Free Preview

Click below to get a free account

Course overview

The Machine Learning Interview course has all the questions and answers you need to ace your machine learning interviews.

  • 30 interview questions and answers

  • Key machine learning terms defined

  • Complex concepts visualized

  • Written by a data science hiring manager with 8 years of experience

Each lesson explains how to answer a common interview question.

  • Overcome common data problems 

  • Choose the right algorithm

  • Properly validate your model

  • Tune your model without overfitting

  • Explain the exact math behind model training

You will use the Machine Learning Interview course throughout the data science interview process.

  • Phone screen 

  • Take-home challenge

  • Onsite interview 

Course curriculum

  • 3

    Gradient Descent

    • Question #1: Why would gradient descent fail to converge?

    • Answer #1: Gradient descent is ...

    • Question #2: What would happen if you choose a learning rate that is too small?

    • Answer #2: If you pick a learning rate that is ...

    • Question #3: How would you overcome the problems associated with choosing learning rates?

    • Answer #3: You can use an ...

    • Question #4: What’s the difference between gradient descent and stochastic gradient descent?

    • Answer #4: Gradient descent uses …

  • 4

    Training Loss

    • Question #1: How does linear regression learn?

    • Answer #1: To understand how a model learns, we must ...

    • Question #2: How does logistic regression learn?

    • Answer #2: Logistic regression models use …

    • Question #3: What’s the difference between evaluation and loss metrics?

    • Answer #3: Sometimes the model is validated...

  • 5

    Linear Regression

    • Question #1 : What would you do to determine the cause of coefficients that don’t make sense?

    • Answer #1: You should …

    • Question #2: How would you improve a linear regression model that has coefficients that don’t make sense?

    • Answer #2: Given that a ...

    • Question #3: What method would you choose to validate your model?

    • Answer #3: The simplest form of validation is ...

  • 6

    Imbalanced Classifier

    • Question #1: What problem could class imbalance cause and what techniques would you consider to counteract this?

    • Answer #1: Class imbalance occurs when ...

    • Question #2: Why is 99% accuracy probably not a good thing for this model?

    • Answer #2: For classification problems, it’s useful to ...

    • Question #3: What’s a better way to measure model performance for imbalanced classes?

    • Answer #3: First, you can look at ...

    • Question #4: How would evaluate a model for which both precision and recall are important?

    • Answer #4: When it's desirable to account for both ...

  • 7

    Decision Threshold

    • Question #1: How would you choose a threshold for classification?

    • Answer #1: The default threshold for classification...

    • Question #2: How would you change the threshold to account for mispredictions of different dollar values?

    • Answer #2: The best threshold...

    • Question #3: How would you change the model to account for outcomes of different dollar values?

    • Answer #3: If they both have the same predicted probability ...

  • 8

    Algorithm Choice

    • Question #1: How would you choose what type of model to build?

    • Answer #1: The most straightforward algorithm selection ...

    • Question #2: What algorithms would you choose to build this model?

    • Answer #2: There are dozens of popular classifiers to try ...

    • Question #3: Next time you build a classifier on a different data set will the same algorithm produce the best results?

    • Answer # 3: The only way to find out ...

  • 9

    Data Problems

    • Question #1: How would you handle missing data?

    • Answer #1: Missing data occurs when...

    • Question #2: How would you handle outliers?

    • Answer #2: Outliers are ...

    • Question #3: How would you handle features on different scales?

    • Answer #3: Some machine learning models ...

    • Question #4: How would you handle multicollinearity?

    • Answer #4: Multicollinearity is when ...

  • 10

    Curse of Dimensionality

    • Question #1: What’s the curse of dimensionality and why is it a problem?

    • Answer #1: The curse of dimensionality refers to ...

    • Question #2: How would you handle high dimensional data?

    • Answer #2: Before moving forward with modeling ...

    • Question #3: What are some common dimensionality reduction techniques?

    • Answer #3: Below are some common dimensionality reduction techniques …

  • 11


    • Please tell me a bit more about yourself

    • Next Steps


  • I'm not looking for a new job yet. Should I sign up now?

    You get access to the content for a full year from the day you sign up. The best time to start preparing for a new job is when you don’t think you need one. Things change fast and nothing’s worse than suddenly finding yourself unprepared to be on the job market. Think of interview preparation as job insurance -- you hope you don’t need to use it, but if do, you’ll sure be glad you had it.

  • Who made this content?

    I founded Decode Data Science after nearly a decade of experience as a data scientist at top tech companies. The content is based on patterns mined from hundreds of data science interviews conducted over the past 5 years.

  • I already know my stuff pretty well. Why can’t I just review a few good blog posts?

    Interviewers may overlook mistakes in other areas, but they expect candidates to know machine learning thoroughly. Too many candidates treat machine learning models as a black box -- data goes in and predictions come out! This is the most dangerous type of data scientist. They can make wrong predictions with high confidence and managers avoid hiring them at all costs!

Full Access

Upgrade to a premium account

For a limited time, you can get the full course with a one-time purchase of $19 (usually $29). Save $10 by going premium today.