Data Science Interview Questions | Data Science Interview Questions and Answers with Tips

so today we have jaded they spawned a he is a senior data scientist at grameena and we will be talking about the best practices to crack a data science interview welcome Jay there Thanks prasad so to start with Jay Dave can you help learners or listeners understand data science in one minute think about the space that you are in that we are in this building objects so on and so forth how are you able to traverse and navigate through this universe of course like the entire universe happens to be in a 3d space except black holes and all but you're not likely to end up in one so the universe exists in a three-dimensional space you are able to navigate this universe and traverse through it because you understand the physics and the geometry of it any piece of data if it is structured also exists in its own space in its own multi-dimensional space where dimensions are the things that you are measuring or the number of things that you are measuring how many columns a table has and so on so data sense again is all about traversal of that space it's about how well you understand the geometry and the physics of the underlying data so that you are able to come up with certain insights on the data you can locate patterns you can locate clusters like for humans it's important to be able to learn to walk without bumping into things similarly for data scientists it's able to be a it's important to be able to traverse this space without distorting it and at the same time capturing things that are important so badly that's what data Sciences so yes it's fairly large data statistics it is geometry it is you need computers to do this so it's computer science as well to some extent it's a lot of engineering because there are so many moving parts and systems that you see nowadays a single standalone mathematical analysis is almost never good enough so it's all of those things they have now that we have defined what data science is and I'm sure you would have done a lot of hiring for your team of data scientist right what are the three things that you look for when you are hiring a data scientist speaking for myself the most important thing that I look for is clarity of thought more than anything else because for because of virtue of things like say the Moore's law you have a lot of data and you will have a lot of computing sooner or later if you throw enough computer data or enough data at a problem it will come up with something or the other that is not how we want things done at all we want the paramount of paramount importance is clarity of thought in terms of how you plan to attack a problem what is the problem it starts with defining the problem itself like I keep saying you want to make what is important measurable and not what is measurably important so that's one thing the second thing that I look for apart from the obvious data science skills is an engineering mindset by that I don't mean that you have to be an engineer that's just a word right you have to be a hacker you have to be a maker you should be able to tear things up and put them back together they won't be perfect but they'll work and you will have a better understanding of how they work other than that the third thing that sort of comes given the other two but I'll mention it nonetheless is an asset II it's very easy to give up and problems can get really frustrating so because like we were talking about the geometry of data earlier because you can't exactly see things in higher dimensions by the way and doing once said famously that if we could see things in higher dimensions you wouldn't need data science we wouldn't need machine learning so because you can't do that you never know whether you've actually hit the end of the road or not so getting a feel for that is also important which comes with experience which comes with so some really experienced data scientists can tell that there's nothing left to be exploited in a problem on the other hand they can also say that you know even if this looks at any difficult I know that there is more so there is that intuition and creativity that comes into the picture also yeah Jenny I would love to pick up you know something from the previous question which is background right and this gets asked to us a lot which is is there a certain background that gets preferred over other background when you are looking to hire a data scientist to be very frank yes I would be lying if I said that overall in the gel in general in industry we don't care about backgrounds that is not the case people do care about wagons speaking for myself I don't speak in form a company we don't there are a lot of very good people who I know who are data scientists who don't come from necessarily an engineering or even a scientific background lot of in fact Geoffrey Hinton is not really a he's he's I think a psychologist and a lot of great work is coming from the humanities like I've mentioned before so there isn't one specific background but there has to be a background and the most successful data scientists are the ones that have applied elementary data science skills to whatever it is that they are doing so I was reading this book recently called smarter faster better by Charles doing and in the chapter on innovation he says that innovation is nothing except being able to compile inspiration from different sources and putting it together and that comes that is something that really effective data scientists do all the time they are able to apply technologies that they have learnt in one particular domain to an entirely different domain and you know that is where money is made that is where problems are solved so because if it was straightforward anybody could have done it or you know simply considering applying a non-conventional solution to a difficult problem if it was so easy in fact we have something called Auto ml I know I might be digressing about but bear with me Auto ml is well relatively new machine learning paradigm which says just put in your data will reply will apply all sorts of algorithms on it automatically and you'll just get the best algorithm out now sometimes I think if that is the question why do you need data scientists anymore right but it doesn't stop there because machine learning is only one part and even Auto animal doesn't cover all possibilities the broader point I'm trying to make is that simply applying techniques is easy it's so easy that it's almost mechanical if I wanted to hire people who could simply use scikit-learn I could do that fairly easily but that's not what I want I want them to know why they are applying a certain algorithm on a certain date I said what do they learn from what do you get and how do they change their approach after that that's as critical as knowing what to apply today does that mean that anyone can do it and anyone can become a data scientist yes but more correct way of saying that is that a good data scientist can come from anywhere so this is also maybe a hint to recruiters the best data scientist that you're looking for may not be in the place you're looking for so yes anybody can become one at the same time good data scientists can come from anywhere so j-liv now let's come back to the original theme because the idea was to kind of create a mock interview wherein we asked the most frequently asked data science questions and see what's the best possible answer should be you know when you are when interviewing a data scientist there is never one best possible solutions but let's give it a shot so the first one that I would love to take up is the difference between univariate bivariate and multivariate statistics the easy answer to that is that the answer lies in the words univariate to leaves a unique variable bivariate means two variables multivariate means multiple variables where we use this is when you model data it usually has different fields which are considered variables because you don't fix a value of a univariate and something that deals with analysis of only one variable at a time my variable is two at a time because you might want to figure out how one changes as the other changes and multiple multivariate is multiple variables at a time for example let's say that you want to write a predictor for people's Heights height is one figure so that becomes you'll first what you'll start with is collecting people sites that's obvious the very first thing you should do since height in inches or centimeters whatever is a single variable its univariate now let's say that you are designing some sort of a BMI predictor now you need to consider weights as well and not only you need to consider weights you need to consider the relationship between height and weight as it's reasonable to assume that as height grows weight grows not necessarily the other way around right and of course even this has outliers there can be very that there are people who could be very tall but very light at the same time there can be people who are very heavy but very short so it's not there are outliers on both ends of the spectrum this is bivariate analysis you're talking about two different variables and their mutual impact on each other now let's say you throw in a few more this was just a BMI calculator but let's say that we want to make an app which assesses the risk of heart attack or heart disease in general you might throw in a few other things which are your genetic factors do your parents have do you have a history of heart disease in your family your gender your age what kind of lifestyle you have all of this becomes now multivariate because you are taking into account so many other things apart from what you are trying to predict so you want to you can't fix any one particular variable you have to change everything and see if impact on everything else so that's what multivariate analysis is Thank You Jerry I think that would be very helpful for the listeners what I would love to come to next is root cause analysis right and how would you define that because analysis is not something that is specific to data science it's a general engineering or even a business problem which again as the name suggests it's about identifying the root cause of a fault for a problem in a business setting you could frame it like why are we not making profit data can argument to this it can help you understand what it means or or identifying the root cause as such in deterministic systems deterministic systems are systems where everything is fixed and you can always predict what's going to happen with certainty data is almost always never deterministic because in data-driven processes are not for obvious reasons but let's say that you have a deterministic system like an internal combustion engine so in an engineering perspective if you're talking about why did a particular wall field then I should be able to narrow it down to a particular failure or a particular unit of the machine failing if I have am carefully sensing the performance of the other parts of the machine similarly in data related problems if I want to if the data represents a physical process or it could be a business process which is like fairly physical in a manner of speaking if the data correctly represents my measurements and my if they represent the underlying system properly then that can become a proxy for the failures and of different components of that process and things that stand out usually point to problems so for example if I have data related to let's say we are about to land or rover on the moon right let's say that I have data it's sending bad data about let's say it's sending bad data which is to do with with its acceleration right so again it any object accelerates along three axis right so let's say it every second it gives me three numbers which is its acceleration in meters per second square in a particular direction if I plot this on a graph and I look at certain parts which don't make sense to me then I might be able to conclude that okay this is probably where the rover fell into a ditch or this is where it got out this is where it got stuck where it was exerting itself really bad but acceleration was zero it didn't move or even if the acceleration is zero it probably kept moving because I can see the distance increasing something like that and God forbid if something happens to the rover then we'll be able to use this data to find out okay walk week what was the trajectory that caused this to happen to cause this caused the rover to come to a particular point and feel so that's how we'll cause analysis usually works it's just like it's a fairly standard process problem control systems business processes engineering systems all of these systems have a root cause analysis it's just that the data-driven perspective is that you don't actually have visibility into the system directly it's the data that is a proxy for the system and often you have an ideal model in mind that no matter what even if let's say my Rover has to go through the harshest of conditions its parameters need to be within certain bounds and I should see how close and if I see outliers then that's when I need to sound the alarm that it might be in danger so if that happens if you see outliers if you see anomalies those are the phenomena that help you zero in on the root cause of failures let's come to data preparation J there because that is also a very critical aspect of a data scientist work you know why is why is data preparation important in the work of a data scientist simply because computers don't understand anything except numbers for example if I want to make a chatbot I know for a fact that there is no compiler in the world that is going to understand text it doesn't understand emotion it doesn't understand centimeter let alone emotion in sentiment it doesn't understand words it doesn't understand meanings so I have to do something which converts these words into numbers and the numbers that are representing my words also have to be created or looked up in a certain way that reflects the relationship between words right so we do this why a method called word I'm adding spy though which is a kind of data preparation you have your raw data which is text but before putting it into a machine learning model I need to put convert it into numbers before putting it now it's not just conversion preparation when was a lot of other things like I have to make sure that like vocabulary is very large the English language contains quarter of a million words three-quarters of a million words easily and other languages contain even more so Sanskrit for example theoretically it's possible to have an infinite number of words in so many languages right you can just put together words and they become a new word compound words right so there is no way that I can design a system which can identify every word possible right so how do I prepare my data what I do is I come up with representations of words which are which can be put together in the same way that the words can be put together right so if I put two numbers together then they should represent the meaning of those two words put together correspondingly in the same way to the model so this is one example of data preparation this is like a fairly advanced example but there are some really routine things that you have to do cleaning is one part of it which we'll come to but it's basically things like what does a particular category or what do particular values in your table mean you might want to encode them in a specific way for example even if I'm talking about say gender coming back to our previous heart disease predictor let's say gender might be a factor and I can't say female and male I can't include my genders as female and mean so I have to convert them into binary true or false digits and then it doesn't become female or male it becomes male or non male or female or non female and if you're considering more genders then it doesn't even remain binary it becomes multinomial right you can have multiple genders which could may or may not have an impact on your likelihood to get a disease so this is also kind of interpretation it it can be something as simple as encoding categories into numbers to something as complex as converting words into vectors so that the entire vector actually represents a meaning of the word now that we are talking about data preparation J there let's come to data cleaning why is the attack leanin important again in the data scientists like data comes from various sources it could be coming again going back to the rover example there are sensors attached to the rover and it's sending back data through well space and then it gets received by some antenna which is on earth and then gets sent to some computer where it is stored right in the entire pipeline it's quite possible that there might be multiple sources where it's kind of where it gets affected with noise so think of data cleaning as removing the noise from what matters so that's a this is fairly common in telemetric applications but even in less physical or more businesslike applications where you have let's say credit card fraud data you might not want to consider fraud just because say a few transactions have been rejected they could have been rejected for multiple reasons authentication not done or you don't have enough balance that doesn't mean they're fraud so identifying what is noise and what is the specific signal that you're looking for is very data cleaning comes into the picture so let's go back to the rover example we have a bunch of sensors attached to the rover and they are sending signals back to space which are then received by cement in on earth through which they may be through a land network or a Wi-Fi wireless network they end up in some computer which stores it somewhere throughout this process there are multiple places where the signal that the rover is sending might get affected by noise so you might want to remove the noisy parts and only focus on the bugs because if you start incorporating noise even unknowingly in your model then there is a risk that the model will start modeling will start learning adapting to the noise which we don't want them right and this is a very specific application but even in more common rudimentary applications this happens a lot let's say think of just a spreadsheet that color that represents your payroll there might be a fake entry or there might be a erroneous entry right let's say that there was some service for which you didn't get billed properly or a bill came in late so before put let's say and then let's say that you want to make a model which predicts your monthly expenditure so you might want to remove these outliers from your data or which are clearly accidents outlier detection is a separate thing which is fairly important and that's not what we're talking about here but you know certain data to be erroneous and you do need to clean it out because again for the same reasons you drove out a model to adopt two things that are noise coming to machine learning Jaidev what is the difference between supervised learning and unsupervised learning let's deal with unsupervised learning first it's simply about inferring whatever possible from data where you have no specific objective in mind or whatever objective you do have is a general one you are not trying to optimize for a specific goal in mind for example it's like asking what's there in this data in general not necessarily okay is this a man or a woman or am I trying to predict the age for you know take a picture and predict the age it's there is no goal in mind specifically in other words there is no objective function or a loss in mind in general you have an objective which is fairly ubiquitous it's something that goes in deep into the data and tries to find out interrelations between the data itself clustering is one application supervised learning is basically more about decision making like and a computer how to do that with labeled instances with which is what we called training data so if I were to pick an analogy from real life you can teach someone how to ride a bicycle but you can't exactly tell them how to balance it right you can yeah sure give them some pointers but other than that there's nothing like if a cycle tips to one side what to do and what to the other side you don't say these things explicitly because there's there's what we'll use in there's not much you can say because it's just an inherent I guess the idea of balance or the whether you your body is balanced or not is something that is ingrained in humans much better than most of that animals which are not upright so for example I'm sure like a snake has no concept of balance or at least not as good as humans right so that is something I would call unsupervised learning because there is something the creature knows that is supposed to maximize something for optimize some particular parameters that doesn't know water so it just learns from its own experiences straight away on the other hand when you are learning by example let's say you're doing your homework and then you must have a concept because you've done of you some sort of things like that that is super slow thank you for that JD you know another very frequently asked question is can you explain decision tree algorithm so let's go back to that analogy of data being having its own space right it's an imaginary space but it's there nevertheless and how decision trees work is they split this space into boxes such that each box is more or less pure by pure I mean that each box contains data samples which belong to one mostly one category if it contains a mixture of coke at agrees the box is split again into two further boxes so you keep doing this until your entire space is split between unevenly sized boxes so that more or less each box contains samples from the same right then when you get a new sample in that same space you can just find out which box it is in and whatever happens to be the majority class in that box that becomes the predator class right so the you it's basically the way it translates into a computer problem is you have a bunch of if-else conditions on parameters of the data and it just it's a team so the very first split occurs at one value then that's not pure enough you do another split on another value and so on and so forth so you keep on splitting the tree it looks like a tree but geometrically again it's splitting the data space into multiple boxes and reaching each box each small box ultimately it's pure box that you get is the leaf of that and we transition from one box into another that becomes a mood of the tree where the tree is split into the recommender system JD you know what is recommender system and how is it used in data science very interesting story there was a movie called into the while or some such thing and it was based the movie was well fairly famous but it was based on a book which wasn't that well read not that often read so people started buying the book right because of that what happened is that a lot of other books which were about the same zone or which was wilderness exploring and survival those books started a lot of books which had never been which hadn't even been print in print for a long time Amazon still started recommending them because they were in the listing they might not have been in stock but they were there and a couple of such books which had been out of publication which their authors had died their estate wasn't there their children had died entirely out right but people found those recommendations so interesting that there are people who went down and drag those books and brought them back in so they are now in printing I regret I do not remember the names of those books but the point here is that there is something called the long tail phenomenon so which is basically says that majority of products pain in a minority of stores like it's a very small bandwidth that contains most of your data there is an infinitely long tail beyond that which you will never see right for example every time there's a sale on Amazon or Flipkart or something I keep saying the same thing keep seeing the same things again and again it doesn't belamis on is still much better at it than most of the people but the discovery of new books or new products is something that is not very easy to do and that is the long tail because there's an infinitely long tail of products which are extremely difficult to find so because of certain things like into the wild some item out of the long tail comes out and you it's like finding a new treasure right so that's what recommendation systems do they balance two things relevance and discovery they want to recommend products to you that are relevant to you at the same time giving you some sort of a sense of adventure maybe that you are finding out something that nobody knows whether you will actually like it or not Amazon definitely doesn't know or a recommendation system doesn't know that there is a movie or a book or an album that you may or may not like but you might still discover something if you listen to it if you end up liking it all the better why not so I recommend the system is something that balances the long tail out it tries to flatten it out somehow of course it has to cater to the market right but it also plays a big role in discovery of new items and recommendations are important clearly you don't want to I think it's it's fairly routine consumer experience nowadays if there was a single platform where I didn't see recommendations or think that something is wrong with it rightly it could be a simple Google App Store it could be your iPhone store it could be Amazon whatever me into deep learning you know what is the difference between deep learning other types of learning and where does neural network fit in in all of this very honest there's not much of a difference it's like asking when does data become big data actually for Hut had an interesting slide about the second of his talks big data is whatever makes Excel crash right so it's just like that deep learning is essentially neural networks with an unusually large number of layers in it neural network is essentially a network of a layer of neurons followed by another layer followed by another layer and when this becomes too large for a single machine or a CPU to handle we usually run it on a graphics card on a GPU or on multiple machines with graphics cards that's when it becomes deep learning theoretically it is not drastically different from other types of learning so it is it is a specific algorithms I mean it's at least not different from neural networks per se it's just a special case of neural networks where you simply increase the number of neurons in it and even that neural networks themselves are a combination of fairly routine mathematical operations just put together in a different way for text analytics jday what would you use Python or R personally I would use Python no reason that you shouldn't be able to use our by the way definitely user I would use Python because R is a domain-specific language it's meant to do a finite set of things of course Python is technically also meant to no finite set of things but Python is a general-purpose programming language you can everything that is required to build a complete end-to-end system can be achieved within Python so if I'm working on a purely text application getting text from either a scraper or from the web or from my local file system reading it cleaning it modeling it making predictions based on it the whole thing I can do an item if I were to do it an hour it's only the modeling in the prediction side that I would be able to do it an hour to some extent I'm sure our has gotten better and there are ways to cover up most of this pipeline in R as well it's just that Python does it way more natively because it is a general-purpose programming language so text is something that is native to a lot of programming languages like if you talk about C or Java strings are primary data structures in them so you don't have to model tree text as like a second-class citizen in your program so when given a problem Jaidev how do you decide which algorithm to choose and when you apply that algorithm it's suppose for example it does not work how do you decide step 2 step 2 also does not work how do you go to step 3 and then finally how do you arrive whether to continue with this problem or not ideally choosing and prioritizing algorithms comes like leanette or funnel first is clearly identifying the problem properly it has to be framed as a business problem clearly because that's what you're there for then you identify whether that business problem is just a modeling or a visualization problem but in this case it's clear from the question that it's a machine learning problem whether it's a supervised problem or an unsuppressed problem you have train data do you have label data or if you don't have then can use and read it somehow could you do it simply by asking a few people or maybe finally bringing a little bit of it if not is it an unsupervised problem if it is an unsupervised problem does it really correspond to the business problem that you have framed around it is it a classification problem or a regression problem that is also important at when you know these things that this is when the modeling starts happening I usually would start with simple models first by simple models I mean models that are easily implementable and that are not really super smart they do a benchmarking of the data why that is important is because if I am able to get some sort of performance metric on a simple model then I know that that is my base whatever else I try which is more complicated has to do better than this and if it's not then it's not go on the travel let's say that let's see will simplest logistic sorry simplest classification algorithm is k-nearest neighbors or even let's say a bench market with something even simpler like logistic regression for example if I end up spending months and months on that problem and I come up with a really fancy really expensive deep learning solution to that which cannot beat my logistic regression that's a pure and simple waste of time in them right just go back to whatever was working list so then what happens is that okay if I am not getting good enough results with even a simple model I increase the complexity of that one that leaves me to bet better models like if I increase the complexity of say a linear model I'll probably end up with something like support vector machines or with decision trees or with random forests or with neural networks that's how the progression goes at each time I keep on evaluating my effort accordingly at no point should I be doing unnecessarily complicated stuff just for a very meager improvement on my matrix so the one constant trade-off that data scientists often have to look at is is the extra complexity in your model worth the money that it's ultimately saving why this matters is because simpler models are very easily implementable if I want to put my model in an Android app I don't want it to be in your network if it's only going to give me point zero zero one percent of the profit right I'd much rather have it a linear model which essentially just evaluates our dot product of my peaches at the same time like let's say again coming back to the Rover example all it needs to do is say navigate ditches or craters on the moon I'm sure that particular problems does require a little bit of computer vision but if it is solvable with a simpler problem we should that's what we should prioritize because you can't put a GPU in right so that's how it is and again like there are there's always there has to be a budget to this in terms of time and effort it's quite possible that you'll never reach 100% accuracy so you need to plan out and advance that this is a so much percent accuracy is when I'm going to break even in terms of member dismissals problem I'll continue after that if I have time and money or I will stop because I've broken even on the other hand you could not even break even and actually that whole thing might be having Stern effect so it's a good idea to also have a set of rules which will help you decide when to give up not not in the negative sense but just you know in order to maintain your sanity and maybe try a different approach altogether so maybe tie a new data set continuing with the theme of problems are they've how many tennis balls can you fit into an aeroplane interesting so I think one of the things you should accept up front is that it's really difficult to come up with an answer to this you don't need to say this out to the interview out loud just accept that you will probably not get the right answer unless you've actually done some really intensive research about it but there is a very interesting course which happens at MIT it's called the art of approximation so I recommend that to all data scientists because it's beautiful course and it's very approachable it you you don't need programming to do that course these need a pen and paper and you can do most of the exercises so let's approximate what would be the diameter of a tennis ball five centimeters let's say ooh that much maybe somewhere between four to five centimeters now it's reasonable to assume that if I lay out a bunch of balls on this table in a rectangular grid that would occupy more space than if I let it lead them out in a hexagonal grid right so let's say that on the floor of the aeroplane I'm laying out a hexagonal grid of tennis balls right now the shape of the cabin itself is not regular it is like a cylinder but not exactly a cylinder it's something like an ellipsoid a 3d helix but fine let's for the sake of approximation again assume it to be a cuboid which has parallel faces everywhere right what would be the of this cuboid what's the tallest person you've seen standing in another plane seven feet high so seven feet there are 30 seats in an aeroplane 30 rows for the sake of convenience ignore the cabins and toilets 30 rows how much would be the leg space easily two two-and-a-half feets feet at least for say two and a half multiplied by 30 gives us 75 yeah 60 plus 15 that sounds that right 75 feet across sounds a little less but anyway let's park that figure aside for now then what would be the width of it you have three people sitting on one row three in the other and an aisle in between usually this is how I sit this would be how much at best one-and-a-half feet not more than that three of those would be four and a half feet so four and a half feet in one row four and half in the other that's 9 feet and let's leave another two feet for the aisle or is that too less fine two feet so 9 into 11 feet by 75 feet by 7 feet right so what I have to do is just figure out how much how many tennis balls I can fit into one cubic feet and then multiply that by the volume of the plane there's my answer you know chair there since you have done a lot of interviews as a data scientist yourself do you and also companies expect aspiring data scientists to know all the tools no we definitely don't expect people to know all the tools it's impossible practically we don't even expect them to know too many tools at once what we do expect is that they ought to be able to pick up a tool fairly foolish without taking too much time so in a lot of ways we also test the ability to learn quickly yeah so because again this is something I tell my mentees all the time is that tools are a distraction it is extremely easy to get distracted by technology they keep telling me that they forget the function for something or they are forgetting the command to start to soup it or notebook or something and I said good please feel free to forget that is not what I meant brain is meant to do that's exactly what Stack Overflow and Google are for if you want to do something just google it over time with practice these things will happen naturally and that that isn't even your that's not what it's supposed to be occupying your RAM right that should be muscle memory more all of us just like you don't exactly remember where each key on a keypad is right you just you can do touch typing right without looking at the keyboard all the time so it's just like that I can mostly write code necessarily without looking up too much documentation because I've been doing it for a while so don't try to circumvent that by memorizing things at all similarly the same applies to knowing technologies and tools it's very it's almost pointless to learn too many technologies one at once because they are all doing the same thing and because they're doing the same thing the overall value proposition is the same for any - of course some they have that all of them have a specific place in an application Garrus may be better than something called lasagna in certain ways tensorflow may be better than paya Tosh in some cases but that's not what matters most of the time that matters when you're optimizing and in general in computer science one very strictly followed or at least it should be strictly followed the rule is that premature optimization is the root of all evil so we do not optimize things prematurely which means that you shouldn't need to over engineer your solution to begin with which means that you shouldn't have to try out multiple tools on the same problem start with something you keep on iterating until it reaches perfection and if you're a good learner you should be able to pick any tool anyway right having said that when I say that learning concepts is important more important than learning tools that by that I don't mean that ignore tools completely as well in most cases especially things where you learn by doing it's much easier to understand how things work and then understand why they work so for example if you are just starting out with neural networks or deep learning figure out just how to make a deep learning model why it works and what's inside it can wait so that's also there right so there needs to be a balance by the way the faster TI courses work a lot in a similar fashion they will show you how to do stuff get started get things done have a viable product Minimum Viable Product and then go into the details and optimize it so there is there has to be a balance between both approaches of course the balance is far more lopsided on the side of concepts rather than boots Jaidev since the theme of this conversation was interview we would love to end with an advice you know an advice to people who might be appearing for data science interview you know what what advice would you have for them tell your story what I mean by that is if you have a portfolio if you have had a bunch of things that you have done especially in data science the journey matters a lot more than the outcome so like I said we need to see your approach so there was this problem where did it come from why did you have to solve it what did you do in order to finally solve it if you failed what could you have done differently if you had the benefit of going back in time so it all I guess it's encapsulated into that one statement when I say that tell your story properly and faithfully write things only things that matter only things that you are clear about write only about your own contributions and be clear about you know like be clear about your expectations and in general make it a short C don't make it 13 pages long Thank You Jay Dave it was a pleasure having you I think I think you know just mocking the interview very yeah would be would be very helpful but he's feeling a lot more studious than I usually do so thank you for your [Music]

Channel: Springboard India

Views: 87,330

Keywords: Data Science, Data Scientist, Data Science Interview Questiosn, Data Scientist interview questions, data science course, Data Science Training, Gramener, Online Learning, Data scientist jobs, Data Scientist Salary, Salary, Jobs, Interview, resume

Id: 7YuTmLvs1Dc

Length: 44min 15sec (2655 seconds)

Published: Tue Sep 24 2019