Patience when things are not going in your way.
Convert your worries to actions- Improve your coding skills with practice.
You should be positive about your future. No matter what. You will get what you want.
First Have the patience to read my long post :-)
It is only my way of looking at things. Please ignore if something hurts you.
Many of us look for easy ways(short cuts/tips) to crack any type of interviews like going through previous Interview questions etc...
Right from fresher to Project Manager level are looking for these jobs. The answer to this question is different for different people.
Ask yourself the following questions:
Are you a fresher, or experienced, If experienced(other than data science), how many years: less than 5, 5-10, or more than 10.
Here I consider (0-3 years) as Fresher Bin1: because they are in the early stage of Career.
(3 -10) years as Bin2.
Above 10 years as another Bin3.
Fresher(Bin1) :
We know what to do but we look for others to guide us. Just I am reminding you what you have forgotten. Before appearing for an interview You should be good at following things.
1. Machine learning fundamentals.
2. Python programming
3. Good presentation skills
You already know this Right. How to get these.
First and foremost thing:
Are you trying on your own or under some guidance like taking the course offline or online?
Because you should have solid Machine learning concepts or material with you.
If you do not have this, prepare your material first. Of course, you should be good at coding as well.
Assuming you know data preprocessing.
How to prepare your material
1. List down the Machine learning Algorithms you are familiar with.
2. Browse the blogs, books, videos, get the intuition of each algorithm how it works.
Write a few lines( not more than 50 words) about the algorithm in simple terms I repeat layman terms. Look into the Maths part when it is needed only.
3. Get hold of the Algorithm even with numerical examples wherever it is possible
For example in K-nearest neighbor Given data points(instances) with target how to do you decide the target of the test data point. Here we do simple distance calculation and we take mode based on k value.
Likewise for other algorithms how to choose root node(decision trees), Clustering (Computing Centroid and forming clusters), Naive Bayes(Bayers theorem), and so on wherever it is possible.
4. Prepare sample code snippets for each algorithm in python.
5. Explain to your friends in very simple terms, so that they can understand.
Likewise, all the major concepts like bias-variance, overfitting-underfitting, cross-validation, splitting data into train data, and test data, whichever concept you read to give your best.
Do not forget the statistics.
Now comes to Interview,
Give simple terms. Explain the concept in simple terms and say the concept name. ( Strictly no shortcut answers)
What is overfitting:
People talk about bias-variance trade-off though they are not clear about that. Is there any other way to answer this.
Think in terms of the error in train data and test data.
(If you are clear with this and confident that you can convince the interviewer you can opt this, but do not get into unnecessary trouble by speaking terms which you don't know)
Know what you speak.
No shortcut answers
Eg; What is Principal Component Analysis?
Many of us give a one-line answer It is a dimensionality reduction technique. which is not recommended.
What if you explain in the following way.
Start with data: In some domains, there is a problem either classification or regression.
Now when there are more attributes/dimensions/columns in the data, If we want to see a quick look at the results. I opt for a less no. of dimensions as much as possible. I can find it by finding orthogonal components where each component is a linear combination of the other components. These are called eigenvectors which we find through eigenvalues from Covariance Matrix by solving equation (A-lamda) * x=0. I select only the top few components which explain more variance in the data through the scree plot. (Though we are not going into mathematics you should have this in your mind).
If at all interviewer asks the next question to solve the equation you should be ready to do it.
The Disadvantage of this is we can't explain it to the business as each component is a linear combination. The advantage is sometimes linear combination will make sense; Eg: predict the score of a dish where attributes are ingredients. Now component becomes a linear combination of ingredients and etc..
You should answer every question with confidence. This comes through practice.
In parallel to the above, you should have basic python programming skills as well. Solve python exercises from different websites, need not be from data science. Type python programming exercises in Google take top10 websites.
Get your hands dirty in Pandas, NumPy, various operations on data, preprocessing data,
Take several datasets to build multiple algorithms without copying code from the Internet.
Improve debugging skills.
Change the approach to Learning by debugging from Learning by Coding once you are comfortable with coding.
If you do everything as specified, It will boost your confidence. This helps you to crack the interview.
Further, have some idea on what is going on in data science; Get some basic concepts in Text mining and deep learning if you are a beginner. You can apply the same method even in these areas as well.
Bin2: By doing above all how do you use your current job experience. Get the domain knowledge; Try to link data science problems to your domain,
From a fresher, we can expect technical. But from experienced think like if you want to take a person into your project What you expect from him.
As you are experienced start exploring git, version control tools, how can you use your previous experience into DataScience? It is just like you are moving to new technology. Use your commonsense and experience, Show your experience in thoughts and talks. You are not a kid anymore.
Bin3: Not much to say; They should think, In What way we can add value to the existing system through data science.
Stop depending on the others, You are an experienced person as a Manager, Team Lead, Module Lead, Tech Manager etc.. You know how the system works. No need to reinvent the wheel.
Convert the industry problems to data science problems and get into the research. Saying is easy I know. But It is reality. You should accept it and move on.
How can you help the company with your thoughts and work? You are the guide to yourself and your team.
Update 2nd September: Go through the very basics of SQL. Having some knowledge in SQL is a plus.
What is SQL?
As I said earlier every one knew this just I am reminding you.
All the best. I hope It is informative though it is long.
Remember one Mantra: Patience, Practice, Stay Positive…..
0 Comments