Scribd
Upload
26 views6 pages

Google Data Science Interview Questions

Collection of all interview questions asked in Google Data science team

Uploaded by

utsav421
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
26 views6 pages

Google Data Science Interview Questions

Collection of all interview questions asked in Google Data science team

Uploaded by

utsav421
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 6

GOOGLE

DATA SCIENCE
INTERVIEW
QUESTINOS
WHAT ARE THE ASSUMPTIONS OF ERROR IN LINEAR REGRESSION
Independence of Errors - The error terms should be
independent of each other. This means that there should be
no correlation between consecutive errors (no
autocorrelation). This assumption is often tested using the
Durbin-Watson test in time series data.

Homoscedasticity - The variance of the error terms should


remain constant across all levels of the independent
variables. If the variance of the errors increases or
decreases (heteroscedasticity), it can lead to inefficiencies
in the estimation of coefficients.

Normality of Errors - The error terms should be normally


distributed, especially for hypothesis testing (i.e., t-tests for
coefficients). This assumption is crucial when constructing
confidence intervals and p-values.

@karunt
WHAT IS THE FUNCTION OF P-VALUES IN HIGH DIMENSIONAL LINEAR REGRESSION?
P-values are used to test the null hypothesis that a
specific regression coefficient (for a predictor) is
zero. A low p-value suggests that the predictor is
statistically significant, meaning it likely has an
effect on the response variable.

In high-dimensional models, testing many


predictors increases the chance of false positives
(Type I errors), meaning some predictors might
appear significant purely by chance. Traditional p-
values need to be adjusted (e.g., Bonferroni
correction, FDR methods) to account for this.

High-dimensional data often has strong


multicollinearity, meaning many predictors are
highly correlated. This can cause unstable estimates
of regression coefficients, leading to unreliable p-
values. So make sure to remove correlated features

@karunt
LET’S SAY YOU HAVE A CATEGORICAL VARIABLE WITH THOUSANDS OF
DISTINCT VALUES, HOW WOULD YOU ENCODE IT?
Leave-One-Out Encoding A variation of target
encoding, leave-one-out encoding, computes the
target mean for each category, but excludes the
current observation to avoid target leakage.
Pros: Reduces target leakage, works well with high-
cardinality features.
Cons: Computationally more expensive than simple
target encoding.

Embedding-Based Encoding - For extremely high


cardinality categorical features, embedding-based
approaches are often effective. This technique involves
learning a dense vector representation of each
category, and you typically use a NN to get the
embedding.
Pros: Captures latent structure
Cons: More complex to implement

@karunt
DESCRIBE TO ME HOW PCA WORKS
PCA is a dimensionality reduction technique used if
you think you have correlated features, noisy data, or
to visualize data in fewer dimensions.

To perform PCA you normalize features, calculate


covariance matrix (to indicate if variable
increase/decrease when another variables does),
find eigenvectors (directions where data is most
spread out) or eigenvalues (amount of
variance/spread)

PCA does assume variables are linearly related, so


cant be used for non-linear relationships. Also, new
dimension are linear combination of older dimension
so interpretation does become harder.

@karunt
WAS THIS HELPFUL?
Be sure to save it so you
can come back to it later!

@karunt

You might also like