Uber Driver to Machine Learning Engineer in 9 Months! (@Daniel Bourke) - KNN EP. 05

and so the three things that I like to think about when I'm like working on anything or any project is I feel I feel really good when I'm when I can be smart happy and useful the last ones really really important hello everyone can hear about with another exciting video for you today I have the pleasure of interviewing Daniel Burke he's a fellow content creator and machine learning engineer Daniels from Brisbane Australia and you know he has a cool accent and he brings a very unique perspective about what data science is like in other parts of the world Daniel is completely self-taught in machine learning along with the help of quite a few online courses and certificate that he has found absolutely amazing and he absolutely lives to experiment in less than a year he went from not knowing Python and driving uber on the weekends to landing a machine learning engineer position aside from machine learning his two favorite activities are writing and fighting he's actually in the process right now of preparing for a Brazilian Jujitsu grudge match it's actually between him and myself and whenever things open up I think we're gonna we're gonna get after a little bit and see who the better man is watching Daniels YouTube channel or reading through his blog are not only fun and interesting activities it can also make you about a learner a better creator and potentially fitter and healthier finally he's currently the owner of YouTube's self-proclaimed best mullet I don't think that's something I even want to consider challenging him for right now so without further ado let's jump into the interview all right Daniel thank you so much for coming on you know I think it's it's really relevant to have you here because you know my educational background of data science is very formal you know I've gone to school for two different master's degrees to get you know into these positions and you've come from a very different background where you were very much self-taught but you made it very structured and and very intuitive and so I think that that is fascinating and you know I obviously actually wouldn't recommend going the path that I took so I think that you know everyone here is gonna get a ton of value from your perspective especially on learning this field what I really like to start with is understanding you know how you got interested in data science and machine learning to begin with well okay it is amazing to be here so yeah how I got started to begin with well we kind of flashback to maybe I think the start of 2017 so put it this way I've always been a tech nerd like just interested in technology in general and 2017 I was working at Apple but at the retail store so one of those geniuses ones that like fixed your computer when you have have something wrong and I like love that to death but I got to a stage where I was like you know what I want to start to like I was working physically on the computers whatever I'm like I want to start writing code like I want to start building apps and whatever on these machines that I'm servicing every day iPhones Mac's whatever I want to start building code or building anything really and so I started to learn to code like three different times or maybe maybe more but I can remember distinctly like three me and a maid had an idea like we wanted to build an app so we were starting to - to learn - to code for iOS because we had iPhones and whatever that fizzled out after a couple of months and then at the start of 2017 me and another friend we were working out together all the time but we kept running into this problem where I was signed up to one gym because I lived in another part of town and he was signed up to another and every time we tried to like train together there was it was just a pain in the ass because you'd have to go to one gym say hey I'm here visiting a friend and they'd be like I'll sign all this paperwork etc just to get in the door and so me being a tech nerd I was like surely there's a way to to solve this like with just an app where you can just basically go to any gym that you wanted to and we were like why don't we just build that so he had some experience building websites and whatever and so I'm like okay well this can be the third or fourth time I'm gonna learn to code and this time I'll try to build something to to solve our problem anyway we've got like three to four months into doing that it wasn't really anything in depth web development it was just like essentially a wordpress site that would show you a map of different fitness facilities in your area and then you could like put in a form of what your details were all using just WordPress plugins but when we actually here's the thing when we actually started go to gyms we worked out that a lot of gyms revenue don't results are comes from people not showing up so our beautiful idea that we had prototype didn't like look at our website we can get more people into your doors and whatever we kind of didn't have the one thing that we needed and that was permission for people to actually use our website to enter the gym that they were like signing up for anyway so after we kind of figured out that we had this massive roadblock and that was that our whole product and our whole idea depended on on Tim's and they didn't really want to help out it kind of fizzled out so it was called a knee gym we had like this cool prototype going and whatever it was all like web development I kind of got a little bit bored too by the end because I found like putting together different WordPress plugins to me was like I'm not really coding here I want to like I want to actually write some stuff anyway that was start of 2017 and at that time it was like peak like MLA I data science in the news everywhere and so I think if you if you get interested in any kind of development you're gonna hear these things and so because I was learning the tech and whatever trying to make it that's all I could see I was like our machine learning is gonna do this and all this sort of stuff then I figured out okay what's this machine learning thing and then I realized I was like oh wow you you don't explicitly program the rules the computer learns it for you and so I'm like well after I'd spent a whole bunch of time like trying to place different CSS tags and HTML layout a website that look really good I'm like you know what this isn't really my style I'm gonna jump into this machine learning thing and so I I did on my where can I start learning machine learning I stumbled across Udacity at the time they just launched their deep learning nanodegree and it was starting in like two weeks and I decided to sign up and then after I signed up I realized you know what I haven't got any of the prerequisites that I need so I then started learning Python like two weeks before entering the deep learning nanodegree and basically finished every project like four days late of the deep learning nanodegree because i was trying to learn Python at the same time as code deep learning models but at the end of that that was about three months I think and so now we're halfway through 2017 and I was like you know what I really love this I'm gonna study it properly rather than just sort of floundering around because the deep learning that Olivia was kind of for me just a hello world to the field and so I put together my own artificial intelligence I'm saying this in inverted quotes if you're listening to this artificial intelligence master's degree where I collected different resources from online and said you know what I'm gonna study this for the next foreseeable future and see where it goes and then I guess that was the inception point and now where there was almost I would say three years ago probably almost this month that I started that IO master's degree and now we're here it's been it's been good fun that's awesome you know one of the things I love most about your story in general is that you you really clearly have a bias towards action I mean even though you know you started to learn coding you know three or four times and it fizzled out you at least start up right and it takes a couple it takes like starting to learn something a couple of times for it to actually gain traction and you know like your initial introduction of this field is because you wanted to build this this any gym platform and you did it right without necessarily having the prerequisites or anything like that same thing with the the deep-learning man agrees he jumped in it you know at least feet first and yeah I think that that's something that buys Ford's action is something they could really take anyone impressed in this field very far because you know if you get if you get past the worrying about where to start it's so much so much easier to learn so you know one thing that I'd love to talk about is you're just you know before yourself created master's degree what was your education background but you know I get a lot of people like what should I study in college to get into this and you know I kind of think and you're a perfect example of this maybe it doesn't matter as much as you think yeah well just just to touch on that I think if I read the prerequisites for for the deep learning and agree if I before I signed up I might not have signed up that's kind of like a little bit of blind ignorance there but it all works out but um yeah what I studied so beforehand I've actually had quite an eclectic fruit this why I finished high school in 2010 and then I I started University in 2011 studying a biomedical science degree so basically in Australia to get into medicine which is what I I thought I wanted to get into you have to you need if you can't get directly in and to get directly and you need some ridiculous like test results and all that sort of stuff and I wasn't that person because in high school I preferred to play call of duty then study so I signed up for this biomedical science degree which is kind of like this stepping-stone to get into medicine and truth be told like the only real reason I I chose that was if I reflect back was because one a girl I kind of had a crush on was studying it and two I thought it would be cool to tell people I was studying to be a doctor so it was more of it it was more of a status thing and just following this this cute girl then me even considering what I wanted to learn and put it this way I really didn't know like I just thought that that going to university after high school was the thing that you do I really didn't know any better anyway so I studied that for two years and failed pretty miserably because first of all I didn't do any biology in high school so I was again trying to like learn pre-medicine biology in University at the same time as I remember actually one of the in like the first lecture of the biology introduction is the lecturer was like look left and right because 30% of you are gonna fail that means one out of three of you I'm gonna pass this course and I look to my left and right and I'm like that's not gonna be me turns out it was me and so that was like two years the first two years I had to have a meeting I got an email one day from a science Dean which basically said hey come see the science Dean about your results I went and saw him and it said and he said look what's going on with your results it was basically like a why should we keep you at University kind of thing if you're just gonna fail all of your subjects and and not pass or whatever and I was like oh wow this is actually kind of serious and so that was like a good wake-up call actually and so I gave a bunch of excuses like why I wasn't doing too well the real reason was just because I was just lazy and it wasn't I wasn't studying the thing that I was that naturally sparked my interest and I mean of course I could have actually got into it and then it probably would have the interest would have came later after I got better at it but I found myself on the side I was into I was into weightlifting and a heap of sports I found myself rather than studying for the exams I had at university I was like researching hardcore online I'm talking watching YouTube till like 11 p.m. at night later of just um how people would eat and train for different sports so I was getting really into nutrition on the side and then I kind of told the science Dean I was like he uh he asked me do you want to change to anything I'm like well I've been learning about nutrition can I study that and he's like yeah why don't you just change to study nutrition and I'm like huh so you mean I can just change to this thing I'm already studying on the side and like do it properly at university and he's like yeah I'm like okay I'll do that and so the next semester which was I think the start of 2014 maybe cuz 2011 two years failing and then the next year anyway the years don't really matter but as after the two years of failing I switched to nutrition and food science and basically got top marks without even really trying because I was already like studying this hardcore on the side like watching videos reading books trying experiments myself and that's and that's I think the biggest takeaway I actually got from University so I end up graduating in 2015 with the food science and nutrition degree it probably would have been like honours or whatever not that I really cared but because my GPA was already so low from the first two years it kind of it wiped out any chance of getting like good marks later on and so 2015 food science and nutrition and then 2016 before I started studying deep learning and machine learning and data science in 2017 I actually studied um languages for two years I mean a year so Chinese and Japanese I've basically forgotten it since then because if you don't use it you lose it right but the biggest takeaway I got from university I think was learning how to learn and you don't necessarily need this to go to college to get that skill but once I figured out like that that light bulb moment of sitting across the science stand going hey you can you can study this thing that you're already interested in that was like like just a mine shattering moment for me cuz that's what sort of got went and told me like I learned the lesson I was like wow I can if if there's something I'm interested in I can just go and learn like I don't necessarily have to follow any set pathway like it's just like oh there's that thing you know what that's already sparking my curiosity I'm gonna go try it out and that's kind of the exact same like learning how to learn lesson that I applied to machine learning that kind of got me to where I am now I mean that's awesome it's a shocking movie we have pretty similar educational backgrounds in in college I think I went through like at least four maybe five different majors I was an awful student and then I granted schools for four sports and I started studying economics and I just learned like top marks all the way out because I thought it was like fascinating right and you know that's one thing that again for anyone watching us I think that is a really big takeaway is that winter at school I mean if you want to do data science it does help to go certain paths but for example if you were choosing between like statistics and computer science like a hundred percent do the one that you're more excited about right if you're excited about the coursework if you're really interested in learning it you're gonna put so much more effort into it you know if you don't really I mean if you know some sort of you know like object-oriented programming isn't gonna like really get you excited if that's like what your whole degree is based on probably not gonna try too hard in your classes you're probably not going to use it too much outside of outside of what you're doing so like I I really like that and I think we're both lucky in the sense that we found something we were really excited about and were able to experience that because so many people they were in the same position that you were where you were studying something you didn't like and you know they're doing enough to get by and they graduate and then they're like oh my god what do I do I don't like anything yeah so I think that that's like a very very important for an idea so after you you know we're struggling through this deep learning a non-degree when you really started kind of getting into machine learning education what type of concepts did you start with you know what's it do you start with some some kind of projects did you start with like you know a couple different forces what are the beginning steps look like for you or what would you recommend beginning steps to be for someone that is again trying to learn this field from scratch as like who has an out exposure to it yeah well so yeah what is I was kind of learning Python at the same time it's the deep learning nanodegree so I I distinctly remember actually one of the first concepts that I came across was um like convolutional neural networks and I remember sitting on my friends so this is like okay so the way the way I sort of prefer to learn is with anything is just trial and error like I like just getting hands-on as soon as possible I can't sit there and go through theory to begin with for me it's just like I need to set up an a string of experiments to even get interested in something and so what I really liked about the deep learning nanodegree was kind of like a just threw me in the deep end basically and I remember sitting on my friend's bedroom floor this is like the friend that we were I was hanging out with heaps while because he was like my partner in building the the web startup they were working on and I was like just coding away on the floor and I like got up and I showed him I'm like dude look at this I think it was a like an image classification model on the CFR 10 dataset or something like that and I'm like look at this the computer knows that this is a photo of a plane and this is a photo of a dog and if I if I press shift and enter which is the Jupiter notebook of course and it feeds in another image it says like this one's a dog but I haven't told it what's an image of what it's just kind of well technically I have it was supervised learning but it just like it figures this out for herself and he's like whoa that's insane and kind of like I was like yeah this is really cool and so I was like that got me hooked right because I I figured that's what a lot of like machine data science is is like especially when you code in like a Jupiter notebook or something like that okay everyone's got like a different opinion of like where you should um like people like ID ease and you should write scripts and all that sort of stuff but I I really liked maybe that's why I failed to learn to code a few times because it was just in a in an IDE in a script and to get the results you kind of needed to it kind of need to compile the program and have it almost ready to run anyone's burger career yeah exactly right so when when I was figuring out like that machine learning and data science was just a whole bunch of just small little experiments to try and find something I was like boom this is this is me and so in saying that what I found worked for me was combining doing the experiments first and then digging deeper into like what the theory was so for example I mean I used Udacity there's a whole bunch of of different resources out there for me personally like a lot for the for a lot of people Udacity is is quite expensive but I found that once I'd invested that much I was taking a lot more seriously even though I'd already spent like a whole bunch of money at university on like loans and whatever I because I couldn't really see it I didn't take it as seriously because I'd like physically seen me put my credit card or debit card into like Udacity payment portal I was like ah sweet this is a is actually taking my money so I'm gonna take this really serious but then I and that's not to say that anyone has to we're kind of getting off track but anyway paid resources free resources it doesn't really matter they're all great but that was just that's just me personally anyway I found that I was using a combination of okay do the heap of experiments and then I found it a lot easier to go down into the the rabbit holes of the theory because it was like okay on the left here I've got the experiments that I've I was like well why is that coming out like that and then so I would go and dig in and I go oh okay so there's the the math behind the last function that I'm using etc etc so it was and in fact that's actually still how I learned like it's all I'll open up some sort of notebook or whatever and I'll go through a series of tutorials or experiments hands-on with code so I can actually see what's happening and then if I don't understand something or if I want to dig deeper or push something towards its limit then I'll go find another resource combine that and go well here's the theory behind it and then it'll just be like a building block from there but it's always it's trial and error driven I don't learn any other any other way like it has to be trial and error so I got started with a bunch of courses to do to learn foundational skills but where I started to really learn is when I started to build my own things or own projects and so I got really fortunate to get some experience at a machine learning company in my in my city and basically that's all we did we um we worked on six-week projects and this is why I kind of think it's it's like if you're learning whatever it is like courses all the courses online basically all of them are great like it actually doesn't really matter where cuz it's like all the same code just very good yeah I'll plug that one but uh find someone who you resonate with with in terms of can explain whatever concepts it is and and just use use that resource as your foundational skills now this is me looking back and like how how I could have how I how I did and whatever and then I actually yeah going back to the role we were working on six week proof of concepts with different companies and so all it was was I would take the things I've learned in courses and various blog posts and various books and whatever and then would apply them straightaway to do some sort of project we were actually trying to solve and then if we didn't get as good results in the project we wanted it go back to the the theory or back to the research or whatever and then back to the project and it was just like back and forth in a in a pendulum type motion so that's like a long-winded answer of kind of how I got started with in terms of like concepts versus theory versus whatever style to sum it all up is basically just run as many experiments as you can your main your main in terms of learning your main like thing or your main idea your main goal is to reduce the time between experiments that you do so if you're like if you're like setting something out that's especially when you're first getting started if you're like setting something out that's like massively along an elaborate something that'll take longer than a weekend to do you're probably like over complicating it and so that's like what I tell people sort of if they're starting to learn right now is that okay of course is a great for foundational skills but if you want to build upon those foundational skills you have to put them into practice somehow and that can be through a project or through whatever the term project is very very very broad and a lot of people are like oh what do I do what if it's not new and all that sort of stuff well take something that already exists and just replicate it and then you'll you'll understand it a lot more than just looking at it awesome well you know one thing I really like that you said is that you know is reducing the time between the projects or the time of iteration you know a lot of people think they have to sit down and do a whole data science project you know the collection they're cleaning all these different things in one sitting right and that's not that's not how it works I mean when I made the data science project from scratch series which I hope is one of the more helpful resources related to this I took each section like the data collection the data cleaning the EDA model building all these things and I they felt like individual projects to me right like I did one at a time and I spaced them out and they you know each one didn't take more than three four hours at a time and you know to your point it's like look I can break a big project down into these little sub projects they still count as projects on you or like one big project when you put them together but when you work try and have like discrete starting and ending points for each you know part of the project that makes it so much more manageable and so much more kind of attainable at the same time one thing that I'd really like your thoughts on is that you know the whole thing with experimenting is trial and error right one of the things that I think a lot of people struggle with is when they get stopped when something isn't working what are some of your best recommendations for you know getting you like dealing with the failure which is very common in data science when your your stuff doesn't run or when you're not getting as good results even though you're in theory using better models you know how do you get past that how do you you know what what encourages you to keep going and keep experimenting yeah what I do recently actually if like I'm running into a whole bunch of boxes it's just uh like take a break like I literally go outside go for a walk it's like it's it's simple it's like this I don't I would always I would always find myself actually in terms of asking myself that question that you just asked and I would always look for like some sort of secret you know like how do i how do I find the answers out of all this and I like reflecting back on it it was like you know what this this this happens to everyone it's not it's not new like you're gonna run into I mean truth be told if you're not running into errors you're not trying enough things so it's like part of the parcel one one big thing actually that I like to do is if I'm working on any kind of project it's always it's always more exciting or easier to get back into it when you when you know what your next steps are and I mean you're not always gonna know it right so if I'm working on something oftentimes all um like if I'm writing a code selling Jupiter notebook I'm actually doing this now with something that I'm working on is that if I'm going to take a break or something I'll like leave a code cell like half half done and then I'll write like a comment like my code will be just for a simple example model dot Fi and then cut like for fit and then it'll be a I'll just do model to Fi and then hash tag like for the yeah it's like finish finish this finish this like that's my next step and then so I'll come back and if I look in if I'm searching in a project I'll often do command F and then I'll usually have next or up to here and so then I can just like go back to it and sit down and just go command F next and then literally just like start going back on the exact same part even though it would have taken me like 30 seconds to to finish that before I took a break having that little bit of like unfinished business is kind of like a momentum builder for forgetting back into it and so I'll even do that when I'm writing as well like almost almost I found it's really helpful but will you run into like where I personally run into I mean it's really similar to a lot of people is when you're not sure of what to do next then it's really hard to like sit back down and and like dig into it maybe you've ran into some sort of CUDA error and which is like the formidable error and you like you've like you've been through almost every single github thread that exists and you're like well I just can't get this error fixed and then you just work out that you're just tensors we're in the wrong shape like at two days later I've had that happen to it most times I think that there's a lot of value in that is that sometimes in the moment or when you need it you know you know I've got to solve the problem and you should give yourself a certain amount of time to solve it and then just move on you know some people that's so stuff that they stop altogether you know one thing that I do is very similar to what you do is that are just like do like six comments like these are the six things I'm gonna try if some of them work if one of them works great I can keep going if none of them work it's like okay I tried everything that I could let's move on to a different part of the analysis or like let's just step away for a little bit and think of more things that I can do I mean they you know like ask you can always ask like I find the amount of so for every question you probably see unlike Stack Overflow or some github issue thread it's probably like a thousand that are unasked so if you are if if everything and this is what I found actually - like if you start to to phrase like the problem that you're having in like English terms or like just in in in semantic terms I talked it out it actually starts to make more sense like when you or sometimes it does sometimes it doesn't so I would have when I was programming working on another project with someone else he introduced me to the concept of the rubber duck and so it we would treat each other as if we were rubber ducks like sitting in the in the bathtub and then we would explain our problems like off rubber duck I'm trying to to merge this data frame with this other data frame I'm not sure like the order that is first oh wait I should just go on that column and oh I see what I have to do now and so like that actually so so yeah what what you just said of like leaving six comments or whatever that's kind of like a similar thing it's like you're just talking yourself through your own problems and it's the same thing with one of the recent projects that I worked on well I had this document that was basically just like a a thread a rolling thread of whatever thought came into my head about the project and it was so I was like I it went for 42 days which is like six weeks and it's like day one this is what I did and I literally it was like a journal and then it's like day two this is what I'm gonna just for everyone that was something I like started doing when I was working at my machine learning engineer role was every day I would kind of send a message through on slack just saying here's what I did here's what I'm gonna do next those like it was just three bullet points like here's what I did like three bullet points maybe less if it was only one big thing for the day and then the the next was kind of like here's what I think I'm gonna do next and it was out in the public like it was with the whole machine learning team and that way it was one first of all was communicating to myself my intentions and second of all it was like well hey this is at me open if any one kind of has advice or disagrees well it's it's right there so you can you can interact with it if you like you know it's funny I do that every night before I go to bed in my little journal here little bear thing but I write down at the top here all the things that I did like pat on the back type stuff over here like the things that I learned and then below I talked about what I'm gonna do the next day I think that's like a really good practice and you also look if you look back over the like over a year you could say holy you know I learned so much over the course of this year even though if you're just kind of thinking about it and you're not tracking it you're like oh you know I guess I learned some machine learning concepts whatever it is it's it's it's nice to keep a list of the things that your accomplishments because we very much forget about those things you know there's two other other concepts I wanted to kind of touch on with you is you know you have all people I think does a really good job of tying your interests into your your work and into the content that you produce I'd really love to know you know how you're able to do that like if you have some recommendations of finding like project inspiration I mean I got so many people that ask me he can like what project should I do and I always answer is always like do so if you're interested in but you know again you I think you've done a really good job at like learning about yourself and understanding what makes you tick and then you're able to transfer that into to a machine learning project or something else you have any kind of insight on I mean I guess that was really big but that kind of idea yeah well yet my my advice generally - is for the projects is find something you're you're curious on because at the end of the day it's um it's also the it's something that might not work that's what intrigues me a lot so something that might not work that's I think what stops a lot of people as well it's kind of like a paradox it's like the best projects are things that might not work but what also stops you is because you you you can't really forecast how it's going to turn out and I think if you asked like anyone who's worked on anything worthwhile it probably hasn't turned out exactly what they thought because chances are like you'll start and you'll be like yes I'm gonna build this amazing thing and then like for example I listen to like a pretty in-depth podcast with Elon Musk's the other day and I mean if you want a great example of just like a company or or someone who's building something that went through a heap of adversity I'm pretty sure everyone knows Eli mask so you can you can go check out the history of Tesla and see Wow okay they had this they had this overarching goal but like the steps that to get there were just like just insane like they had to change yeah exactly right and so at the end of the day that's what's gonna happen with whatever project you're working on so if it's going to happen to to a company that at the moment looks is incredibly successful as Tesla well it didn't start out that way I'd say that's the same weave no matter what and I'm not saying like whatever project you want to work on it's like compare it to to the one you worked on like the if you're if you're worried about like oh this is not the the latest and greatest because these are the worries that I have if it's not the latest and greatest thing or if it's like not not mind blowing you knew or whatever it's like no don't don't don't compare like your progress to someone else's progress because like it's it's it's it's two different lanes there's two different things and so the three things that I like to think about when I'm like working on anything or any project is I feel I feel really good when I'm when I can be smart happy and useful the last ones really are really important so it's useful to to others so I like to think about in terms of okay what's gonna make me feel smart that I'm working on like whatever project because that's intellectually challenging what's going to like keep me happy so again part of that that keeping me happy is like that intellectual challenging they it's kind of like a circle it's not like three separate categories and then useful it's like what's something that but when it's finished that would be either useful to to me six months ago or useful to to someone else who who might might want to learn more about whatever it is that I'm working on or be entertained because that's kind of like the crossover and it's like if whatever project you're working on it kind of it needs to serve a purpose it needs to educate someone or it needs to entertain some some of them have the combination of all three so I go back and it's like okay something that might not work it has to be cuz that's intellectually challenging if I know exactly how it's gonna turn out well then that's kind of boring for me I only really learn anything when I'm surprised and so it's got to keep me happy so it's got to be like something that I that aligns with my my values and interests and three it's got to be useful in at least some way and so so useful again is a broad term and so I could I could work on an entire project to just to have a blog post skeleton of what it's like to work on your own project even if the project turns turns out to be null and void like whatever result I was going for it turns out not not possible like the whole purpose of the project could be to have that blog post and go you know what here's what I learned here's the steps I went through take it or leave it if you're working on your own project here's the things that that I did that may may not work for you may work for you but there's a there's a story attached there so yeah smart happy useful and things that might not work that's uh that's my take for projects well you know I really like that I think a lot of people are scared of failure and that you know you don't realize is that I think data scientists our character by their failures like a good data scientist has filled more than the other data scientists who are not as good because you it's about experimenting it's not trying new things and if you worried that oh I didn't get great results I can't put this on my resume that's not true at all I want to see a project I like beautiful projects that didn't turn out as you thought they would on your on your portfolio or on your resume that's something that shows that like you're human that like you realize that not every project you're gonna get a great result you know I think that we might be programmed a little bit differently just because of our experiences you know I like I relate a lot of things to sports I play golf growing up right and golf is a game that's characterized by failure like the greatest golfer ever is one less than 20% of the tournaments that he's you know participated in right and and to me like you know I want a lot less than that right and so you know what they decide it's like you start help and you you know you you don't you're not successful with like the first 20 models Iran you're not successful even after you run like a grid search sometimes they mean there's a bunch of different things where you're gonna keep bailing keep tailing keep bailing and then you'll finally maybe get something that you're like content with all right I don't think I've ever like this is the greatest thing like the results are well beyond my imagination but you know it's the same with for example your experience with Brazilian Jujitsu right is you go to the gym if you're working with people who are more experienced or you know who are leveled up further than you you're basically just getting your ass kicked every day and like yeah that's part of the learning process though you never go home and you feel defeated you're like wow like you know these people that I'm working with are better than me nad like I got better because I worked with these challenging people or I got better because I worked on these really challenging problems so you're exactly right like I think the other night I was talking to my coach and I'm like I actually prefer to be told when like don't go wrong I'm human I like praise I like people going oh you're doing a good job and all that sort of but I actually prefer like to be told like when I'm doing something wrong or when it could be better especially if it's someone who's like God got more experience or something like that actually I actually like an older version of me would have and don't get me wrong like I'm not I'm human why I've still got like the ego and whatever it's like if someone tells you are you could be better or you're doing this wrong it's like okay yeah take that in take it on the chest but yeah if it's in terms of if it's someone especially if it's someone who who has your best interest in in mind like if they're offering like hey this is some feedback like a potential for you to be better it's like that's like a gift like that's like here you go well if you want to be resilient idiots is a great example I cuss I was literally talking to my coach the other night I was like where where am I going wrong where could I improve and so I think that's for for almost anything like if you're working on whatever project you share it out and you're gonna get like let's be real though no one's no one's like if they are like being like hateful or or saying oh this didn't work and all that sort of stuff well that that's like null and void that opinion doesn't matter whatsoever like because most of those times they're those sort of comments and whatever or either don't happen because people are generally like this is just my experience of putting a lot of things out online people are generally like nice and if if they do like coming across as that well then at the end of the day it's like well that what I understand how that person's feeling because I have also looked at other people's work and gone you know what I'm jealous of that because my skill level isn't there either so my first reaction is is well that that person must have had had this advantage or they've got I don't know whatever I'm trying to pick out the flaws in whatever they're putting out because realistically they're just the flaws in in my own skill or my own character and so they're that's what I'm quickest to see in in other people but I've I've since I'd like to think I'd like to believe that I've got better at that but I mean it's a it's a work in progress right but yeah with these things it's as if someone's pointing out it's like hey we're your as long as it's not like malice or whatever it's like well your work could improve here that's like who doesn't want an opportunity that's right yeah well you know I I have two quick laws I really want to give a lot of credit to all the people who I've done resume review and project reviews for they've taken criticism like incredibly well and like I mean I absolutely commend them for that not a single person yet has been like what are you talking about dude like like there's been no pushback which I'm okay with pushback I'm not always right but I really want to thank everyone who submitted for that as well you know what yeah you know it's kind of put you put yourself out there takes guts man so that's that's kudos kudos from from me to them as well well you know it kind of goes to something you're talking about it I am really impressed with people when they're not all that sure of their work yet and they still put it out there to be judged because you know that I mean that that takes some serious guts like you said I mean nothing not actually nothing impresses me more then then someone trying to fulfill their potential like that is the most beautiful thing in the world to me it's awesome so you know one thing in terms of filling potential a lot of people view that in terms of job success right I would love to hear about your your job experience you know how you ended up in your machine learning engineer role I think that that's like very interesting to a lot of people and I think your story but also really help them understand what they could do to put you know potentially land a job that they've been dreaming about yeah so a question I get asked a lot is what am I going to get asked in a in an interview and I don't really ever have a good answer because the only tech role I had I didn't really interview for and so I'll just I'll just tell the story of how it happened so I was learning I was studying my online AI master's degree like in this same bedroom and at the same time I was documenting a lot like posting a lot of just I would almost maybe once a week a video and then once a week an article just like hey here are the things that I've learned this week and here's what I'm working on etc etc and so I was doing that I would say I think it was I don't know the exact timeline again but I think I've used in the past nine months let's call it nine months I was nine months into studying my own curriculum full time and then a girl on LinkedIn sent me a message and it turns out I used to work with this girl because I was posting on LinkedIn right I was sharing while sharing all these things that I was doing I had a blog I had YouTube and look this is it's not for everyone but this is just what I did and here's what happened from it a girl from LinkedIn messaged me and said hey I've been seeing your you'll post on on machine learning I think that's like really cool and all that sort of stuff I think you should talk to Mike and I'm like okay I'll talk to Mike and I talked to Mike and it turns out Mike knew chem who worked at a company in Brisbane and they were working on machine learning problems and I'm like well here's what I'm working on this is me - Mike I was like well I've been doing this I've been a creator moan oh I must disagree because I didn't really go on and back didn't want to really go back to University I'm studying online I'm just making all these things I'd like to to somehow get into machine learning and healthcare and all that sort of stuff and it's like oh wow you should meet Kim ended up meeting Kim I told him the same story hey cam I've been studying my machine learning master's degree I want to get in machine learning healthcare I'm thinking about oh this is actually real I was like I was like at that point like nine to ten months in I'm like you know what I'm gonna buy a one-way to to the US and find a job at a start-up somewhere and and just see what happens because in my mind I was like we could have hung out man come on in my head I was like well I've a I and machine learnings like taking over the world so there's gonna be a bunch of startups in America's kind of where all the the technology and things happen so why don't I just go there and just work it out so that was literally like my tentative plan I hadn't bought the ticket yet but I was like that was what I was in my head I was set on doing and then cam was like well why do you have to go to the u.s. you could just do it here I'm like what do you mean and he's like well we work on those sort of problems do you want to come in on Thursday and this was like on a on a Monday that I'd met camp and so shout out to cam shoutout to my legends and I went in on the Thursday and it was kind of like a very informal internship I like met the machine learning team which is that at that time was two people so I was the third person and we just had a bunch of structured data that we were going through and so I spent the day just exploring it with pandas so they asked me at the end of the day I was like do you want to come back next Thursday and I was like yep and then I went back on the next Thursday did it's very similar thing just more data exploration with pandas not even I didn't even build a model yet and then I think it was the third or maybe the fourth Thursday let's say the third because I think it was only two and then the third one it was like about lunchtime and then I went out with the CEO and the head machine learning engineer at the time and they're like hey do you want to do on a roll and I was like sure I'll have a I'll have a roll and then I threw it the next year and a half working there on machine learning problems and a wide range of industries I actually started building models after that but there was a lot of data exploration and so my takeaway from this is how did I get a job I was first of all with anything of course she needs some sort of skill right so that's a foundational thing so it wasn't like I was just walking I was posting these things online of like doing what I was doing without actually doing them so I was really just documenting what I was doing which was studying all day because I had nothing else to do I didn't have a job I was driving I did I was driving uber on the weekends to pay for courses and just studying this this new machine learning thing as much as I could because I was basically obsessed with it and so whenever someone asks like like what should I do for a job or what should I do in the interviews I don't really have a great answer because I didn't actually I didn't actually have a formal interview it was literally like a sit-down conversation we had coffee do you want to come in on Thursday went in on Thursday then got offered a role but the first thing is of course builds skills and I think everyone kind of knows how to do that because it's like you can go through some of course you work on some projects yes the second one that I think really helped me out because if we look back at that scenario it was I got lucky I got lucky but someone saw my post and knew someone and then referred me to someone else but the the luck was kind of like pushed along by the fact that my work was was there and so if I made a little bit yeah so there's that saying you create your own luck and so whether I did whether I didn't like if you just it's a thought a fun thought experiment to think if I didn't put anything out there would I have got that message probably not I don't think so who knows but that's what I say they're like people it's like okay you're building skill second of all is to share your work and we've we've talked about this throughout this this little conversation here is that okay at the beginning it's might not be very good like but that's not very good compared to the people that you're comparing to if you compare yourself if I compared my my 9-month cell of when I first started learning machine learning it was like a world apart when I first started of course I didn't didn't really know much but after I'd been into it for for nine months it was like wow I'm leaps and bounds of where I was and so day to day is like yes okay progress is going to disappoint but over the long run like if you're if you're being consistent if you're putting in effort progress surprises so that's that's a fun little saying there too it's like in the short term progress disappoints but in the long term it surprises so the way you can kind of and I'm really bad at um so I've always had this thing as well he's like with job applications I pretend that if I apply through like a apply for this job button like on a careers page or whatever I pretend like it's an automatic no for me so with any job or whatever in the past that I've tried to go for I've always tried to have either an in or like I knew someone or and I've like gone above and beyond rather than just sending through a resume because I'm like this is just me I've always hated the concept of a resume I remember when I was first making one for like some first job in high school I'm like what the hell I could I can't put all like this crap about me on like one page and then someone looks it in like yeah this person but that's just that's just my take on it so I kind of had to if I go for like any role in the future not that I'm looking for a job now I probably won't apply directly through any court of any sort of job portal I'm gonna need to be doing everything I can to first of all start the job before I have it as in figure out what I'd the most ideal position I'd like to go for and if it doesn't exist I'd probably just start creating it that's kind of what that's my advice to anyone who are who's thinking about going a job just create your own but if you're actually going for a job somewhere at a company is figure out what it like figure out what what that person does day to day and if you're smart enough to do research to like how to learn data Sciences how to learn machine learning how to learn AI and all that sort of stuff how to learn to code you're kind of smart enough to do the same sort of research to figuring what you're going to be doing day-to-day and again it's sometimes it's just as simple as asking someone and then you set up this is a beautiful way to get experience before you even have a job is you go you take what someone's told you you set up your own little project and you go well I'm gonna do this for the next six weeks and work on this thing and then when someone asks do you have any experience you go well yeah actually talk to this person they said that they do this and so I designed my own project and worked on this for X amount of time it didn't quite work out how I'd want it to but here's all the code that I wrote here's like the analysis that I did after it and here's probably what I do next I mean to me that's if someone that says that that's not experience well then you probably don't want to work there so that's a very again the long-winded approach to you like how I got a job for me personally it was building skill sharing my work and then somehow the universe through through an opportunity my way and I hopped on board well you know my biggest thing is that if you didn't show your work like you said there's perfectly no probability of that happening at all I talked with the guys over at sharpest minds so they do like a data science mentorship program and they have all this data on the people that they mentor and through like LinkedIn or through some of these platforms the like interview acceptance rate for applications is literally it's I think it's between one and two percent so yeah like if you're applying you put a hundred jobs you'll get one or two Collins for an interview for like the average person it's absurdly low and that was average yeah if you know that stats the average yeah if it's on average four foot deep yeah but so you know people will get really disappointed they're like oh oh they got like two calls out of twenty back I'm like those are great that's great numbers man but at the same time almost every data scientists that I've talked to they did not get their job through a traditional route and I think that that's a really really important insight and it's just highlighted in your story that there are so many other ways to get noticed there's so many like different things and you brought up a bunch of really good ones of like you know actually just talking to these people and learning more doing even informational interviews like if you come in with a better understanding of the company than any other candidate even if your data science or machine learning abilities or like they're okay like you're gonna stand out so much because you've just done more homework I mean yeah and as why people don't treat the interview process like they would treat the data science problem you know I always say like okay you know people ask what skills should I learn to be desirable in a job market my advice is take 20 job postings that you're like really excited about and just aggregate what skills are looking for you'll have like a ranked list of which ones walk which skills and so just changing that thinking and looking at it like no it's not an optimization a problem where if I apply to 500 I'll get this and like it's not a funnel yeah yeah that's putting there yeah that's putting the cart before the horse that kind of way cuz you then you look I just sort of this like I'd rather instead of optimizing for the interview like I'd rather optimize for the the problems that I'm going to be working on cuz I kind of doing it that approach like it kind of just it felt it just felt backwards to me and so I and who knows maybe that's just because I just had no experience or whatever but even now that I've worked at a company and I've like hired hide people I would yeah any today like if I've seen like someone's worked on something like then that's that's like miles above like not having anything happy awesome so I one last questions I know you recently published a video which was you're 2020 machine learning roadmap I'd love to understand from your perspective what's your best advice for learning machine learning from scratch doesn't have to be long but just one or two things also people should definitely check out that video I watched it it is very in-depth so what would love would love any thoughts there if you want a feature film introduction to machine learning future film length video that's probably for you but yeah the kind of a the really really high level overview for me it's a look people are gonna you're gonna you're gonna if you especially if you just getting started like one of the first arguments you'll probably find yourself in is people deciding whether what programming language they should learn so that's probably the cuz look edited a if you want to get machine learning you're writing either math or code and I'm I'm biased for the code part so I like I like writing code if you want to get deep into the research you're gonna be writing a lot of math so you're gonna have to learn to code so this is from scratch if you don't know anything you're gonna have to learn to code you're gonna the first thing you're gonna face is okay I want to learn machine learning you go up there's a whole bunch of programming languages that I should learn or could learn and now I spent probably two weeks when I first started learning like in this eternal debate and I just ended up being a whole bunch of waste of time I should have just gone with the just just tried something out before because you're a lot smarter when you're doing things than you are thinking about things so if you want my very biased direction I would say learn a programming language it's probably gonna be Python of course you can learn R but this is this is just might a learn Python why because there's a lot a lot of resources out there especially when you get into the more advanced stuff it's basically all Python and then someone's gonna ask argue god oh why not Julia it's this new language is that okay where you can do that just be where that is because it's newer there's not as many resources out there so my approach learn Python very like approachable programming language I'm not gonna say easy because I hate it when people say things are easy to learn sometimes they're hard for people sometimes they're easy anyway its approachable so learn Python and then you're probably going to so this is when say each of these steps don't think of it as like something you can do like just over a weekend like this is like a three month per step so learn Python and then you probably going to learn how to work with data and so if you're using Python it's probably gonna be the main three libraries you're gonna work with pandas numpy and mad hot lib so pandas is to represent structured data the things that you see in an excel spreadsheet which is probably the vast majority despite what you see in terms of our what deep learning can do with unstructured data which is images video audio that sort of thing the vast majority of data you'll probably likely to work with this structured data so and even if you want to go into deep learning and things like image recognition video analysis etc it's still very very important to be able to just manipulate data so this is where pandas numpy matplotlib come in those three you could spend three months total on those and now again i've been learning this for almost three years and i'm still learning all of these things so what this is just what I'm I'm like giving like a let's say year-long sort of headfirst dive into machine learning - and at the same time as you're doing all this you want some concepts so here we've got four things we've got Python just the programming language itself we want machine learning concepts on the side so you just get a broad overview of like the general concepts of machine learning like what the different types of learning are what machine learning actually is then we want to be out of MIDI mp8 data with pandas matplotlib and numpy numpy is numerical python so it's very similar to Python but very a lot faster than Python at working with numbers which is basically what data is is just patterns and numbers and then we want to work with possibly the most robust machine learning library that's out there at the moment is so I can't learn so that is like the how do you say just you're probably going to use that if you're working on machine learning problems because a lot of actually here's a good thing a lot of the deep learning libraries are actually based off like cycle loan structure so we've got five things now - concepts pandas matplotlib numpy scikit-learn and if your vet brand-new I'm at me saying all of these things is probably gonna sound like gibberish to you but trust me they'll they'll become part of your workflow then after that the natural progression to that is probably getting into some sort of deep learning and for that you're going to become you're gonna find another debate is you're going to go oh should I learn tensorflow or should I learn PI torch which is again a waste of time debating that because you'll probably end up going to using both and these days they're much of a muchness there they do very similar things it's just different syntax and so while you're learning these things you've probably going to be like Daniel where should I learn them and you can do your research you're smart enough there is basically an unlimited amount of research that's out there I'm sorry of resources that are out there to learn all of these things all of them are basically the same thing just explain differently so find something that resonates with you and what's really helped me is working in small little sprints of of like learning things so it's like say this month I want to like two months ago I read this book and to end I don't get below I love that yeah so if you're a complete beginner I probably wouldn't recommend dumping straight into this like at this is like you you've never programmed before if you have some Python experience or then you could probably start reading this but if you've never like written any code before which I mean if you're watching this video it's probably unlikely you're probably already into data science and whatever so if you are watching this video get this book if you can if you can't get this book don't worry all the code is on github you just have to to go through it yourself which is see this is why I like the what you're often paying for in paid resources for learning is someone has done the the curation organization and put it yet is put it into like a structured manner because as I said you if you wanted to learn Python you could just read the Python dogs but there we have like the foundational tool tool belt that you're going to be using now to put your knowledge really into practice and this comes back to sharing your work is you want to at least every one to three months depending on what you're working on is work on at least some small scale project and whatever timeline you dedicate to that is up to you but I would usually recommend something that that's at least like a month long or if you're working maybe it's a month-long of weekends so it's like four weeks of like just going hey I'm working on this one thing and I'm gonna the first week I'm gonna do data collection the second week I'm gonna do data modelling this the third week I'm going to try and improve my model the fourth week I'm gonna wrap it all up and and and share what I found now again timeline is arbitrary how long you spend on each step is arbitrary but the most important thing is courses books etc etc lay the foundational knowledge you can't really start working a project if you don't have any idea of anything so you you do need the foundational knowledge but where the specific knowledge happens is when you start to work on your own things and can and I talked about this before projects something that might not work something that interests you something that makes you feel smart happy and useful and the fifth one there can be share it share your work awesome well you know that's that's so really all I had I think that this is super super informative not only for me but hopefully for everyone watching as well if you want to hear more from Daniel you can check out his youtube channel i'll link that above and below you can also check out his medium posts he has a lot of great videos a lot of great content so definitely pay attention and feel free to subscribe to his channel you closed a lot of really good stuff so thanks again for coming in and looking forward to doing more similar stuff in the future thank you Ken it's been a blast see you later

Channel: Ken Jee

Views: 18,885

Rating: 4.9396229 out of 5

Keywords: Data Science, Ken Jee, Machine Learning, data scientist, data science journey, self taught machine learning, self taught data science, self taught mle, machine learning engineer, mle, daniel bourke, data science projects, data science motivation, data science experiment, python, learning python, learning coding, interview, data science interview, machine learning secret, kjp, podcast, ken jee podcast, knn podcast, ken's nearest neighbors, ken's nearest neighbors podcast

Id: hO_YKK_0Qck

Length: 68min 32sec (4112 seconds)

Published: Wed Jul 15 2020