# Microsoft Interview Experience | Set 54 (For SDE)

**Round 1 (45 mins):**

Tell me about yourself.

1. Given an array of strings with **R**, **G** & **B** as characters. **Sort** the array such that all R comes first, followed by Gs & followed by Bs. Do it in **O(n)** time complexity. (**Paper Code**)

For eg. array is:

** I/P: G G R B R B B G R B R**

** O/P: R R R R G G G B B B B**

Hey geek! It's time to become a success story instead of reading them. Check out our most renowned **DSA Self Paced Course**, now at a student-friendly price and become industry ready. And if you are looking for a more complete interview preparation resource, check out **Complete Interview Preparation Course**** **that will prepare you for the SDE role of your dreams!

Feeling prepared enough for your interview? Test your skills with our **Test Series** that will help you prepare for top companies like **Amazon, Microsoft, TCS, Wipro, Google** and many more!

**Hint:** Same problem as sorting strings of 0s, 1s & 2s.

Done using the same logic as that of sorting 0s, 1s & 2s in no time. Interviewer was happy. *:) happy

2. Can you do the same above program using **Singly Linked List (SLL)** & **Doubly Linked List (DLL)** in **O(n)** time complexity.

Done it using DLL comfortably.

But using SLL I got stuck as we’ve to keep track of PREV pointer every time. He wanted it to be done in a SINGLE PASS if possible. Was not able to do this for SLL.

**Round 2 (1.30 hrs):**

Tell me about the projects you’ve worked on.

1. **Matrix Multiplication** of 2 matrices. Both of the matrices are **Distributed** across a cluster of nodes. And matrices are of huge size (**BigData**). How can you multiply these matrices efficiently. He wanted to know the approach only but not the actual code.

(He asked me this Distributed problem esp. as I’ve worked on **Apache Spark & Hadoop Distributed architecture** in my project. So he was very keen to check my knowledge in the Distributed & BigData domain.)

**Hint:** Think in terms of **MapReduce** jobs.

**My approach**: As we know that in matrix multiplication, ith row is multiplied with jth column, (i+1)th row is multiplied with (j+1)th column, & so on till (n-1)th row with (m-1)th column.

**Map** each ith row to jth column, (i+1)th row to (j+1)th column & so on till (n-1)th row to (m-1)th column.

The idea is to make **Keys as ROW** index **& Values as Column** index.

While doing the **Reduce **operation, Multiply the respective Rows & Columns using above Mapped Rows & Columns. B’coz generally in the Reduce operation of MapReduce jobs, Spark/Hadoop brings all of the mapped data into one Server to do the computation & stores the result. So in the Reduce operation, Spark/Hadoop will bring only the mapped Row & Column while doing the actual multiplication. We need NOT to bring all Rows & Columns data at once.

This was my approach. One can think of the other approach also & share their better approach too.

Laterally he asked me about MapReduce jobs in further details & asked me about Hadoop & Distributed architecture (NameNode, DataNode, etc.). Some questions on Scalability.

2. Given an **N-ary tree** with **thousands of nodes** in it. **Pair** (Join) the **Leaf nodes** which do **NOT SHARE** the common path. i.e. Two Leaves can be Paired only if they do NOT have **Intersecting path**.

For example,

A / | \ B C D / / | \ E F G H

**Leaf nodes**: E, F, G, H & D

**Possible Pairs in O/Ps:**

a) (E-F), (G-H) or

b) (E-G), (F-H) or

c) (E-H), (F-G) or

d) (E-D), (F-G) or

e) (E-D), (G-H) or

f) (E-D), (F-H) or

g) (D-H), (F-G) or

h) (D-G), (F-H) or

i) (D-F), (G-H)

**Note: **If we pair(join) say, **(E-F)** then we can NOT pair any of the **(D-G)** or **(D-H)** as they SHARE the COMMON path from A to C.

i.e. E-B-A-C-F —> (E-F) pair

D-A-C-G —> (D-G) pair

D-A-C-H —> (D-H) pair

So the above case is **NOT** possible.

Also if there are **EVEN** number of Childs, then exactly n/2 pairs are possible. In case of **ODD** number of Childs, 1 of the Childs will be left **Unpaired**.

I’ve told him using **BFS** as it traverses each Edge exactly **ONCE**. The idea was while doing BFS traversal, Pair Childs of a Visiting node if both of their Childs are Leaf nodes.

If the Visiting node has only **ONE** Child (Left Child or Right Child only) then store this Child in the TEMPORARY array (Array of Pointers to the Childs).

Once the BFS traversal ends, Pair the remaining Childs in the **TEMPORARY** array one by one.

Again he was NOT satisfied as BFS works **LEVEL-WISE**. So he told me what if some of the nodes have **SKEWED** sub-tree with thousands of nodes.

I got stuck. 🙁

He told me to apply **Mathematical Induction**.

If we can prove for the Base Case that at least **2K** Childs can be Paired. Then we pair & prove it for **2K+2, 2k+4, , , , , , ,2K+n** Childs too.

I was Blank in MI concepts.

Was **NOT** able to solve this problem. 🙁

He was so impressed *:) happy with my 1st problem’s approach using MapReduce. He was a little bit satisfied with the 2nd problem’s approach. So I’ve been forwarded to 3rd round.

**Lunch time.** Soooooooo Good. Food was so good. Really liked it.

**Round 3 (50 mins):**

He had quickly looked at my CV. Then he asked me to tell something about the projects that I’ve worked on & working on. As I’ve worked mostly on **Machine Learning** & **NLP** domain, he asked me few questions about **NLP**. How will you Scale your current project.

1. Given a sentence, Reverse it **Word-by-Word**.

For example,

**I/P: “Microsoft Azure is one of the best Cloud solutions”**

** O/P: “solutions Cloud best the of one is Azure Microsoft”**

So easy. Given on GFG. 🙂

Done it in no time.

He asked me do it in a **SINGLE PASS**. As the solution requires **2 PASSES** of the sentence. (1st for reversing each individual word & 2nd for reversing WHOLE sentence).

Got stuck. 🙁 Taken a long time to think about this. Couldn’t able to do this. Finally he asked me another question.

2. Detect a **Loop** in a **Singly Linked List**. *:) happy

Given the GFG approach. He was Happy.

3. **Remove** the Loop in above SLL.

Again given the same GFG solution. 🙂

I’ve been told to wait for the 4th & Final round.

**Round 4 (Design Round):**

What do you know about **Cloud Computing**.

Do you know anything about **Microsoft Azure**.

Design a **WindowsBox** just like **DropBox** Cloud storage.

Provide all the services **IaaS**, **PaaS** & **SaaS** using **Service-Oriented Architecture (SOA)**.

Have given very Low-level design. NOT up to his expectations. As I’ve NOT worked on Cloud Technology earlier. 🙁

NOT able to perform well in this round. 🙁

**Advise**:

Prepare for the **Design** questions **& Scalability**.

One can go through this link: https://www.quora.com/How-should-I-prepare-system-design-questions-for-Google-Facebook-Interview

Watch out this video. Its very good.

They look for the basics of **Data Structures & Algorithms**. (esp. Linked Lists & Tree)

Lastly, GeeksForGeeks is the best source for all. GFG really helped me a lot in the preparation.

Thanks 🙂

If you like GeeksforGeeks and would like to contribute, you can also write an article and mail your article to contribute@geeksforgeeks.org. See your article appearing on the GeeksforGeeks main page and help other Geeks.