Hlo Sir, thanks for that interview session that helped me in several ways. Today, I got job offer as Associate software engineer AI. This was the message by Sahil within a week after this mock interview.Congratulations👍👍
@@carlmarx6436 He was nervous throughout the interview. Also there is always a luck factor that plays an important role while interviewing in different companies.
I just realized my interviewer conducted entire interview today by picking questions from this mock interview.. I should have watched it earlier to give better answers.. What a coincidence 🤣
Please stop posting negative comments on the candidate. Atleast he was gracious enough to allow this interview to be posted. This is how people learn. Not everybody knows everything even if they r learning or have learnt in IIT.
From my experience, this is the best replica of real time data science interviews in companies. They dont go to ML. Theyll finish you in statistics itself.
@@snehangshubhattacharjee8072 ah here we have a typical person on the internet who thinks he knows the other person behind a screen. Get a life instead of trying to be a smartass on the internet ;)
@@anmol3457 @Anmol I hope you realise that a interview in which a young boy is getting bullied by two people who has a little bit more experience than him and everyone else celebrating that by calling it 'best replica' and people like you coming in support of that is a beautiful example of 'human stupidity'. However, I would really like to thank you for putting forward such beautiful examples.
Questions 1.When everyone talks about sample at that point of the time generally we use to take n-1 in the denominator ..so what is the reason behind that? 1.1 Have you heard of Bessels correction? 2. Can you please tell me a scenario of where I can apply CLT !? Effectively and able to get results out of it ? 2.1 what are the outcome of statistical tests? 2.2 what are the outcomes of normal distribution? What assumptions do you make from Normal distribution? 3.If data is given to me and do statistical analysis to find out outliers and what is the approach u are going to take to make it a normally distributed data set? 3.1 approach to find out and approach to Handle the outliers ? 3.2 without discarding how will you handle data? 3.3 what do you understand about normalisation? 3.4 what is the difference b/w normalisation and standardisation? 4. What is diff b/w Z statistics and T statistics where we are supposed to use that which one you r suppose to use and which one your are not supposed to use !? 4.1 can you please talk about diff b/w z and T in terms of standard deviation? Machine learning concepts : 5. Why root mean square error some times call as a worst errored calculations in case of regression!? 5.1 what is the probable solution for where we have one data in one scale and other as another scale 6. Can you please talk about error function or loss function in logistic ? 7. What’s your fav algorithm? 7.1 if you have to teach me a linear or logistic regression happen how you gonna explain me ? 8. you’ve worked on time series just tell me how you gonna do train test split ? 8.1 why sequential arranged ? (Time series) 8.2 will there be too much impact if the data has lot of variations!? 9.how do you decide no of clusters?( k mean ) 9.1 if you under sample the data what are the results (ans: loss of information) 9.2 what does over sampling do ? 9.3 have you tried any class weights !? 9.4 which algorithms you applied for supervised ? 9.5 are there any algorithms which will not get impacted by imbalance datasets ? 10. If have a problem statement you are a placement co ordinator you are facing a challenges there are many companies which used to visit your colleges and every company is having same set of requirements as a placement co ordinator you get stuck between company and student ! There are 1000 of resumes you have to map a correct resume to correct company ? You have to solve it using Ml not NLP!? 10.1 if you take Dt how you are going to define classes 11.you have done one hot encoding to change categories in to numerical what if you have a category like pin code ? 11. 1Why do we convert categories to numerical 12.You’ve built a NN how do u decide how many no of layers and hidden neurons to be used?(ans: keras tuner ) 12.1 why k fold cross validation is used? 12.2 what does cv is?
I have done M.phil in statistics & from this live interview session I understand ... Interviewer focused on basic statistics and I did this all statistics in graduation. So I think Data science is not hard for Statistics students 😉.....
Hi, this mock interview was very good. You guys asked simple questions, I knew most of the answers but explaining them properly is also important. Just a request Krish, when you think that you are not getting a proper answer then please give the right solution too, it can be brief. Thanks
Hi hope you are doing well. Actually I am looking for someone to practice and prepare data science interview+ improving communication. My background is in ECO and statistics. Currently I am pursuing MSc in quant eco.
"Yeah you can have 30 seconds." All the interwier is doing is imposing his thought process on the student, trying to have a kind of dominance. It's bad example of an interview.
@krish the virtual interview is good. If you could also share the candidates resume or CV for our reference. So that it would be beneficial for aspiring DS like me and others for preparing for future upcoming days.🙏
I am working in ML more than 8+ years, from my experience I can say,even if candidates in prestigious school perform poor it does not matter! ultimately they are the only one who will go to FANG due to college brand, other good candidates will only keep applying to these companies and will get filter out in HR screening
bat to aapki 100 percent sahi hai.... Recently Bank of America ki job aayi but they mentioned only graduates and postgraduates from iit are allowed....
@@shivamkrathghara3340 Ya they will choose Metallurgy, Civil, Mining ,Biotech or guy from any branch with these so called prestigious colleges who hardly have any interest in his/her core branch for Data Science job but will not give fair chance to students of other universities who are equally talented, hard working and from relevant streams
The technique of standardization and normalization both are used for feature scaling. Standardization we do when we know our data is normally distributed and our model makes that assumption like in linear regression. .normalization techniques, like min max we do if our model does not make any assumption about the data distribution like neural networks
Normalization : between 0 to 1 Standardization : mean 0 , SD 1 (preferred , because in normalization we might loose significant outliers) Logistic Regression : We can use for multiclass classification (one vs rest scheme) loss func : Cross Entropy
Just watched the first few minutes !! The interviewer needs to give the kid a break ! Rather than encouraging him, he is more focused on undermining his thought process. shame !
This is how most interviews are, companies will go through a large pool of candidates for a position and will not be so liberal with their time for each candidate.
Inter quartile range, and box plot are most common methods for outlier dtection. And capping the outlet at 95 or 98th percentile should work for handling outlier
He doesn’t know half the stuff undergraduate students at majority of the top universities are expected to have mastered by the end of their 2nd year Mathematics/Statistics/Econometrics courses (at most). For example the difference between a Students T and Z score is that Students uses the sample standard deviation (because the population variance is unknown) incurring a penalty in degrees of freedom whereas the Z score is calculated using a known population variance. This is taught in 1st year and as a professional Statistician it’s painful to watch him go through this mock interview. In saying that I harbour no disrespect towards the IITs as I still believe them to be premier engineering schools however this is not the quality of students I’d like to study with if I were doing my masters in Statistics there.
@@shafqat1138 what should i do to have a solid foundation in machine learning or data science ???? any suggestions because i am just getting started and i am in final year of engg would be helpful...should i focus more on mathematical part ???
@@sastashroud7646 Depends on how strong your background in Mathematical Statistics, Inference and Analysis (Real/Complex) is. I'm not a big fan of the whole "Machine Learning/Data Science" gimmick because these "Algorithms" have been around for quite a while in mainstream Econometrics/Statistics/Actuarial Science. It's just that computer aided Data Visualization tools have given ML Algorithms a bit more of a crunch as they're now more prone to being understood by your average non-Technical individual. My suggestion would be to focus on Big Data Analytics (Hadoop Architecture, Database Design & Management, et al) cause they are going to be the thing of the future. Big Data is in direct conflict with Classical Statistical Inference in that it allows real time processing of Population Scale Data whereas Classical Statistical Inference focused on using Sample Parameters to estimate Population Parameters. So my suggestion would be to focus on Big Data Analytics rather than Data Science.
Now that tells us how important are the basics to the ground i really like how you guys ripped it apart at the basics and the approach you guys took towards applications of a basic and then build it up gradually and slowly towards the end and for sahil from my past experience I would say that it was all show at the beginning and then it turned out to be a no go but he should have appeared for this when he was prepared to some extent atleast
34:54 sequentially arranged because the correlation with the recent lags will be higher and that will help us to arrive at the accurate future values. The correlation decreases exponentially from recent lags to past lags.
Ford Fulkerson algorithm with super sourse and super sink. Solve for max flow. Connect students to source and companies to sink. If student is interested in company have a link. Flow capacities b/w student and company is inf. B/w source and student is 1. B/W sink and company is 1.
Normalization is rescaling the data in the range of (0,1) we can do it by using min max scaler. Standardization is a form of normalization using z score.
Nice Interview. Sahil was completely nervous and couldn't think clearly for most of the questions. He was getting confused between the actual logic and the implementation part. For the resume assistance problem, he kept talking about the algorithms, but he seldom spoke about how the data could be extracted from the resume and how that could be handled. Basically, he somehow kept forgetting about how to extract and label the data and concentrated mainly on the algorithm. Anyway 2 technical interviewers are definitely going to make anyone nervous. Even I would have got nervous and forgotten some of the concepts. At least this was a good experience for him and viewers like me. I am new to this field and started learning Data Science a few months back. Anyway I can conclude that the resume keywords play a key role in the interviews and concepts are more important. Thank you for the wonderful session, Krish Sir and Sudhanshu Sir.
Tree-based algorithms perform well on imbalanced datasets. Boosting algorithms are ideal for imbalanced datasets because higher weight is given to the minority class at each successive iteration. 40:01
ONE interview can`t decide the level of IITK student, Mr. Sahil is also DPC at IITK. One non IITian person can`t think the effort he is putting in this course to maintain CPI Moreover any IITs are not bad It`s up to you how you want to learn just to get job or to go for PHd towards research or teaching and enhancing the knowledge of the subject. So who all are blaming Mr. Sahil here think once more.
I think the guy getting interviewed seems to not understand what statistics is, as what it is from first principals perspective. He only seems to know how to do something but now why it is done that way .
please provide the answers that the guy failed to give and also some detailed answer about the Sub- question like any other way can we do it ... Please make a separate video
Discrete Choice Modelling is all about using Logistic Regression for Multi class classification. I didn't think there is any such reason of not using it. In our university we are having an entire course about this.
logistic regression always use for binary classification because the sigmoid function=1/1+e^-y, gives you value between 0 to 1, you make limitation if predicted value is 0.5, then it belongs to class 1. there is a 'S' shaped graph of sigmoid function which tell this. that is it.
For the resume question, medium.com/@Synerzip/resume-ranking-using-machine-learning-implementation-47959a4e5d8e Basically we can use a logistic regression which outputs probability score of suitability of resume. Training data will be resumes selected and rejected in past. That being said this seems to be inefficient to me and unfair, because requirements might have changed from training set to now
First question answer is degree of freedom, basically if you have a sample and you know its mean and you know 9 values, 10th can be calculated so similarly using ‘n’ in the denominator introduces bias,just the way we avoid dummy trap by deleting the first dummy in similar way we decrease n value by 1 to remove that bias
If Say 20 companies are there and our main task is to classify each resume into two categories that candidate can seat in interview or not, this can be done by selecting features =what skill companies want ,we also need to think about the those candidates who have experience or have done project in specific field compared to those candidates who have only knowledge of that ,so that Candidate with experience should be given preference.Also try to keep no.of features less .Now we can train a model based on all features(input) for each feature we also need to associate no with it like if company want expert of any skill and only basic knowledge is sufficient then I can assign different number ,based on that I can create a model ,just like we define region in LPP like y>=3x1+4x2+..... Something like that and . Then I can make 10 models (no. Of companies coming for placement drive) , for each model we have output But if there is any constraint that a single candidate can only appear for 3 or 4 interview, then we need to think further. Now we have models ready with me ,and I have data as key skill of students , I will pass all these data to our model and i will measure the shortest distance of this data to each contraint line for specific model and sum it over all straight line boundary relating to each constarint like y>=6x1+3x2. And we will find such distance for all data(student skill) given in resume of student and then apply scaling so that ranking of student can be done a/c to best suitable canditate , Similar thing can be done for all model . But I think In this approach I haven't use ML . 🤨
In pin code can't we break it down in to components like 1st few characters and like mid and then last component to finally do a mean,target or one hot encoding of each component?
I think this will clear the use of CLT ..this is the post that I had posted on LinkedIn before some days...hope this will be helpful to understand it.. CLT states that the distribution of sample means approximates a normal distribution as the sample size becomes larger assuming that all samples are identical in size and regardless of population distribution shape.. Best part of this theorem in our real life problems is that regardless of distribution(i.e If we haven't idea about distribution of population) shape one can compute the probability of an event for that distribution using samples and CLT theorem For a layman who is not from statistics stream Can understand it by one simple example- Let we want to check what is the probability (i.e chance) that an Indains have income more than 50,000 then we can compute it using Central Limit theorem and sample (i.e a sub part population data) even we haven't idea about the distribution of income of Indians
Can I say that if my sample is having a lot of values and I am supposed to be calculating the standard error of a lot of samples(a lot) from the Indian population then I can use the CLT here as calculating the SE for every sample won't be feasible... Right?? M a student.. ignore if I made any wrong assumptions
@@greatgreek3569 What you have written is littile bit confusing...like you are talking about a sample or a group of samples...as per I understood if you are looking to calculate standard error then you can calculate standard error for a single sample or standard error of group of sample means normally ..here CLT need not to be used ..CLT is only used when we haven't idea about the distribution of population and we want to get some information about that population using their sub part like chance of occurrence of an event or something.. And one more thing you can notice that the idea behind using sample means is that it will estimate population Better than a single sample as it consists of many samples Hope you got it..
Normalization/standardization does not help treat outliers. They only change the mean and scale of the data. I would like to know if there is any normalization technique that helps treat outlier. Yes, certain transformations can be applied to data to treat outliers.
Yes at some I want to clarify u that abs -transform can treat outlier as it also a part of normalisation as some of the people got confuse normalisation as minmaxscaler only that actually normalisation have many sub-category.
for someone who has an IIT background, I expected him to be able to explain his understanding on the concepts asked about in a more clear manner. Pretty disappointing.
Eh, why? Because he had better profs? (not true btw) The biggest determiner of your understanding is how much you've self studied and ruminated on the topic yourself. No amount of teaching can give you that clarity
He is a post graduate student ..and the real iitians who have passed Jee advanced and earned a bachelor degree from among the top 5 iits would be more knowledgeable than the two interviewer..
The attitude of Sudhanshu isn't good, infact very mean towards the interviewee which could diminish his confidence. Pls don't laugh at them and say this like I ll give u 30secs
Guys pls stop degrading that guy he isn't some god he as much as human as us. Maybe he is scared and have some other issue god knows. Stop degrading Just because you failed once that doesn't mean you can't get up again. At least he gets a chance to improve. This is so stupid.
Yeah,these are like that they only want to hear those predefined bookish answers,not anything new,because when they are taking interview ,they think themselves as the boss and thinks that they know everything
Im stoked at the fact that people think this is how interviews are conducted. Rather than encouraging the candidate and asking him probing questions they are undermining his thought process
Main IIT ni ja paya.. Iss video ko dekhne se pehle mujhe bahot guilt tha. Thanks to this video, now I feel happy and I won't send my kids to IIT ever. 😜
Teacher's in IIT'S are best teachers India Could produce but not everyone you can see many NPTEL courses and check their teaching techniques. Many teachers just read those PPT's and , the least amount of effort. The main problem is blindly accepting the fact " IIT's are absolute" (andh vakti), Even Major companies follow this.(I am not from Engineering background Physics student). Getting into IIT's is harder due to Reservation.
As around time stamp 34:05 Krish sir asked sahil about *Why do we select time series datapoint which are located in the end of the dataset as test data, as we can't split time series data like universal method : my answer : Because of data shift ? (if i am wrong please correct me ) Thank you.
Can we use features and classification using neural networks by inputting 1000 datapoints regarding those 10 features and processing it on train test CV and getting a proper model with it ,with one vs all output format
Am looking for a machine learning/data science internship and don't know how to go about it. Please any tip will be helpful. And I will like to know your domain and how you chose the right domain for yourself
Hi hope you are doing well and have done some internship. Actually I am looking for someone to practice and prepare data science interview+ improving communication. My background is in ECO and statistics. Currently I am pursuing MSc in quant eco.
Knowing of central limit theorem help us to know the spread of data easily because as normal distribution helps us to calculate pdf, cdf and we can conclude at what percentage of data lying under the limit. Correct me if i am wrong
Yes. You are wrong. Central limit theorem allows you to expect almost types of distribution to asymptomatically follow normality. There is no relevance of data spread here.
@@TAF3000 he is not even listening to.Nor he is giving him time to think. In real life, the interviews are never conducted in this way or may be I never faced such interviewer, not even at FANG
i watched this till 31.36 sec, thanks god this boy didn't said his fav algoritham from deep learning, otherwise sudhanshu gonna r*p* him. but this session will definitely help this boy.
Hlo Sir, thanks for that interview session that helped me in several ways. Today, I got job offer as Associate software engineer AI.
This was the message by Sahil within a week after this mock interview.Congratulations👍👍
howwww!!!!
Logistic for multiclass lol ...
@@carlmarx6436 tag bro 😂
Hahahaha
@@carlmarx6436 He was nervous throughout the interview. Also there is always a luck factor that plays an important role while interviewing in different companies.
Now this is called real interview. And this is how normally interviewee behaves while interview. Thank you for sharing this with us. 😊
I'm also a so called statistics student.
After seeing this interview, I really think I should study
I just realized my interviewer conducted entire interview today by picking questions from this mock interview.. I should have watched it earlier to give better answers.. What a coincidence 🤣
Please stop posting negative comments on the candidate. Atleast he was gracious enough to allow this interview to be posted. This is how people learn. Not everybody knows everything even if they r learning or have learnt in IIT.
Correct
From my experience, this is the best replica of real time data science interviews in companies. They dont go to ML. Theyll finish you in statistics itself.
If you think this is the best replica of real time data science interview, I am very sure about the kind of interviews you have attended!
Get a life!
@0_0 There are various other comments which would answer your query.
@@snehangshubhattacharjee8072 ah here we have a typical person on the internet who thinks he knows the other person behind a screen. Get a life instead of trying to be a smartass on the internet ;)
@0_0 perhaps he's frustrated with his life. But I'm not him so I'm not gonna assume anything.
@@anmol3457 @Anmol I hope you realise that a interview in which a young boy is getting bullied by two people who has a little bit more experience than him and everyone else celebrating that by calling it 'best replica' and people like you coming in support of that is a beautiful example of 'human stupidity'.
However, I would really like to thank you for putting forward such beautiful examples.
It's not about college, knowledge is a practice . The more u practice and be in live scenarios u learn more.
Can't believe if questions are going to be this straight forward in an interview setting!
Questions
1.When everyone talks about sample at that point of the time generally we use to take n-1 in the denominator ..so what is the reason behind that?
1.1 Have you heard of Bessels correction?
2. Can you please tell me a scenario of where I can apply CLT !? Effectively and able to get results out of it ?
2.1 what are the outcome of statistical tests?
2.2 what are the outcomes of normal distribution? What assumptions do you make from Normal distribution?
3.If data is given to me and do statistical analysis to find out outliers and what is the approach u are going to take to make it a normally distributed data set?
3.1 approach to find out and approach to Handle the outliers ?
3.2 without discarding how will you handle data?
3.3 what do you understand about normalisation?
3.4 what is the difference b/w normalisation and standardisation?
4. What is diff b/w Z statistics and T statistics where we are supposed to use that which one you r suppose to use and which one your are not supposed to use !?
4.1 can you please talk about diff b/w z and T in terms of standard deviation?
Machine learning concepts :
5. Why root mean square error some times call as a worst errored calculations in case of regression!?
5.1 what is the probable solution for where we have one data in one scale and other as another scale
6. Can you please talk about error function or loss function in logistic ?
7. What’s your fav algorithm?
7.1 if you have to teach me a linear or logistic regression happen how you gonna explain me ?
8. you’ve worked on time series just tell me how you gonna do train test split ?
8.1 why sequential arranged ? (Time series)
8.2 will there be too much impact if the data has lot of variations!?
9.how do you decide no of clusters?( k mean )
9.1 if you under sample the data what are the results (ans: loss of information)
9.2 what does over sampling do ?
9.3 have you tried any class weights !?
9.4 which algorithms you applied for supervised ?
9.5 are there any algorithms which will not get impacted by imbalance datasets ?
10. If have a problem statement you are a placement co ordinator you are facing a challenges there are many companies which used to visit your colleges and every company is having same set of requirements as a placement co ordinator you get stuck between company and student ! There are 1000 of resumes you have to map a correct resume to correct company ?
You have to solve it using Ml not NLP!?
10.1 if you take Dt how you are going to define classes
11.you have done one hot encoding to change categories in to numerical what if you have a category like pin code ?
11. 1Why do we convert categories to numerical
12.You’ve built a NN how do u decide how many no of layers and hidden neurons to be used?(ans: keras tuner )
12.1 why k fold cross validation is used?
12.2 what does cv is?
Great job
@@boopalsingh08 thank you
Thank you for 😊 your efforts
51:00 we can use clustering and then depending on skillsets we can cluster them and sort the resumes
We use t-test when std deviation is not known used for comparing 2 population means pair or independent...and Z- test is simple..!!
After watching this interview mock test my confidence level has been boosted.
I have done M.phil in statistics & from this live interview session I understand ... Interviewer focused on basic statistics and I did this all statistics in graduation. So I think Data science is not hard for Statistics students 😉.....
, lol , now days focus on Machine learning and deep learning when you consider those things, you have to have good understanding about Algorithm,
Hi, this mock interview was very good. You guys asked simple questions, I knew most of the answers but explaining them properly is also important. Just a request Krish, when you think that you are not getting a proper answer then please give the right solution too, it can be brief. Thanks
Hi hope you are doing well. Actually I am looking for someone to practice and prepare data science interview+ improving communication. My background is in ECO and statistics. Currently I am pursuing MSc in quant eco.
Congratulations Sahil Saini..it's so motivating to see you here.
Outliers can be detected by Inter Quartile Range.
All values out of range Q1-1.5IQR & Q3+1.5IQR are outliers.
Wow some hardcore fundamentals discussion and very conceptual as well
Data science k coaching se jitna sikhne ko nhi mila..Usse kahi jyada sikhne ko idhar milgaya..thank you
"Yeah you can have 30 seconds."
All the interwier is doing is imposing his thought process on the student, trying to have a kind of dominance.
It's bad example of an interview.
@krish the virtual interview is good. If you could also share the candidates resume or CV for our reference. So that it would be beneficial for aspiring DS like me and others for preparing for future upcoming days.🙏
I am working in ML more than 8+ years, from my experience I can say,even if candidates in prestigious school perform poor it does not matter! ultimately they are the only one who will go to FANG due to college brand, other good candidates will only keep applying to these companies and will get filter out in HR screening
bat to aapki 100 percent sahi hai....
Recently Bank of America ki job aayi but they mentioned only graduates and postgraduates from iit are allowed....
@@shivamkrathghara3340 Ya they will choose Metallurgy, Civil, Mining ,Biotech or guy from any branch with these so called prestigious colleges who hardly have any interest in his/her core branch for Data Science job but will not give fair chance to students of other universities who are equally talented, hard working and from relevant streams
Yaa this is what happening in present situation
@@annamdurgashivaprasad4041 this only happens in India. In countries like usa they test your skills
The technique of standardization and normalization both are used for feature scaling. Standardization we do when we know our data is normally distributed and our model makes that assumption like in linear regression. .normalization techniques, like min max we do if our model does not make any assumption about the data distribution like neural networks
Normalization : between 0 to 1
Standardization : mean 0 , SD 1 (preferred , because in normalization we might loose significant outliers)
Logistic Regression : We can use for multiclass classification (one vs rest scheme)
loss func : Cross Entropy
How does normalization remove outliers
@@rafibasha4145 it doesn't..instead it just remove the skewness by normalizing that data
@@ArunYadav-qq1cj ,it shouldnt even transform the distribution i belive
Just watched the first few minutes !! The interviewer needs to give the kid a break ! Rather than encouraging him, he is more focused on undermining his thought process. shame !
Krish was still okay and encouraging the candidate but Sudhanshu might have good amount of knowledge but the pride he has, is not favorable.
You are right - agree with you.
This is how most interviews are, companies will go through a large pool of candidates for a position and will not be so liberal with their time for each candidate.
This is so typical of Indian Interviewers. Grilling an interviewee isn't going to help!
@@swatantrachib5916 I don't think so - it won't go this much longer.
great interview .. keep posting mock interview .but sir also keep telling answer or hint so v can further research it
51:50 me in every interview
Inter quartile range, and box plot are most common methods for outlier dtection. And capping the outlet at 95 or 98th percentile should work for handling outlier
dil kush hogya interview dekh ker... great help for u...Tumb up for sahil for facing the master of datasciience
Sahil, you're doing great and I believe that this interview gonna push you to do MORE GREATER things in life.
All the best buddy!
This guy won't be same person after this interview.
He doesn't know half the stuff. His basics are not clear.
He doesn’t know half the stuff undergraduate students at majority of the top universities are expected to have mastered by the end of their 2nd year Mathematics/Statistics/Econometrics courses (at most). For example the difference between a Students T and Z score is that Students uses the sample standard deviation (because the population variance is unknown) incurring a penalty in degrees of freedom whereas the Z score is calculated using a known population variance. This is taught in 1st year and as a professional Statistician it’s painful to watch him go through this mock interview.
In saying that I harbour no disrespect towards the IITs as I still believe them to be premier engineering schools however this is not the quality of students I’d like to study with if I were doing my masters in Statistics there.
@@shafqat1138 what should i do to have a solid foundation in machine learning or data science ???? any suggestions because i am just getting started and i am in final year of engg would be helpful...should i focus more on mathematical part ???
@@sastashroud7646 Depends on how strong your background in Mathematical Statistics, Inference and Analysis (Real/Complex) is. I'm not a big fan of the whole "Machine Learning/Data Science" gimmick because these "Algorithms" have been around for quite a while in mainstream Econometrics/Statistics/Actuarial Science. It's just that computer aided Data Visualization tools have given ML Algorithms a bit more of a crunch as they're now more prone to being understood by your average non-Technical individual.
My suggestion would be to focus on Big Data Analytics (Hadoop Architecture, Database Design & Management, et al) cause they are going to be the thing of the future. Big Data is in direct conflict with Classical Statistical Inference in that it allows real time processing of Population Scale Data whereas Classical Statistical Inference focused on using Sample Parameters to estimate Population Parameters.
So my suggestion would be to focus on Big Data Analytics rather than Data Science.
Now that tells us how important are the basics to the ground i really like how you guys ripped it apart at the basics and the approach you guys took towards applications of a basic and then build it up gradually and slowly towards the end and for sahil from my past experience I would say that it was all show at the beginning and then it turned out to be a no go but he should have appeared for this when he was prepared to some extent atleast
My man nailed it - successfully answered most of questions wrong 😂😂 worst part is most of these are covered as a part of Krish Naik channel
34:54 sequentially arranged because the correlation with the recent lags will be higher and that will help us to arrive at the accurate future values. The correlation decreases exponentially from recent lags to past lags.
doing a refresher course on stats teahces all this, idk what bullshit has this senior read about.
after looking at people finding interviewer rude i think people haven't given any vivas
Ha bhai yaha aise hi hota hai 😂😂
Hii guys..your interviews are good..one REQUEST is ..please do not smile or laugh when the candidate is not able to answer the questions..
Ford Fulkerson algorithm with super sourse and super sink. Solve for max flow. Connect students to source and companies to sink. If student is interested in company have a link. Flow capacities b/w student and company is inf. B/w source and student is 1. B/W sink and company is 1.
Normalization is rescaling the data in the range of (0,1) we can do it by using min max scaler. Standardization is a form of normalization using z score.
he is having trouble understanding technicality of the questions
does not think like a programmer
Nice Interview. Sahil was completely nervous and couldn't think clearly for most of the questions. He was getting confused between the actual logic and the implementation part. For the resume assistance problem, he kept talking about the algorithms, but he seldom spoke about how the data could be extracted from the resume and how that could be handled. Basically, he somehow kept forgetting about how to extract and label the data and concentrated mainly on the algorithm. Anyway 2 technical interviewers are definitely going to make anyone nervous. Even I would have got nervous and forgotten some of the concepts. At least this was a good experience for him and viewers like me. I am new to this field and started learning Data Science a few months back. Anyway I can conclude that the resume keywords play a key role in the interviews and concepts are more important. Thank you for the wonderful session, Krish Sir and Sudhanshu Sir.
Tree-based algorithms perform well on imbalanced datasets. Boosting algorithms are ideal for imbalanced datasets because higher weight is given to the minority class at each successive iteration.
40:01
Big thank you for such videos. They are very useful and informative.
Good question and are not so tough...!! Wish same question would have been asked to me...!!
Can you type the answers you know. Will be helpful for all.
Good to know type of questions we should prepare for.
ONE interview can`t decide the level of IITK student, Mr. Sahil is also DPC at IITK. One non IITian person can`t think the effort he is putting in this course to maintain CPI Moreover any IITs are not bad It`s up to you how you want to learn just to get job or to go for PHd towards research or teaching and enhancing the knowledge of the subject. So who all are blaming Mr. Sahil here think once more.
You r 80 percent right. Not all students got the jobs from IITS
Great Job 👍, and Sahil Did very well
The boy was just a bit stressed. I think he did pretty well.
Normalization is forcing to behave normally by using transformation like Box Cox
I think the guy getting interviewed seems to not understand what statistics is, as what it is from first principals perspective. He only seems to know how to do something but now why it is done that way .
yup i think he just did what he was told ...he didnt ask question why we do. so ?
please provide the answers that the guy failed to give and also some detailed answer about the Sub- question like any other way can we do it ...
Please make a separate video
Discrete Choice Modelling is all about using Logistic Regression for Multi class classification. I didn't think there is any such reason of not using it. In our university we are having an entire course about this.
logistic regression always use for binary classification because the sigmoid function=1/1+e^-y, gives you value between 0 to 1, you make limitation if predicted value is 0.5, then it belongs to class 1. there is a 'S' shaped graph of sigmoid function which tell this. that is it.
Please answer the questions krish sir so that we can learn.
For the resume question,
medium.com/@Synerzip/resume-ranking-using-machine-learning-implementation-47959a4e5d8e
Basically we can use a logistic regression which outputs probability score of suitability of resume. Training data will be resumes selected and rejected in past. That being said this seems to be inefficient to me and unfair, because requirements might have changed from training set to now
First question answer is degree of freedom, basically if you have a sample and you know its mean and you know 9 values, 10th can be calculated so similarly using ‘n’ in the denominator introduces bias,just the way we avoid dummy trap by deleting the first dummy in similar way we decrease n value by 1 to remove that bias
All the people hatin the interviewer ..that's how mock interview should be .. if you are prepared for the worst then you can do good in any scenario
If Say 20 companies are there and our main task is to classify each resume into two categories that candidate can seat in interview or not, this can be done by selecting features =what skill companies want ,we also need to think about the those candidates who have experience or have done project in specific field compared to those candidates who have only knowledge of that ,so that Candidate with experience should be given preference.Also try to keep no.of features less .Now we can train a model based on all features(input) for each feature we also need to associate no with it like if company want expert of any skill and only basic knowledge is sufficient then I can assign different number ,based on that I can create a model ,just like we define region in LPP like y>=3x1+4x2+..... Something like that and .
Then I can make 10 models (no. Of companies coming for placement drive) , for each model we have output
But if there is any constraint that a single candidate can only appear for 3 or 4 interview, then we need to think further.
Now we have models ready with me ,and I have data as key skill of students , I will pass all these data to our model and i will measure the shortest distance of this data to each contraint line for specific model and sum it over all straight line boundary relating to each constarint like y>=6x1+3x2.
And we will find such distance for all data(student skill) given in resume of student and then apply scaling so that ranking of student can be done a/c to best suitable canditate ,
Similar thing can be done for all model .
But I think In this approach I haven't use ML .
🤨
U can use a convert into flow problem. Its a graph problem. Matching problem.
Now, it feels like real interview. Great session. Thanks you for sharing it.
2nd question answer is inventory management.. by creating confidence interval we can predict easily the stock for a given month at 90% -95% level
Holy shit! You destroyed the guy!! You should have invited a ug guy! He would have destroyed this interview :)
In pin code can't we break it down in to components like 1st few characters and like mid and then last component to finally do a mean,target or one hot encoding of each component?
maksuda be tu
sir apka video dekhta hun 😍😍😍 apne upar confidence a ara hai sir
Great one krish and sudanshu..also plz get somebody from oracle apps (ERP) with 2/3yrs Experience and who has made the transition in DS
U give too much confidence sir to learn more
I think this will clear the use of CLT ..this is the post that I had posted on LinkedIn before some days...hope this will be helpful to understand it..
CLT states that the distribution of sample means approximates a normal distribution as the sample size becomes larger assuming that all samples are identical in size and regardless of population distribution shape..
Best part of this theorem in our real life problems is that regardless of distribution(i.e If we haven't idea about distribution of population) shape one can compute the probability of an event for that distribution using samples and CLT theorem
For a layman who is not from statistics stream
Can understand it by one simple example-
Let we want to check what is the probability (i.e chance) that an Indains have income more than 50,000 then we can compute it using Central Limit theorem and sample (i.e a sub part population data) even we haven't idea about the distribution of income of Indians
Can I say that if my sample is having a lot of values and I am supposed to be calculating the standard error of a lot of samples(a lot) from the Indian population then I can use the CLT here as calculating the SE for every sample won't be feasible... Right?? M a student.. ignore if I made any wrong assumptions
@@greatgreek3569 What you have written is littile bit confusing...like you are talking about a sample or a group of samples...as per I understood if you are looking to calculate standard error then you can calculate standard error for a single sample or standard error of group of sample means normally ..here CLT need not to be used ..CLT is only used when we haven't idea about the distribution of population and we want to get some information about that population using their sub part like chance of occurrence of an event or something..
And one more thing you can notice that the idea behind using sample means is that it will estimate population Better than a single sample as it consists of many samples
Hope you got it..
We can also convert it into optimisation problem - linear programming
Before commenting that person do know where you are guys. Just encourage these questions are not that easy... Just try to learn
Z score , can only be used when data is normally distributed and also IQR?
Z is used generally when data is normal, if not normal t test is used
If I join your channel with data science materials can I view members only streams? Or do I need other membership
Invite IIIT Delhi peeps to interview. If you want I can be a voulenteer.
Sure drop a mail in krishnaik06@gmail.com
@@krishnaik06 Okay definitely sir, let me handle my End Sem Exam first ;)
@@krishnaik06 Sir, how about a applied Statistics and Informatics student from IIT Bombay?
@@deepanshuchoudhary4598 Gentleman be patient I'm waiting for a suitable time.
I am machine learning beginner and I could literally answer most of questions. All thanks to Applied AI Course online.
Which course
@@HarisRehman search for appliedaicourse on google
Normalization/standardization does not help treat outliers. They only change the mean and scale of the data. I would like to know if there is any normalization technique that helps treat outlier. Yes, certain transformations can be applied to data to treat outliers.
Yes at some I want to clarify u that abs -transform can treat outlier as it also a part of normalisation as some of the people got confuse normalisation as minmaxscaler only that actually normalisation have many sub-category.
Sahil bhai, deep institute mein jo padha tha sab bhool gya kya?? :)
Sudhir sir ka lo interview feminism pe
for someone who has an IIT background, I expected him to be able to explain his understanding on the concepts asked about in a more clear manner. Pretty disappointing.
He is a first year student. And people generally learn prob-stat and data mining in 3rd year.
@@himanshudaksh4957 he is a masters student in statistics, MSc statistics, not an engineering student.
@@mthetree Sorry for my statement. I didn't notice that.
Eh, why? Because he had better profs? (not true btw) The biggest determiner of your understanding is how much you've self studied and ruminated on the topic yourself. No amount of teaching can give you that clarity
He is a post graduate student ..and the real iitians who have passed Jee advanced and earned a bachelor degree from among the top 5 iits would be more knowledgeable than the two interviewer..
The attitude of Sudhanshu isn't good, infact very mean towards the interviewee which could diminish his confidence. Pls don't laugh at them and say this like I ll give u 30secs
Is this how they conduct data science interview ??
Nope, not at all !! sudharshan passes statements just like that !! they dont make you suffocated
it's possible, that the same type of people are present in Indian companies. AI engineers are everywhere, so these questions could be copied by all
Guys pls stop degrading that guy he isn't some god he as much as human as us. Maybe he is scared and have some other issue god knows. Stop degrading
Just because you failed once that doesn't mean you can't get up again. At least he gets a chance to improve. This is so stupid.
z statistics we can call if population variance is known and t-statistics we call if p var is unknown.
Why is the interviewer not ready to hear a word outside his dictionary? Lol it's like that teacher from 3 idiots. No offence tho
Yeah,these are like that they only want to hear those predefined bookish answers,not anything new,because when they are taking interview ,they think themselves as the boss and thinks that they know everything
It is shame that a Master student in statistics is being questioned by those who have no knowledge of statistics themselves.
Absolutely. I was about to write this. Ratta marke ake hero ban raha. Asal mein zyada dimag hai nai.
Im stoked at the fact that people think this is how interviews are conducted. Rather than encouraging the candidate and asking him probing questions they are undermining his thought process
Focal Loss is also good for mis balanced data
This guy is super brave!😂 although he wasn’t technical enough in his answers, he was courageous enough to do this publicly🤣. Good job fam!
Main IIT ni ja paya.. Iss video ko dekhne se pehle mujhe bahot guilt tha. Thanks to this video, now I feel happy and I won't send my kids to IIT ever. 😜
Teacher's in IIT'S are best teachers India Could produce but not everyone you can see many NPTEL courses and check their teaching techniques.
Many teachers just read those PPT's and , the least amount of effort. The main problem is blindly accepting the fact " IIT's are absolute" (andh vakti), Even Major companies follow this.(I am not from Engineering background Physics student).
Getting into IIT's is harder due to Reservation.
Bro there is a huge difference between UG and PG students. You may think anything that helps you have a good sleep.
@@devanshsharma3502 but this pgs will be earning much more than ugs in other colleges .. so it's an win situation I guess
Pin code is a tuple made up of (LAT, LONG). lat and long are numerical data by themselves.
I will never sit with u guys😃
🤣🤣🤣
😂😂😂😂😂😂😂
As around time stamp 34:05
Krish sir asked sahil about *Why do we select time series datapoint which are located in the end of the dataset as test data, as we can't split time series data like universal method :
my answer : Because of data shift ? (if i am wrong please correct me )
Thank you.
Can we use features and classification using neural networks by inputting 1000 datapoints regarding those 10 features and processing it on train test CV and getting a proper model with it ,with one vs all output format
Am looking for a machine learning/data science internship and don't know how to go about it. Please any tip will be helpful.
And I will like to know your domain and how you chose the right domain for yourself
Hi hope you are doing well and have done some internship. Actually I am looking for someone to practice and prepare data science interview+ improving communication. My background is in ECO and statistics. Currently I am pursuing MSc in quant eco.
N-1 is to factor in the Degree of freedom where you are taking a sample and not the full population.
Hope my point makes sense.
Knowing of central limit theorem help us to know the spread of data easily because as normal distribution helps us to calculate pdf, cdf and we can conclude at what percentage of data lying under the limit. Correct me if i am wrong
Yes. You are wrong. Central limit theorem allows you to expect almost types of distribution to asymptomatically follow normality. There is no relevance of data spread here.
Thanks 👍 for this info
You were good Sahil 🙂
I think one should take one or two days to prepare before interview
36:38 we do differencing to try to reduce those shocks in the series
Can u share the resume also
So that we can get to know the format for reference
The interviewer just wanted to hear what he wants to hear.
Nothing more nothing less
Exactly...
@@TAF3000 he is not even listening to.Nor he is giving him time to think. In real life, the interviews are never conducted in this way or may be I never faced such interviewer, not even at FANG
Lol, not really. That's minimal expected from IIT statistician. They heard him.
11:34 What's with the attitude man ? we know you are good but sir give the guy the time he deserves. sudhanshu
exactly, That man Sudhanshu is so triggered, he gets agitated in every sec of the interview. Crass and egotist attitude. No patience at all.
@@parthtool you are totally correct.
Logistic uses binary cross entropy as loss function
These guys are gift for all aspiring DS!
One is Krish Naik . Who is second person
i watched this till 31.36 sec, thanks god this boy didn't said his fav algoritham from deep learning, otherwise sudhanshu gonna r*p* him.
but this session will definitely help this boy.
That's ranvijay from roadies 😭
Sudhansu with op questions man.. shit even if I can explain he will find another question from tht obvio boss level..
This guy (Sudhanshu Kumar) is incapable of any job in India, if anyone can find, please let me know how many times did he say "LIKE A". !
Just dominating doesn't makes you smart - is this the way interview is done lol .
Nice to see question answer 💐💐