How To Get Machine Learning Job

How To Get A Machine Learning Task In 6 Months

Follow this study guide to be successful in a Machine Learning interview at tiptop tech companies.

I take given and took many Motorcar Learning Engineering (MLE) interviews at companies like Google, Twitter, Lyft, Snapchat and others. Since I accept been on both sides of the interviewing process for MLE, I can tell you that unlike software applied science, the interview process for MLE is cluttered. By chaotic, I mean every company has their idea of what skills a candidate needs to have; thus, every visitor has a different hiring process. However, I will requite yous a template to follow so that past the end you will have the cognition to pass about any MLE interview at FAANG and other tech companies.

Photo past Christina @ wocintechchat.com on Unsplash

Ane practiced matter that is happening inside the last year or and then in the field now is that most companies are (slowly) converging on a somewhat similar process:

[Some people consider the phone call with the recruiter the starting time step, but in all honesty, unless you lied on your resume or say something offensive, this is merely a chat almost the next steps]

Telephone Interview — This usually would have either ii forms: a complete software engineering interview or a mix of software engineering and a few ML questions
Onsite Interview:

Software Technology/ML Coding Interview
ML Theory Questions — I have seen this interview split into 2 parts: ML theory and ML coding
ML Arrangement Design
Company Specific — I put this hither because nearly companies tend to have i of the to a higher place interviews repeated or something specific based on the job (e.g. Facebook has a System Design interview, note this is non an ML Organisation Blueprint interview)
Hiring Manager (behavioral) — I failed an interview for a company by being unprepared for this round. I aced the technical interviews and was confident I would get an offer. Even so, when I did non, the feedback was that I did not exercise well on the behavioral interview. After that feedback, I spent a lot of time preparing for these questions.

And then we can see that at that place are iii main rounds that you lot will run into in almost any MLE interview:

Software Technology/ML Coding
ML Theory
ML System Pattern

We will go through each of these rounds and I volition provide you with the all-time (IMO) websites/documents that you tin can use to succeed in your next MLE interview in 6 months. Since nosotros accept 6 months we tin build a stiff baseline of knowledge and and so branch out.

Software Technology & ML Coding

Photograph past Florian Olivo on Unsplash

We tin can split this into two sections: pure software engineering (SWE) coding and and so ML coding.

For the SWE coding, the plan would be:

Leetcode : This should not be a strange inclusion every bit information technology is ubiquitous for SWE interviews and for good reason! I would recommend post-obit the below steps in order equally they build on each other:

[2 weeks] https://leetcode.com/explore/learn/ — this is good to refresh your memory on data structures and algorithms
[2 weeks] https://seanprashad.com/leetcode-patterns/ — this is a smashing resources containing many patterns that you would face up in a coding interview
[two weeks] https://leetcode.com/problemset/top-interview-questions/ — you should exist aware of the pop questions that tend to come up in an interview
N.B: While answering these coding questions, you should create a spreadsheet of difficult questions and the algorithm (not the code) that is used. And so go dorsum to the spreadsheet later on on and attempt to solve the questions a second fourth dimension. This will help you sympathize your weaknesses. Once you lot complete the iii links higher up, yous should answer at least ane leetcode question every day (preferably medium and sometimes hard). When it is closer to your interview date, I recommend buying leetcode premium and doing the company questions.

2. [one one/ii months — optional but recommended] Elements of Programming Interviews: Many people recommend Neat the Coding Interview but, I would recommend you purchase Elements of Programming Interviews (EPI). I brand this recommendation because y'all can choose between Java, C++ or Python as your programming linguistic communication and can utilise their coding judge arrangement to exam your lawmaking. This book was hard for me to become through at first and I could non solve nigh of the questions. However, as I continued, I started writing python code efficiently and inverse how I think about solving problems.

For ML Coding, the plan would exist:

[one week] ML Coding Algorithms: I would recommend this site because getting proper ML coding resource became an issue for me when studying for interviews. Most websites are lacking in 2 ways: the code is far too complex than what is required in an interview or does not contain time/space complexity. This website solves both problems.
[one week] Github ML Algorithms: This resource contains near all of the possible ML algorithms you lot will probably run across in an interview. The lawmaking is written in a very modular way, making it is easy to understand the lawmaking.
[ii weeks — optional] Machine Learning Mastery: A popular website containing the nigh popular ML algorithms that the writer builds from scratch. It is optional because in that location are parts of the code that may not be relevant in an interview (eastward.g. data ingestion). If you lot have extra time, you tin browse the code.

ML Theory

I found myself spending hours upon hours trying to sympathize what topics I should know for these theory interviews. After endless hours of research, these are the topics you should know:

"Foundational" ML Topics (topics you lot HAVE to know):

Linear Regression
Logistic Regression
KNN
Tree-based models- decision copse, random forests, bagged/ additional copse, KL Difference, Entropy)
SVM
Clustering (basic ones like Thousand-ways, Hierarchical) & EM
PCA (SVD & Eigen Values)
Naive Bayes
Maximum Likelihood Estimation, Maximum A Priori Estimation

Specialized ML Topics (you lot should know at to the lowest degree Neural Network Architectures; for the other topics, they tin can be reviewed based on your resume and what the job specifies):

Neural Network Architectures: FFNNs, RNNs, LSTMs, CNNs
GANs
NLP
Calculator Vision
Recommendation Algorithms
Information Retrieval
Reinforcement Learning

Misc (these are VERY popular questions which you lot should know):

Activation Functions
Optimization schemes SGD, Adam, RMSProp…
Losses — log-loss, swivel loss, Huber loss, L1/ii loss…
Metrics — Precision, Recall, Intersection over Marriage, F1, RMSE…
Feature Selection
Feature Engineering science

For studying, I will requite two options based on your preferred method of studying (note, yous tin can mix and match if you would like):

Books:

[3/4 weeks] Hands-On Motorcar Learning with Scikit-Learn, Keras, and TensorFlow: This book is an first-class book that covers most of the topics above (other than some specialized topics). It is an easy read with engaging diagrams and code to assistance you understand each topic.
[1/2 weeks] The 100 Page ML Volume: This is some other neat book that covers near of the topics above. Information technology goes into more mathematical particular than the Hands-On ML volume but, the writer does an splendid job of being concise while still being an like shooting fish in a barrel read. Likewise, information technology is free :)

(OR) Video Courses:

[ane week] Udacity's Machine Learning Course: An easy but comprehensive grade on ML. The instructors requite yous an overview of most of all the foundational topics and some of the misc topics. It is an fantabulous refresher if yous are looking for somewhere to showtime.
[two/3 weeks] Coursera'southward Machine Learning Course: This has to be the nearly popular course on the internet for machine learning and for good reason! It goes into more of the mathematical foundations of many of the traditional ML algorithms while still being accessible to anyone. Note, if you desire a summary of the class rather than watching all the videos you can check out these notes.
[4/v weeks] Coursera'due south Deep Learning Specialization: These courses cover most of the neural network architectures above and go into sufficient detail for an interview. I highly recommend this every bit it has almost all the data you need for succeeding in questions on Neural Networks and deep learning. Again, if you lot would like a summary of the course you can read these notes.

Once you lot are finished with the in a higher place resources, you lot probably need questions to practice. Here is a collection of question and answer docs:

https://github.com/andrewekhalel/MLQuestions
https://github.com/IndyNYU/Auto-Learning-Interview/blob/primary/Algorithms %26 Theory
https://www.mle-interviews.com/ml-questions

ML System Design

An ML system pattern question is (nearly) a given for any MLE onsite interview. Since in that location are no courses or books on this as of this writing, we should use blogs (we can complete all the below in 2 weeks)

ML System Design Template For MLE Interviews: A general framework that can exist used for most ML System Design interview questions. There are specific questions and answers that most companies will enquire.
Architecting a Machine Learning Pipeline: A blog for agreement the engineering side for ML System Pattern
Machine Learning Organisation Design Draft PDF: A skillful overview of themes that y'all should include in your response to ML System Design
Motorcar Learning Arrangement Design Weblog: A great resources of links to many visitor blogs which explain how they build their ML systems.

Wow, that was a lot! I never said studying for MLE position would be easy but, if you enjoy ML, this should be enjoyable. A terminal recommendation is to take notes, whether it be online or using the old fashioned pen and paper and don't forget those leetcode questions once a day!