google and type in like Data set search and it'll pop up. But there's all sorts of stuff that you can kind of sift through in here. And then what happens is, you know, maybe you'll find that. OK, well, I don't have the data that I'm looking for, but you still get links to different websites and you get links to scholarly searches. Right. And if you can look at research papers that were done, most research papers would have accompanying datasets with it. So you kind of just go down a rabbit hole and and disappear.

Speaker5: [00:34:03] Yeah, well,

Speaker1: [00:34:05] It does take time to find good data, but you could definitely track it down. Right. Part of it is trying to find it straight up through a website, through some API, or you could try to find studies that were done in the field, that or other thing that you were interested in and see what studies for audit preparation time right at completion time for academic studies were done along the lines of this, because I'm sure you're not the first person who has thought about doing something like this and then go read the research. And then in the references section you can see, OK, great, there typically will have the data in the reference section or they'll link to where they're getting their data from. And you can use that as kind of a baseline. Right, or just collect it yourself, which is time consuming and takes

Speaker5: [00:34:47] And technically, that takes me a little bit into the next well related question. I think these tools, as I'm using tools that I developed, which I am stating is going to say I used to type. Now what I'd like to do on. The website, of course, is as people start using the tools, you want to have that kind of ongoing confirmation that is continuing. Now, can that be integrated into some sort of a national solution with the machine learning is kind of following up on the time spent? And what would I need to do to kind of track that?

Speaker1: [00:35:27] I mean, that's a really difficult question to answer from the get go, but it depends on what it is you're trying to do. Right. So what what do you want the end solution to look like?

Speaker5: [00:35:37] Technically, it's the parameters that are involved in very simple terms. I created a tool with four parameters that you choose, and based on those four parameters, it will then estimate the resource and time required for performing or not. OK, so you can do that in 10 seconds. You just put into parameters. Now, this is kind of very early in the stage. Later on, you will do the actual workflow processes in preparation where you prepare the risk assessment and other things that can track Machard, time spent, etc. and then you physically do the audit again. You're tracking the time span, what's been performed, not performed, et cetera, et cetera. And you have a summary. Now, what I'd like to do is to have some sort of a verification. Are the initial estimations that were made on time spent, didn't recommend to resources. How do they actually physically compare to the actual results? And then I want that to kind of feed back into the early stages to correct evidentally if there are large discrepancies. Now, there are many other parameters that come into effect. One is where the location of this audit will be performed, which country, which continent. There's other parameters like the company you're auditing. There is the team constituency, how you set up your data. There is the financial complexity, the risk assessment. So there's a lot of parameters that I know will go into these, quote unquote, algorithms or equations to start to measure. So for me, the initial stage is that I am using my own experience to separate the feed into not on the P&L, but in time. I want the system to learn from all the various variables and then feed back into the beginning again. I don't know if that makes sense.

Speaker1: [00:37:41] Or are you trying to ultimately are you trying to make a prediction and is your prediction based on historical data projecting forward?

Speaker5: [00:37:49] It's it's a combination of historical data

Speaker1: [00:37:53] That's like some raw

Speaker5: [00:37:54] Plus a learning process of actual data, all going to predict what is going to be the next stage and how the next stage will be performed. So that that's kind of like what I'm trying. And right now it's just a lot of ideas running around my head. I know what I want or I think I know, but I'm finding it very difficult to kind of understand how and meld and Praksis good support in that process because I am.

Speaker1: [00:38:21] So maybe it doesn't, because if you're not right, prediction like on the fly real time, it sounds like a lot of what you're talking about is rules based, like there's different rules that are governed.

Speaker5: [00:38:35] That's my excel thinking. I think you're hearing into my Excel saying to me that for me to be able to predict or to adjust, I have to build all the rules, which is what I've done so far. But the the concept as a whole, it cannot be managed. It's not manageable. I already know I have twenty four people testing tools that are created right now. And the Data amount that's coming up, just those twenty four testing is beyond even my own capabilities or

Speaker1: [00:39:08] Even just an excel. Like if you just want to see if a machine learning thing is applicable, just try a simple linear regression and see what happens. It sounds like you have enough data or columns as input to see if you can come up with a simple linear regression, which is easily you know, you can easily do that in an Excel that Dave Langer has, of course, that Dave on Data they can check out, but that might be a good first start. OK, well, can I do this with a simple linear regression? If so, how well do I do? I you know, if if it's worth some within some acceptable range, then we can start saying, OK, well, let's see if I could try some more complex. Right. Maybe I want to move on to a random forest or a tree based methodology, which I think probably would work best in this case. Or you're talking about learning from previous mistakes incrementally that maybe a boosting type of algorithm might work. I'm not sure. But yeah, any any one of those Christophers you have a hand up

Speaker2: [00:40:05] And just the question. I thought, are you interested in learning machine learning? I mean, like really learning?

Speaker5: [00:40:14] I have to admit that I'm joining this to get a try to get a better understanding of machine learning, artificial intelligence, etc.. I don't think I ever will get to a point where I'm actually going to do the physical programing. OK, to me, there are professionals out there like yourself and out there that will be doing this on my behalf. However, to better communicate with those professionals like yourself and others, it is I need to understand the language. I need to have a level of understanding that I can kind of create some sort of a it's like my relationship with my developer right now, all the models that's been created, I've created in Excel so that there are functioning models in exile with its limitations. Then the developer takes that and then I explain, OK, here's a limitation in exile that doesn't allow me to do this and this and this. And then he will actually do the programing to make sure that it works more as a database function with the analysis part of whatever else that's required. So so to me, I'm not here to to become a data scientist or tell their story, any such thing. I see myself as a data analyst because that's what I do in my regular day to day work. And I do it even with my project that I'm working on, etc.. But no, I'm not here to learn the programing itself. What about like, to me, that will it will take too much.

Speaker2: [00:41:52] Why do you say that? I mean, if I understand you correctly, you've got already some experience with Excel and I assume some visual basic,

Speaker5: [00:42:05] But not by the ethical basis that I have goals 20 years back in time when I was just learning to create the queries, etc., I worked in some databases, but I I'm how should I look at myself more as the high flier? A little bit like I have a lot of ideas and concepts, and yet I do analysis and excel at the moment. Is it easy? Should it be possible for me? Yeah, it should be. But given my time and present status at this moment, I just don't have how shall I say? I don't think I will benefit enough because I won't be using it. It's like I mentioned. I have to use it to prove it's not you can be the textbook, but you have to utilize your learning to really get to that now when it comes to my exile and the experience there. I've never taken an exile course in my life. What I have done and how I started with exile was that in a project I worked on, the procurement manager, he had everything built up in Excel. And I was doing the controlling a business controlling and I was just starting out with two plus two equals four. And basically that level now over the next year and a half, we challenge each other. We gave each other challenges. OK, how would you solve this problem? How would you build it? How would you do this next year and over time? That's how I built my skills and and really utilizing Excel to an extreme level without going into visual basic or macros or any such thing, OK. For me, there's a cost benefit as well. The cost of time put in. I don't foresee that I would have enough benefit because like I said, it's not something I need to do every day. And in time there are people that are going to be a lot better at doing it that I ever will be. And those are the people I want to use in my project, for example, if that makes sense.

Speaker1: [00:44:14] Yeah. So for for this particular project that you're that you're thinking about and trying to think of this for any machine learning project, one for to see if it's applicable. Right. Ultimately, at the end of the day, in order for us to use machine learning, we need to have an example, set a training set of data that we can use to identify relationships with. Right. So all these you know, this phenomenon you're trying to encapsulate and try to make a statement again, just start off by saying, all right, can I can I create a data set where it is? Here's one row for a particular audit and for this particular audit. Here are several columns that describe this particular audit. And finally, here's how long that audit took to prepare or whatever it is that night. And, you know, we have one row per audit with a number of bits of information about that one audit, whether it's location or whatever. But how how many different factors they have, I don't know. Like, you get what I'm saying. There are some features there. And, you know, you end up with a whole, know, one row for every single audit with all the features and how long it took. Right. Then you make that your training so that if you can represent the phenomenon you're trying to understand in this type of tabular format, you made one step towards potentially using some type of machine learning algorithm to learn from that. As I think as a first step is what you need to do is can I conceptualize this phenomenon into a tabular format? And if I can, then great, I'm on the path to being able to do something that

Speaker5: [00:45:45] I can give you a quick example. I have, for example, one of the tools is risk assessment, where you have your. And you have a rating from one to five, so based on your last question that you're asked, you don't respond and there will be a rating. So this means that when you go through the 27, you answered all the questions. You are then calculating an overall risk of that. Now, the next program that it said to into is that when you're actually going to go into 90, you would need to have what they refer to as a work program. So from that risk assessment, you will then go through a new set of questions, a new set of five variables per question, which in turn will automatically generate this work. Right. So here's you find the red line. Now, when you go to the audit and you're working on the work program, you will again start making documentation, comment summaries, etc. And when you're finished, you're going to bring that back. And that's when it's going to be interesting to match what physically actually happen if Data you collect what you analyzed, how you did and then trace all the way back to the original risk levels that were generated in the original answers. The data sets that we're talking about here, it's just a question of imagination. That's the limitation of what you. But for me, what's important is to have that red light and then a machine. Learning to me is the solution to find the common ground throughout that process. So you can eliminate.

Speaker1: [00:47:33] Yeah, that's that's depends on what your end goal is. Right. But ultimately, just think about it this way. In order for any machine learning algorithm, whether it's depending on whatever task you want, whether it's predicting some class, whether it's doing some type of clustering, whatever it is, is if you can compress or it has to compress. But if you can summarize in aggregate all the information from this real world Data generating process down to one line and just have a set of examples where it's here's one phenomenon and everything that happened with that I and one real per phenomenon, then you're on the right path. Right. And that we're going to use machine learning to do some type of prediction or group stuff together or, um, you know, ultimately

Speaker5: [00:48:18] Like

Speaker1: [00:48:18] This like this like red line thing that you're talking about. Like that's that sounds to me completely rule based. You're not really making a prediction. You're just trying to that

Speaker5: [00:48:29] The prediction comes in when the system in my mind, like, well, I have X amount of years of study using this and doing the risk assessment, going through the workflow. The next time I come in as a user to do an assessment, the risk assessment, my goal is for the system to actually come back and give the indication based on the location, based on what was done before, to give suggestions. So if I ranked that this particular item as, say, very low risk, but yet my dataset is actually telling me that that is not true on average or in the past or historically, it's shown that it's a very high risk that I want the system to actually come up with a bubble. Like just say, listen, this is what statistically or whatever you want to call it that's happened in the past or in that area. Now, are you sure that what you are saying is correct? So, yeah.

Speaker1: [00:49:30] So that that definitely like like I said is if we have previous examples that we can say that historically this is what happens when we encounter something that has all these features and and then now you're just creating a software that will collect information and that those implementation details will be left whatever developers you have. But the way you're framing it. Yeah. Can machine learning potentially help you in the situation? Yeah. Is going to be easy. Probably not require a lot of effort. Right. Because now you your entire system is not like a simple Excel sheet where you know what I mean. Like, we just, I think excel in something in a pop up and say, no, actually you shouldn't do that. Right. Like you're designing an entire piece of software and out of which machine learning is going to be one tiny component. And that one tiny component can be, you know, just alerting you that you might potentially be entering a value that's not valid. I mean, now, you know, you're just describing the entire software architecture, right? Instead of just one piece of machine learning, can it be applicable here? Yeah, maybe stuff is possible. It's there's not is it's not easy way. I can just tell you I do this, this and this and you've got your solution. But is it possible? Yeah, absolutely. It's possible in this case. So how much time energy investment you want to put, how much development know how you have at your disposal. I'm not sure, but. A significant bit of that, and

Speaker5: [00:50:56] That is a challenge that the next round of, that's one of the reasons that Lawless to meet this group of people, but also some other groups, is to try and get a better understanding of the processes behind and leading up to because to me, honestly, anybody that does Data data analysis that to me it's like I don't see myself as a data analyst as such, but I do use data to achieve things. And to me that is the key for pretty much any job. Every job is a Data you have to evaluate whether it's documents or contracts or data sets or whatever it is to perform. But but like I said, for me, it's really the key to try to understand the principles behind and where it can be used. One of the challenges facing today is that I'm talking to some other software companies, etc. They're using artificial intelligence that's artificial. And to me personally, I don't feel artificial intelligence is there. It's we're still talking logarithms and some sort of a black box. But algorithms say that there's no real I think they're using the term wrongly or incorrectly machine learning, different story. And to me, that is more of a hands on. It's something that is actually working. And to me, that is more the term that should be more regularly used. That's my personal opinion. OK. I don't know if you feel the same way, but that the artificial intelligence for sure is just being thrown around. That's the magic

Speaker1: [00:52:41] Word. Yeah, probably is, most definitely. But whatever, they can't really change people's perception of it. That's not

Speaker5: [00:52:49] What I really

Speaker1: [00:52:51] Care about, changing people's opinions and how they frame stuff. But let's open up for one last question. I see.

Speaker2: [00:52:59] I just wanted to add something because I think many people don't realize that about machine learning is only a subfield of artificial intelligence. I mean, artificial intelligence is much bigger than machine learning. So when they throw artificial intelligence into it, they can also say machine learning. I mean, when they say machine learning, they could have said artificial intelligence.

Speaker1: [00:53:26] Yeah, it's just semantics, right? It's just it's just marketing semantics at the end of the day. Right. It's just marketing semantics how you use it. Whatever I say, that's what it is. At the end of the day, it's it's marketing takes. So are they all if you have a question, let me know. I know you've been sitting here quite patiently saying your name right, Data, who, if you had a question, feel free to add me to my car and and let us know. If not, I said, gee, I had a question. Maybe so if you want to go,

Speaker3: [00:53:57] It's more of an input on foris comment for I'm an internal auditor for Mike, for a biomedical company, and I read tons and tons of standard operating procedures to identify risk. And I can tell you it's a very painful process because I read tons of documents. I'm also trying kind of like, you know, she's trying to figure out if I can use it to identify the risk in this processes and procedures and all that stuff. I'm still I did find one company in Canada who is somebody who's created this internal audit for biomedical. But they don't they they they devise their own system, I guess, in auditing biomedical documents and stuff like that. But, yeah, I'm in the same shoes, too, because the pipeline is long, because you have to audit different departments sales and you need customers and department and FDA and then you have the FDA, the picture, because they have to, you know, make sure that the biomedical assays they are producing is good and non FDA products have their products. So I'm also thinking the same lines of like, what can I do to automate this entire process so that, you know, if I can do something much more useful in other fields and yeah, I'm still figuring that that part out like I want to I mean, to automate the entire internal audit process and I can do something else. So and the other thing is, I think natural language processing would be a great one, because when you're reading documents, it's a lot of text and you can definitely identify stuff NLP can identify a lot of things and, you know, kind of stuff. But I don't know how to go about doing is a huge project in my head and I'm still trying to learn NLP while doing this at the same time.

Speaker1: [00:55:50] But you don't need to do it yourself. They're products out there that are. Developed so that they take these massive documents and compress it down to give you the most important bits, right. So you don't have to necessarily reinvent the wheel if you can just get your hands on these products to make your life easier. Again, I won't say you can look up, quote unquote, internal audit called machine learning. And you can see that this is an area that people are exploring. Right. So you guys definitely aren't the only ones looking into how to do this. My recommendations is just to kind of read through the research. Yeah. Yeah. And see what other people have done. So here I had to just I like to read white papers that are more academic just because that's easy. You can look at blog posts and stuff which are more high level that hopefully have references to other places they can go learn more from. But yeah, it's a thing, right? Internal audit, machine learning.

Speaker3: [00:56:43] So this is great. Yeah. Because I if somebody has already done this already have like a like a good Harp report or something like that, it's already been done. I would love to take that and repurpose it for, you know, organization. They would be great in that entire thing.

Speaker1: [00:57:00] Yeah. I mean, they probably won't find a GitHub repo for it, but we can always check. Yeah, but you can see what other people have done and and be inspired by their process and apply that to to what you are working on. And I tend to do this quite often. So I spend their time in

Speaker5: [00:57:18] Research and stuff. If I, I'm doing the exact same research because this is something that I want to integrate into my own software solution, which is targeting the oil and gas industry. But right now, today and this past few days, I've been looking into teammate who truly has advanced quite a lot since the initial stage a few years back when it was just that simple software package. But they're definitely moving in that direction and they are specialized knowledge. So I wouldn't say it's precise, but they do offer for the biochem industry and among others. So you may want to have a look. I'll send you a link on teammate solution that they can look into them.

Speaker3: [00:58:02] Ok, yes, that's great.

Speaker5: [00:58:03] Yeah, NLP is definitely a way to go because personally, when I go into an audit, I will be reviewing between 15 and 20 contracts. I would love to have a tool that would just given Tallentire contract take out all the terms and conditions, match, give me page references I still need to read. But if I could just get a page reference, that alone would be a huge, huge help for me to do my job. Instead of going through the index and page by page by page, I can then go directly to a click on that. It brings up the page, the paragraph, it's yellow out. You know, that's how you should be able to do it.

Speaker3: [00:58:46] What happens for me is I did the standard operating procedures and then I have and then I have to interview the individuals who are working the department go step by step. Is this the process working right now? And, you know, why aren't you doing this step? Is this line being used? And if the farm is being are you following this that so there's a lot of manual work involved. So that's why I think NLP or any other tool out there can automate the entire process.

Speaker1: [00:59:16] And then there are software solutions that exist that I implore you to look at. You don't have to build stuff up from scratch. It'll take you forever. Right. And leverage other people's experience and stuff on this. I'll let you two talk about this offline.

Speaker5: [00:59:30] Yeah.

Speaker1: [00:59:31] So any other questions last time for for a question. Otherwise we could begin to wrap it up. I see Christoph as a quick non defense related question. That's also

Speaker2: [00:59:40] Not important. So if you were in harridans, you can wrap it up for today.

Speaker1: [00:59:46] No, no, no. Is that we've got a quick question.

Speaker2: [00:59:48] We came out OK, because I'm just curious. And I think it's the question on you, Harpreet, because your network on LinkedIn is like a hundred times bigger than mine. So my question is, did you already deal with haters?

Speaker1: [01:00:05] Um, yeah. I'd post some stuff and then people will just like comment on the blog, although that's not it. What about this thing? What about that thing? And I was like, OK, well dude, I've got thirteen hundred characters and maybe a few slides like I'm not going to cover. This is not a academic research paper. Yeah. Great. A thank you for your input. Appreciate it. Like I mean at the end of the day, like I just don't really care about those people. And if you, if you, if you have nothing better to do than try to pick the semantics or talk about some really obscure edge case related to my post, then yeah.

Speaker2: [01:00:43] Yeah, yeah.

Speaker1: [01:00:45] I mean, at first I was worried about it, right. At first I'd get so worked up people I could comment or say something because I missed one. Or maybe they just didn't agree with it. All right, well, cool. That's that's right. And you are completely entitled to your opinion, and I'm completely entitled to not care about it. Right. Um, so, you know, just just as much as people are entitled to say whatever the hell they want, I can just ignore it. It doesn't really affect me that I think. But I don't worry about haters on LinkedIn. There's a lot of people.

Speaker2: [01:01:17] No, I'm not worried. I was just interested because there are different people. And it's obvious that what you do is great for the community. But because there are so many different people. I was just asking myself, did you already have something in my experience?

Speaker1: [01:01:36] Yeah, it's possible. Again, possible people. People probably if there's probably comments on YouTube, I haven't checked them, but they're probably comments on YouTube. Like this guy doesn't know what the hell he's talking about. And it's probably true. I don't like I just haven't seen, like, every single use case of machine learning that ever existed. Right. Like, that's the beautiful thing about this field. I'm always going to be a learner. I'm always gonna be learning things. I'll never be an expert in Data science. And I don't want to be an expert in data science because I always want to be learning right. I always want to be pushing to the next level. And if it comes to a point where I do become an expert, Data science will then let me try to find an intersection with Data science. That is interesting that I could then start exploring. But yeah, that's kind of my viewpoint on it.

Speaker2: [01:02:16] I just say thank you. Yeah.

Speaker1: [01:02:19] Yeah. Don't I encourage you. Just post, keep posting, keep posting.

Speaker2: [01:02:24] And it wasn't about me, it was like asking who was really out of curiosity. Yeah.

Speaker1: [01:02:33] Yeah right. I mean um well if there are any other questions I want to give it a chance to ask a question because Adeola has been sitting here patiently. So if you have a question my friend, please let us know this is your opportunity to go up. Uh, does not look like you have a question. I said, do you have any last minute questions or anything. Cause I know you were having cuz

Speaker6: [01:02:57] Yesterday I did not know. I keep dropping off. My Internet provider is not on the best side today. So I have a question that I have a question in regards to data science. I enjoy the work. If they are feeling data science where you do not have to be forward facing with people to get it. I mean, yeah

Speaker1: [01:03:20] There's there's with any I think any technical career path, there's always two different routes you can go. Right. Typically if you want to go on that management route, then you'll have to be dealing with people and, you know, Ford facing and things like that. But they also have there are also parts where you can climb up the ladder and just be a technical employee, but still get paid just as much as like a manager would. Um, and I think that's that's a huge thing in the tech field, I think, is that there's actually parallel career paths. You can continue to move up and just work primarily on building models and writing code and things like that. You probably more likely to be an engineering type of position. So maybe Data engineer, machine learning, engineering or research, Data scientist, um, those type of roles. If you're working, I think in a product Data scientists kind of role, you're probably more likely going to be working with, um, people. And in that case, does that kind of answer your question? Sorry, I saw somebody working as Chris I've just about now. I said that kind of answer your question.

Speaker6: [01:04:23] Yes. Yes, it did. And I also saw on the email you sent out the newsletter, there was there was a resource you had attached for communication, which also helped. So thank you for that.

Speaker1: [01:04:35] Nice. Which one is that? I forgot.

Speaker6: [01:04:37] I have to open up my phone. I'm sorry. My network is a bit shaky and I

Speaker1: [01:04:40] Know no worries. Well, I'm glad you found it useful. Really glad you found it useful. I try to give away as much stuff as possible, the newsletter. So hopefully that's helpful. But yeah, there's you can you can climb up the ranks and Data science without having a Ford facing or a managerial role. Those type of career paths do exist where technical people can move up. Um, you know, Google is a great example of this. Google kind of has two career paths, right, where you're just moving your way up and technical role, but not having to deal with the communication, storytelling people aspect of it. Um, and most companies have tracks like this.

Speaker6: [01:05:17] So in the meantime, I have to work till I get there.

Speaker1: [01:05:21] Yeah. I mean, it depends on the organization that you are in and the culture of the organization. Sometimes if you're in a small startup, you have no choice, but you have to try to deal and work with with people. But in large organizations, they typically have multiple career paths. Right.

Speaker5: [01:05:37] Thank you.

Speaker1: [01:05:38] Do you not like working with people for some reason?

Speaker6: [01:05:40] You know, I just sometimes I tend to be very introverted and very get lost in the computer and I would rather get lost on the computer forever if I could. But clearly, it's not maligning. With my father right now, so that was an option to stick to that that you pointed me to. So thank you.

Speaker1: [01:06:00] Yeah, probably machine learning engineer, Data engineer type of roles, most likely guaranteed you will be lost in computers all day. So those are two paths right off the bat. Data science rules themselves. They could be either or you could be a data scientist just researching and doing stuff for now, working out products. So something is available for you there for you. Thank you. Personality type. All right. Cool, man. Well, it looks like there's no other questions. Thank you guys so much for being patient as we got this started a little bit later than usual. Be sure to check out the podcast released the awesome interview a couple of days ago with Dr. Paul Pagad. He's a cognitive scientist, philosopher. We talked about ethics. Um, he's teaching at the University of Waterloo in Canada. He's professor emeritus from there. Just pretty awesome. Waterloo's like the Harvard of Canada, I think called that were the MIT of Canada. So definitely had a great conversation with him. And there's officers coming up next Friday as well. So make sure you join us for that and again next Sunday. At the same time, our guys will take care of the rest of the weekend. Remember, you've got one life on this planet. Why not try to do some big cheers? Everyone says.