Feb 8- Live Virtual Mock Interview For Data Science Role

Key Takeaways

like yeah so i think we can start so sorry if you can talk about yourself again little bit and yeah chris can we have his resume uh i've already dropped your mail so i'm sure i'm just making it live in youtube okay so we are in that stage and i've already sent you the resume you can have a look um and now we are live i'm just dropping you the url the chat yeah we are live we'll wait for some time so that yeah so we will wait for another two minutes so that everybody comes and then we can probably start i hope you got the resume um okay so many people have started joining uh so hello welcome everyone welcome sudhanshu once again uh uh welcome saurabh uh again uh for this live virtual market review uh for data science role and uh what we are going to do is that again as usual uh probably this entry will be for 45 minutes uh the way that we will proceed probably will go which will start with python programming language some basic things in that and then we'll move towards statistics machine learning and then we'll probably follow your resume what all the things we have and to talk about the resume we already have your resume me and sudan and based on that this entire interview will be going on so yes you can start with the introduction sort of and then i had done my bachelor of technology degree from united college of engineering and research nani pragaraj in which is affiliated to dr fpj abdul kalam technical university like now in a specialization of computer science and engineering i have done my schooling from marie lucas school and college and my interest is from the last two months is in machine learning with python and my hobbies are teaching and searching and exploring the machine learning project and till now i have limited knowledge about the machine learning but how to deal with the real world problem and how can i automate that problem so that the person cannot do that starts manually so i want to think about that and how can i approach to that problem through the skills which i have gained till now that's awesome okay uh so fine sorrel so here i think in your resume so you have mentioned like a machine learning things and python you have mentioned and you have mentioned that uh so summer training at python developer time technology duration two month and many more things so again you have participated in a workshop based on python programming at ecc lava that understood a second okay and uh yeah so i can see each and everything in your resume and okay so since how long you have started learning machine learning i think you mentioned two months right yes sir from the 1st of december i have joined the course yesterday which course by the way applied air course okay fine so you have joined a course uh there and like okay so in machine learning so what in all portion uh that you know said till i completed the supervised learning supervised part the last thing i have learned the ensemble technique okay okay fine so the last thing was in symbol technique and starting with the basics of pythons required for the data science basically stats and the basics machine learning algorithm and the maps behind the each machine learning algorithm optimization problem and all the thing till the supervised learning computer but automation right so how and what kind of a work that you have done over there so in terms of automating things so like i have did one case study on the question uh quora question pair similarity because in the quran question database sir because there are hundreds of billions of users in the quora and each and every user asks question daily so there is a chance that the new person asks the question maybe that question is already present in the quora database so the connected user used to answer the question so how i can i automate this process that is the if the question is already present in the database so how can i map that that question which is asked by the user to the uh to the person who has already answered to that question because if we treat this question as a new question so the connected user answer to the question one more time which is i don't think that it isn't good because that question which answer is already present in the question that's why that's why i'm thinking that i want to automate this process like if the question which is already present in that database or database i want to map that question with the answer which has previously answered by the sum of the connected users how you have mapped those things i've said i have did only the similarities like if the question present in the database i have checked that is that question present in the database or not that portion i have it that [Music] so based on what algorithm so what is the algorithm what is that technique that you are using for similarity check so for the samsung request first app did the because in this second [Music] data set there are four columns question id one question id2 question one and question two so first i applied the preprocessing tags and all the tasks then i applied the machine learning algorithm like you know first ever first i converted that text into vectors like first i've converted it to the two parts like first i converted into the bag of words and the second with the response coding like and what is what said response coding okay so and i think it is a part of your curriculum itself right yes sir okay so what you have done from your side so it is not the part they have to provided me that the link that you have to this date this type of word you have this type of they provide the object means that if you they give you the headline that you have to do in this tag this step we have to write the code itself as an assignment in the attack so i have did that assignment parts okay so what is the similarity score that you were able to get what is the like a performance of the algorithm that you were able to get sir i have used the log loss for this for this problem so for the first time check the first object as a random model because to because in hog glass there is a no benchmark like no base because like as a random model i have got the log loss of 2.53 so as a random model if i am getting that the log loss is 2.53 so basically if we apply the machine learning algorithm so that log loss should be less than the 2.53 so first i have applied the univariate analysis so first i have considered the only the question both question pair how it is affected in predicting the question pair one question to that's the process i have did in that house okay and how you are thinking of productionizing this model so let's suppose if i'm going to productionize this model right yes so what will be your approach on that sir i will use the so first i will use the question quora database and i will send the question as you check the the that question is already present in the quora database so here we apply the indexing technique like because in the quora database there are hundreds of millions of questions so we cannot compute similarity with each question so we apply a indexing technique like which is used as a searching algorithm in the google so we applied that technique so we get the subset of the question from the quora database that this question may be similar to that question so then we apply to questions but my question over here is again if you are trying to take a subset so how you are trying to decide that particular subset that okay fine so this is the subset which i am going to choose as you have mentioned millions of questions so how are you deciding that because because we like the regular we use some like the like the substring is present in that [Applause] in the quora database like if i ask a question who is the president of the india so if that question is already similar question like who is the president of india what is the name of the president of india means the searching question text is similar means the question takes like what is the means on the basis of the text we sub take the subset from the question from the database again so intent how you are going to define c what is and all those things that's fine right so but intent of the questions will be different all the questions will be different so how you are going to decide that this is the intent of the question and based on that you have to create a subset this is my question sir on the like we do the question or text searching like we first we match the what is let's suppose if i'm talking about text searching right so how it is to be how it is going to help me out so maybe do i need a machine learning over there to do that for text search so i think we can write the python code and then with the help of that or code in any language right so we can write and then we can try and do a uh search and i think people have been doing these things since age ages right since a decade so people are doing these things so how your solution is unique this is what i'm trying to understand so how your solution is unique when you are saying that you are trying to do a search based on the string which is available in my question so uh i can't see like i'm not able to visualize like a machine learning approach over here so what is so excited said we can use the concept of what to wreck also i like if we because they preserve the semantic meaning also so if i pass that what is the name what is the name of the president of india and if that semantic meaning present in there because if we pass the word word to work whatever for each vector and in the question and we check the semantic meaning of that so in that way i think that word work will be helpful is that going to work for you for millions of data sets sir it can extract sample of the database because we do see my semantic searching and on the basis of semantic sensing we extract that question that because if we pass the each question it will act as a data point and if you compute that similarity with that questions and if they if we find that they are both similar on the basis of word to back then i can extract the question because what do i preserve the similarity similarity check is fine right similarity thing we can do but like the things that you are trying to propose is not not an optimized one right so we have been doing a semantic search semantic check right since a decade i'll say it's not a new concept at all right it's a very old concept i'll say so based on the semantics so we'll try to find out what is the similarity between two entities that is fine but the thing over here is so how i will be able to get a context and what is a best possible approach according to you what is the best possible approach that i will be able to apply on that one so till now i am able to recognize this only technologies okay any other project that you can talk about sir sir i have worked on the personal cancer cancer technologies this is a part of this of my course as an assignment i have did that part in which there is a in which i have to in which there are four columns like gene and the variation and the text what is the literature behind gene and variation and text and we have to predict that out of these three variable variables what type of level of cancer it is there are nine level of cancer on the basis of gene variation and the text so the real world problem here is i don't want to read the text by manually by the doctor or in uh to other domain experts so we want to automate this process by the machine learning to read the text because the text is very very much large in quantity so we want to automate this process on that basis of text we want to predict that what level of cancer the patients have so what kind of vertex can you please tell me the example so what kind of a text syntax is the like there is a gene and the mutation or variation like in the gene theory if there is a variation so on the basis of gene and variation all the literature it is written about the genes and the variation but if that gene variates to that level then this then the cancer is of one one level of cancer if that gene variates from because all the literature which is returned in there there are some words like keywords so on that basis you decided which level of cancer that the patients have okay so uh do you think that the data set that you have uh like as close to a real time yes sir i think do you know how chemotherapy happens how do we record all those like uh like a relations from a chemotherapist and then based on that so we try to diagnose something inside a human body no idea sir i think that the cascading on trouble technique i think that the hair applied simple technique so i'm just talking about your data as you said that it's a real time right so i'm trying to understand that whether the data set that you are using and you are trying to say that it's near to real time right so do you have any idea like how this data set has been collected and then how like actually we diagnose any patient as a cancer patient and then say this is a kaggle problem so this is the database offer right it's not a real time it's open source dummy data set you're talking about yeah yes yes yeah it's not a real time this is not a way how it's been done uh someone has open sourced our data that doesn't mean that your data is a real time first of all we don't do actually and this is not an exact approach at all i would say right so we have a budget we are working on that day and night and this is not a way at all uh like uh we uh take it ahead any kind of a healthcare project okay fine coming to your like uh python part yeah so can you please like share this video and uh show just just open up any id any id yes go with google collab suggest anything anything is fine yeah any id is fine i don't have an issue yes okay so chris you have something in your mind to ask in terms of python yeah sure uh just tell us the difference uh have you heard of yield keyword yield keyword closer do you know iterator yes sir what is iterator just to show me an example of iterator sir iterator like there is a list okay [Music] just just write it down just write it down just zoom in your screen little bit okay what what this is called as my iterator you're saying yes because the list of because list is a sequence or and we can iterate through then what is what is it yes at least because we can uh traverse through the list from the start to the end so it becomes an eye treble it does not become iterator right so do you know the concept of eye trace what is the basic difference between iterable and iterator in iterator you are basically traversing through the valley so because of that you're using for loop right over here yes how do you create an iterator now so by using the for loop sir no no no how do you create an iterator which function do you use inbuilt function item have you heard of fighter yes sir i've heard about twitter tools yeah eater just just write eater i t e r i t e r yeah and i insert the inside this brackets use brackets and probably if you supply this list no no no open open and close a bracket no not like this see it is a function right heater is a function right so yeah open and close the bracket open and close the bracket have you heard of this this key this function anywhere no sir okay have you heard of generators at least no but i am not able to recognize that you have mentioned that you have completed python and then you have like a stand stood second in 2018 python programming in a workshop right yes fine not an issue so uh can you please like uh show me how to iterate through an integer so how to iterate through an integer yeah no so range will give you like a range will give you a table object right so not by using a range so is it possible to iterate through an integer integer like 123 244 like this you have an integer value okay yes sir how do you iterate through that just just to check your basics programming skills so first i will convert them into any string then i can use the you know without connecting into a string so let's suppose if i have to iterate over to my integer so is it possible so by using the divide and modulo i can no no so just let's suppose like one two three i is equal to one through three like uh let's suppose you've created some variable i right yes sir is equal to one two three right it's just i is equal to one to three no not nodding list so just normal integer one one two three no one twenty three one twenty three one two yeah is it possible to like uh as an integer so is it possible to extract one two and three out of this one so by coding i can extract so just just from like an integer itself so you have you don't have to like convert these things into any other data type so just through our integer so i'm looking for one two three separately is it possible no sir because every integer is not any table so i can't so how you will be able to check that so i'm looking for a proof right so like integer is not uh i'll say like a iterable object so you have to give me a proof so can you please give me a proof so but if i use a uh like if i use it later for i in for j in i then it will show an error and object is not a treble okay okay that's fine so just try to like uh do one thing yes just try to like uh take one string right and without using a for loop so i'm looking for each and every element from a string so is it possible now just try to make i is equal to a string so just try to keep these things in double quotes double quotes single quote whatever you like same same one two three inside out same yeah one two three now without using a for loop so i'm looking for one two and three separately i'm looking for one to three separately without using a for loop you can take help from google i don't have an issue okay alright so we can use the membership operator uh sorry which one a membership operator like in i i just put one condition over there that you don't have to use for loop anywhere so apart from for loop you can use whatever you like simple but no for loop so we can use means we cannot use loop any loop any loop yeah any loop you are not supposed to use have you heard of indexing yes sir no but still like uh indexing uh okay okay so okay fine uh next thing wait ready so then okay uh just try to create a csv file you have a csv file anywhere in your local system yes sir just try to read a csv file without using pandas without using pointers yeah it's okay you can google not a problem and yeah like uh don't don't get nervous it's completely fine yeah that's okay if you cannot remember it's so good to google you know do well says take your time so uh like uh we are not but don't don't check out any other website check out stack overflow python documentation you can check out those things okay one suggestion will be that change your default page make it to google you know you'll get better results so yeah so that should continue i think the code is fine but he just has to change the csv name will work fine yeah okay so okay fine uh sort of so just try to like uh write one class right so just try to like create one classes and inside that so probably you can try to create a two separate uh method with the same name yeah this is this is fine yeah this is only fine so like i just try to create uh like a class and uh inside that class so just try to create a two separate method with the same name so sort of just zoom in more your screen little bit so that you know people no not that much yeah this is sufficient yeah it is sufficient sorry so what was the question yeah so like a question is like just try to create two methods so like he has created two method right inside one method so just try to return maybe like a a plus uh b and inside other methods just try to return a congratulation b so let's suppose in a first add method that you have written right so let's suppose just try to return a plus b into a first method and uh into the second one so just try to uh like again maybe you can do the multiplication of any anything so that is fine so can i say create a constructor so fine if you want you can do that if you don't like so inside add it's fine without any constructor so it's fine there is none okay so now okay so now you know first add method right so let's try to like uh add z plus x so just try to return z plus x subject the six yes z plus x just try to return self dot y is equal to thanks okay now so return return these things the multiply and in the second one so maybe you can try to out for z into x okay that's fine yes so now if i have just return so i think you have not written return statement so fine um now so here you have to call up like a both the method and what i'm looking for is so in one place it is supposed to return me a addition and in other places it is supposed to return me a multiplication so can you do that yes you um yes um this so now you have done many mistakes while creating your classes and its method so just try to figure it out many mistakes you have done actually many basic mistakes you have done it's okay take your time sort of not a problem go slow you go slow it's okay first of all be calm right take your time so a simple thing is take your time 10 minutes 15 minutes 20 minutes whatever it needs take it and then try to like uh write a code it's completely fine okay you want to use internet it's okay use it if you have any confusion somewhere if you are stuck with some kind of syntax things just use whichever makes you comfortable use that anyhow when you're working in the real world company you will get an opportunity to see the internet in interviews very less probably since you're a fresher you know they'll actually see your coding skills also uh oh so okay so so okay i think so uh you can proceed with other questions i guess too much time is taken in this okay fine so it's fine so like uh uh we can we can stop sharing your screen so i think we can talk about machine learning uh knowledge yeah so fine it's okay thank you yeah you can stop sharing skin okay so uh fine sorry so just like uh tell me like uh which visual you you're comfortable with which algorithm so which in all like a stack of the algorithm you are like us you could think that okay fine so you are very comfortable so like recently i will study it on sambal technique so i think i think that sir all the algorithm like like random forest uh like used for that because basically there are four ensemble techniques which i know okay yes sir so like bagging i know bagging and boosting cascading and stacking okay so bagging boosting cascading and a stacking one okay so uh can you please talk about xg boost algorithm and mathematics behind that because as you have mentioned that you have gone through algorithm and its mathematics in depth so can you please talk about a mathematics behind exe boost yes sir like uh in the xc boost algorithm like so first of all we know what is boosting algorithm so like in the boosting algorithm we want to reduce the bias like if the data set is there so the data set should have the low bias and low variance so in the boosting algorithm if i know that a data set has low variance and if if it has high bias so we can use the concept of boosting so so instead so to proceed with the like so can i explain this with examples yeah yeah that's that's fine uh first we create a random model and a regression model like and if you create a if we pass the data set suppose that we pass the data point and the actual output is 28 but it is that it has predicted like 23 so the error would be five to five so in the next stage uh we don't use the y data like a response level instead of that we use the error so we compute the error and then if compute the error instead of taking the error we know that the error or error is is same as the loss so instead of taking error we compute the loss function and under loss function i will compute its derivative so instead of derivative instead of taking in the next step as the error we take the x i the data part which i have taken in the first step and but in that step we take the minus of the uh gradient of the error so that is i think that as i proceed with that till that till that till i reach the optimal zero fifth okay so uh like uh it was little bit difficult for me to understand your explanation so i'm just trying to like uh recollect uh all of those things so i can explain once more yeah yeah fine go ahead you can you can take a second try i don't have an issue like suppose there is a regression model so regression model has a let's suppose there is a data point x i which has which is d dimensional and there is a y which is a real number okay yes sir so if i first i use the random order and predicted that its output is 23 but in the actual i know that it is 28 so there is a loss of error is 5 so instead so there is a proof that the loss is similar to the minus of gradient okay it is proven in the mathematics that if the loss is similar to the so here here i am taking the regression problem so the law minus in gradient yes minus ingredient is same as the loss okay so in in the second step what we do is instead of taking the error we are taking a data point so has the model uh which started in the first step but in this class label we take the instead of taking error we take the gradient of that error function okay so why we are taking gradient except because we take the current because it has been fast convergence to the model because if we because in the in the less number of steps we compute to the error okay so we we all used to talk about gradients right so can you please explain me let's suppose i'm not a person who understands this gradient part right so can you explain the gradients in a better way so everyone says that like we will try to do a gradient descent gradient ascent and all those things right we'll try to find out the derivative of errors and then based on that we'll try to change some values some like factors in between equation so what is a actual meaning of a gradient so what we are actually trying to do in gradient so so let's as we all know that in the regression problem there is a squared loss mean the squared loss is a i think uh is a parabola in a parabola shape like this so what we want to achieve in the regression problem is that we want to get to the momentum of that point [Music] okay so you are saying that fine so i will try to get into a minima but just a minima so let's suppose like uh first derivative i'm trying to do and then i have received some value so how i will be able to uh know that this is a minima which i have to achieve or maybe there is some other minima which is also available said as there is a concept in maths the concept of minima if i dedicate that point if i derive the loss function and equate is equal to n equal to zero then this point will be the either the maxima or it will be the minimum part okay is it yes sir either maxima or minimal product yes sir so if i equate to zero then i'll try to listen up this question clearly right you are saying that that i will try to achieve a minimum right yes you are saying that you will try to achieve a minima through a gradient right yes sir so my question over here is that fine let's suppose i received some value okay yes and then based on that so i'm trying to like achieve a minima so let's suppose i i received some value right so i got like a five in terms of value so five so maybe that can be a minima for me or maybe there is a possibility that i will be able to achieve a value lesser than 5 right possible so how i will be able to justify or how i will get to know that whatever value which i am able to receive is actually a minimum that is my on the left hand side can you please like uh share your screen and then write it down uh without a mouse or something and then try to explain me because it's a little bit like uh difficult for me to understand in this way sure so where is so like this is a lot the regression loss function okay and we want to reach to this to this place suppose we have randomly taken that the weight should be here just try to write down your like a loss function so just try to write your loss function as you said a mean squared loss you are going to use right just write a loss function and then with the help of loss function can you please show me that how you will be able to like get a derivative and then how you are going to consider something as a minimum or maybe a maxima uh that is fine so i can use the as gradient descent algorithm like suppose yeah you're fine fine you can write your like a loss function over here and with respect to loss just try to show me yes minimum maximum so anything or maybe a local medium so first i compute this left hand side of that point because if if i equate equal to it if i differentiate it and get a value suppose this value is zero take a equation try to differentiate it so let's suppose there is x square x square and if i equate if i differentiate this it will equal to 2x so if it is 2x [Music] take a loss function and then do it i know the differential of x squared that is fine so i'm not looking at let's suppose this is y i just do one thing no so you can you can try to use maybe notebook or something so in that you can type so no need to uh like uh write it down everything over here so we can take a notebook and then write it down there maybe open up notebook notepad plus plus yes so my notepad is visible oh no i think you will have to reshare your screen again so questions yeah it's visible yeah so maybe y square is equal to y2 you can write so i will be able to understand if you have written y square in this way you can write yeah so let's suppose sub predicted value w transpose dot x i is my predicted value and my actual value will be y i so here i will use y i minus and how this predicted value will be w transpose of x i so because when we find the slope that is the best fit line if that point is above above the plane then the w transpose x i will be positive and w transport if the point is below that uh below the slope then w transpose x i will be negative but that is not a prediction right so prediction wise you will be able to get some value equivalent to y right yes sir [Music] value scalar value i can understand things that's fine so you have return predicted value minus predicted value whole square so after getting the predicted value minus this value will be the error so your input okay this value is error yes sir are you going to find out a derivative of dy by dx or maybe with respect to error you are going to do something sir error will be so error will be also said except we can't we can't use that because sometimes the differentiating the uh error function loss function is not trivial so instead of that we can use the sjd algorithm or the gradient descent algorithm like because sometimes the different setting or losses is not revealed is not easy and equate is n equal to zero so so we can use the to say that we can't uh create a differential okay so how you are able to get a new value so let's suppose you are able to get a predicted value which is nothing but w into x i transpose that you have written and your actual value that you are expecting is equal to y y i you were expecting as an actual value now you will try to find out a difference difference of x i and y like predicted value w t x i and y i right yes now that will give you basically a error so in a function you have written so based on this you will be able to get a difference between predicted and the actual right yes now based on this error right based on this error so i am supposed to adjust a w yes sir yes sir now show me how so first randomly i select the bit suppose i have randomly assigned the weight w is equal to 0.002 randomly first random value of w is equal to something you can pick and choose any any number or maybe 0.002 let's suppose so by this we can update our weight what is dl by the way what is the like meaning of cdn is that loss or the error sir okay so you can error by dw we differentiate [Music] [Music] old minus learning it into d error by d w will be closer first of all yes i agree with that that is fine so first of all show me how you will be able to find out so you have to find out a new weight value right w you have to find out w old i told you w old is point zero zero two let's suppose that you will start from there now learning rate let's suppose i'm going to fix as point zero zero one yes so you know a value of w old you know a value of learning rate if i will be able to find out value of d error by dw right yes then i will be able to compute w new right so i am looking for this one d error by d w then only i can say that i am able to converge right or i am able to find out that i'm able to approach towards the minimum yes so it will be the dnr by dw will be yourself yeah show me how you can do that show me that so d error will be a squared function so if i is divided by pi squared so it will be as a vector differentiation means because the error error in the vector is uh so because our vector will be set this is a differentiation it will be a vector depressor because because we have different weights in different direction so we can use suppose in the x direction we use x uh and for we do partial differentiation so and the weight which will do it it's fine so whether it's a partial vector we'll see do it i have already given you a value of for like a w so you know w old and uh you know our learning rate so it's it's fine so you can like to like uh take it ahead from here w news is equal to w so what will you have given for the w also world is point zero zero two let's suppose one zero zero two yeah and learning rate is equal to point zero zero one yes so by this we can use so first i w new one will be the zero zero point as you have suggested and we calculated sorry this is my point i don't understand dl by dw so you have to show me you have to convince me that how this equation anyone can write right so this starts um anything there's anyone everyone knows that we will try to find out w new based on world there will be a learning rate and there will be rate of change of error with respect to w rate of change of error with respect to c right we all know that right in case of a differential yes how we calculate it i'm looking for that as you said you are good with mathematics of the algorithm so i think this is something we start from a linear regression itself right so just try to explain me that part nothing else looking for an explanation of dl by dw simple or d by dw whatever you say so dl by dw we can exactly say that if the limit like that like if in a loss function if it a little bit travel to like it del x will be equal to tends to 0 and in that space when we say tends to 0 it simply means that we are referring to a infinite number right yes between a scale so we can try to draw infinite number doesn't uh like uh uh doesn't matter that how small our scale is right so out of that infinite number which number i'm supposed to put over here because to calculate a w new i nee i know that number right so that's the reason so you have to show me an approach that by which i will be able to get this one simple question question is clear right yes i have to calculate w new you have all the values you have w world you have a learning rate if you will be able to calculate dl by dw for sure uh like as you said we do that right we do that everywhere so somehow if you will be able to do this calculation right so yeah so just show me that particular calculation right this is how i will be able to change just show me that so i think we all know what is our differential of divine by dx of like x square right so we all know that don't show me differential so many differential of error function so error function is error function is also a square function yeah that is fine so then show me so like the error function is like this so suppose this is an error function no no no i'm not supposed to like i suppose something else okay chris you have some question it's fine i think okay so uh you can stop sharing sort of uh okay uh we'll just go back to your uh review feedbacks you know uh because we have almost done 45 minutes of interview so please stop just uh sharing your screen okay my basics are my basic uh part of like i i think yeah it's okay that's why it's okay it's okay let let us tell right so let's let's hear some feedbacks from sudanshiw uh see you cannot say something like suppose okay you they'll see all your resume and they can ask you from anywhere right so now i think he's told me that uh he is good with this yeah and from there so we get into a gradient uh decent and uh it was just a problem for me whatever you said we asked you that only nothing uh separate from that or nothing because python also i feel it was very very easy not uh very difficult you know simple simple questions it was uh with respect to the fresher level whatever things are there we asked you that but it's okay uh so we'll hear the feedbacks from sudanshi and then i'll tell my feedbacks and based on that please prepare yourself i know you have just started at december uh uh still you have time uh just in two months definitely will be difficult to cover up but yes sudan should please to let them know about your feedback so you can go ahead and sort of what i feel is like uh i think you should practice uh again so when it comes to a fresher job right so let me tell you what you are going to need so as a fresher so for sure any will anyone will try to like uh check you on a coding part right and uh basic at least right like how you write a classes inside that methods or maybe how you try to pass uh data through by using some of the libraries or maybe uh like a basic things like iterator or maybe generator or maybe or something like like yours has asked right so this is uh like a basic expectation so i don't think that it's a like a too much uh from my interviewer's side in terms of expectation and when you try to mention clearly that i have studied or maybe i have gone through also i've studied only python scripting parts are not the object oriented on that part because the python which is literally yeah so it's rated table from the scripting part itself right and when you say python uh so you can't put a clause over there first of all that like i i don't know oops or i don't know this one and again if someone is asking you just to create a class and method right uh because i have seen that you were even struggling with that like uh how to write a classes and its method and argument uh right which is a very basic one it's not like a sub cordless method so i think you should practice on that particular part right and industry-wise so we don't use a scripting we use a complete modular coding i hope you are aware about it no something has been used at all right and uh we don't even write a code in that way we always try to follow a standard coding practice uh for writing all of those things so expectation will be there for sure at least a basic right basic means abcd and this is what has been asked now second coming to your like algorithm part right so whatever explanation that you were trying to give for xg boost i'll like say that i think we should look into that explanation right recordings will be available so you can hear yourself out uh that you have to say uh in terms of executive boost or boosting algorithm explanation plus gradients again and like i said uh you are good with the mathematics but what i feel is like you should practice you should focus on that part as well and don't try to overcome it overcome it in terms of like i know this i know that i know these things completely uh no one knows everything completely right even uh we don't know anything completely that is true and i think we all accept that that is fine so uh and whatever you are saying that okay fine so i know these things completely i think you must be in a position to justify it immediately like you said okay fine so i i know the gradient descent i know what is uh gradients and then uh like how we calculate the new weights or maybe how i would be able to calculate the error rate but uh when it comes to explanation so i think uh we are supposed to like a good with that with the fact right not just a verbal expression because everyone will be able to tell me that okay fine so we used to find out d by d w or d by dc that is fine right but when it comes to our computations of uh that one just it's it's again it will take two different three lines you have to write right but again many people fails over there then there is no meaning of saying that that i know graded descent or sgd or whatever name you were talking off now talking about right sir i know that but i no i know that i am not able to explain that part knowing means at least you are not supposed to explain but you should you must be able to write some equations immediately right at least this will be the expectation so just make sure that don't try to overcome it in any interview right once you know this okay fine so till end i will be able to explain i will be able to close this topic just try to uh like uh talk about that particular topic so that interview will go in your direction right it will not go against you at any point of time so i think you have a time and just just take your time and try to like practice on this front as much as possible right and initially you were talking about uh the project right so again uh think through that first of all that whether that is uh like uh you you have already mentioned that uh real-time project we all know that what is real time right and we even know that if you have mentioned some project and if you have mentioned a name or some maybe data set link we uh we all belong for the same industry right so it just take hardly a second of time a second of time couple of seconds of time to uh know that that okay fine so from where you have picked up which project right and you have said that it's a real time finally i said that okay fine so it's from kegel it always creates a bad impression right first you're saying that it's a real time then you're saying that's kegel about your second project about your first project you have mentioned that no so like uh this is project which i have done then uh you said that okay fine so this was a problem statement given by aicc i think this was a name right not sure about it yeah so yeah things and then so don't don't try to like uh do this thing that will be my sincere advice to you right and try to follow you have time right you have time it's just been three months when you have started so yes i think uh if you are like a following a complete pattern right yes for sure you will be able to uh like uh uh see very good things uh coming in front of you a good opportunity right and uh yeah so with that you will be able to get many more things okay so that would be my like advice suggestions well not advice it's a kind of a suggestion of friendly suggestion from my side right so yeah okay thank you uh sudan sure uh now sarah we're not coming to my feedback is that first of all you have to work on your presentation skills probably you may be knowing many things but uh in part of explaining things i think you're lagging in that because you will definitely see this particular video you know at that time you know you'll be able to see that you'll only not be able to understand what you're saying that will be the scenario and this will definitely even though you have a lot of knowledge even you know all the maths but again this will go against you that is the thing that i want to suggest apart from that see when you say when you are planning to become a data science data scientist should not say that okay i'm good at scripting only or i'm good at only this particular part i know only how to work with pandas in a nice way or probably i just know how to do model training you know so usually during model training only you'll be writing this kind of scripting code not only that so anything that will be related to python you really have to do let it be debugging let it be writing object oriented code let it be regarding modular programming exception different different things and that basic things also you are lagging which i saw because i asked you the basic difference between i treble iterators these all are very important things all together so you really have to work on that and as again suggested say you told that you started the preparation from december i would suggest try to practice more that will be the thing and try to showcase your skills showcasing your skills one way is that you write blogs you know try to stand in front of the mirror and try to explain someone if if possible try to explain your friends you see that how they are are they able to understand or not it is fine even we initially had a lot of knowledge but sometimes to present that skills usually was very very difficult so other than that i think anyhow you have to keep on working on it until you become better and probably in the future will again conduct this kind of interviews for you also you know so there you'll be able to compare your uh changes with respect to the knowledge that you're having okay so from today i will work very hard no problem okay so other than that i think uh thank you saurav again for attending this interview thank you sudhanshu uh for your time again thank you everyone you have been a nice uh people nice audience who are motivating sora and all yes please do hit the like button yes guys we'll see all in the next video have a great day thank you