Prepare for and practice interview questions from Lawrence Berkeley Lab across topics like Statistics, SQL, Machine Learning and more.

Lawrence Berkeley Lab Interview Questions

Lawrence Berkeley Lab Interview Guides

Machine Learning

How would you encode a categorical variable with thousands of distinct values

Encoding Categorical Features

Interpret coefficients of logistic regression for categorical and boolean variables

Coefficients of Logistic Regression

In which case would you use a bagging algorithm versus a boosting algorithm

Bagging vs Boosting

What are some reasons why measuring bias would be important in building this particular model

Food Delivery Times

How to model merchant acquisition in a new market?

Merchant Acquisition

Product Sense & Metrics

How would you measure the success of Facebook Groups

Group Success

How would you measure the success of the Instagram TV product

Instagram TV Success

What data points and metrics would you look at to decide if amateur creators are doing worse and only superstars do well

Amateur Performance

What are the top five metrics you would start tracking to understand the health of Google Docs

Docs Metrics

What are the most important metrics for WhatsApp

WhatsApp Metrics

Data Structures & Algorithms

Given an integer N, write a function that returns all of the prime numbers up to N

Prime to N

This problem involves identifying duplicate numbers in a list of integers. The function should return a list of the duplicate numbers.

Find Duplicate Numbers in a List

Write a function digit_accumulator that returns the sum of every digit in a string

Digit Accumulator

Write a function that tests whether a string of brackets is balanced.

The Brackets Problem

Write a function to return the value of the nearest node that is a parent to both nodes.

Nearest Common Ancestor

Behavioral

Describing a data project and its challenges

Hurdles In Data Projects

Strategically resolving misaligned expectations with stakeholders for a successful project outcome

Stakeholder Communication

How would you answer when an Interviewer asks why you applied to their company?

Why Do You Want to Work With Us

What do you tell an interviewer when they ask you what your strengths and weaknesses are?

Your Strengths and Weaknesses

Making data-driven insights actionable for those without technical expertise

Simple Explanations

Business Case

How to boost presence in high-demand city areas

Incentive Scheme

How would you initially filter users who post fake comments on Amazon and scale the effort with minimal manual intervention

Flagging Users

Diagnose why the pricing algorithm is under-pricing a consumer product

Underpricing Algorithm

How would you create a policy for refunds with regards to balancing customer sentiment and goodwill versus revenue tradeoffs?

Food Delivery Refund Policy

Demystifying data for non-technical users through visualization and clear communication

Accessible Data

Write a SQL query to select the 2nd highest salary in the engineering department.

Note: If more than one person shares the highest salary, the query should select the next highest salary.

Example:

Input:

<code>employees</code> table

<table>
<thead>
<tr>
<th>Column</th>
<th>Type</th>
</tr>
</thead>

<tbody>
<tr>
<td><code>id</code></td>
<td>INTEGER</td>
</tr>

<tr>
<td><code>first_name</code></td>
<td>VARCHAR</td>
</tr>

<tr>
<td><code>last_name</code></td>
<td>VARCHAR</td>
</tr>

<tr>
<td><code>salary</code></td>
<td>INTEGER</td>
</tr>

<tr>
<td><code>department_id</code></td>
<td>INTEGER</td>
</tr>
</tbody>
</table>
<code>departments</code> table

<table>
<thead>
<tr>
<th>Column</th>
<th>Type</th>
</tr>
</thead>

<tbody>
<tr>
<td><code>id</code></td>
<td>INTEGER</td>
</tr>

<tr>
<td><code>name</code></td>
<td>VARCHAR</td>
</tr>
</tbody>
</table>
Output:

<table>
<thead>
<tr>
<th>Column</th>
<th>Type</th>
</tr>
</thead>

<tbody>
<tr>
<td><code>salary</code></td>
<td>INTEGER</td>
</tr>
</tbody>
</table>

Select the 2nd highest salary in the engineering department

2nd Highest Salary

Tell me about a data project that didn’t go the way you expected. What did you set out to do, what surprised you, and how did you handle it?

Let’s say that your company is running a standard control and variant AB test on a feature to increase conversion rates on the landing page. The PM checks the results and finds a .04 p-value.

How would you assess the validity of the result?

How would you assess the validity of the result?

Experiment Validity

A/B Testing

Given an integer <code>N</code>, write a function that returns a list of all of the prime numbers up to <code>N</code>.

Note: Return an empty list there are no prime numbers less than or equal to <code>N</code>.

Example:

Input:

<pre tabindex="0" class="chroma"><code>N = 3
</code></pre>

Output:

<pre tabindex="0" class="chroma"><code>def prime_numbers(N) -&gt; [2,3]
</code></pre>

Let’s say you have a categorical variable with thousands of distinct values, how would you encode it?

Which method can be used to extract communities from large networks, that does not require a pre-determined number of clusters like K-means?

Which method can be used to extract communities from large networks, that does not require a pre-determined number of clusters like K-means?

Let’s say you work in air traffic control. You are given the table below containing information on flights between two cities.

Write a query to find out how much time, in minutes (rounded down), each plane spent in the air each day.

Note: Both cities are in the same time zone, so you do not need to worry about converting time zones.

<code>flights</code> table

Example:

Input:

<table>
<thead>
<tr>
<th>Columns</th>
<th>Type</th>
</tr>
</thead>

<tbody>
<tr>
<td><code>id</code></td>
<td>INTEGER</td>
</tr>

<tr>
<td><code>destination_location</code></td>
<td>VARCHAR</td>
</tr>

<tr>
<td><code>source_location</code></td>
<td>VARCHAR</td>
</tr>

<tr>
<td><code>plane_id</code></td>
<td>INTEGER</td>
</tr>

<tr>
<td><code>flight_start</code></td>
<td>DATETIME</td>
</tr>

<tr>
<td><code>flight_end</code></td>
<td>DATETIME</td>
</tr>
</tbody>
</table>
Output:

<table>
<thead>
<tr>
<th>Columns</th>
<th>Type</th>
</tr>
</thead>

<tbody>
<tr>
<td><code>plane_id</code></td>
<td>INTEGER</td>
</tr>

<tr>
<td><code>calendar_day</code></td>
<td>VARCHAR</td>
</tr>

<tr>
<td><code>time_in_min</code></td>
<td>INTEGER</td>
</tr>
</tbody>
</table>
Please give <code>calendar_day</code> in the format <code>YYYY-MM-DD</code>

Write a query to find how much time each plane spent in the air each day in minutes rounded down

Total Time in Flight

Let’s say you want to test the close friends feature on Instagram Stories.

How would you make a control group and test group to account for network effects?

What could be a potential risk when Facebook segments the metrics of the user interface change by market or demographic groups?

What could be a potential risk when Facebook segments the metrics of the user interface change by market or demographic groups?

Network Experiment Design I

What is the primary metric that Facebook should monitor if it decides to make the user interface of its posting feature more like Instagram's?

What is the primary metric that Facebook should monitor if it decides to make the user interface of its posting feature more like Instagram's?

Network Experiment Design II

If Facebook changes its user interface to mimic Instagram's, which adverse effect should they anticipate and monitor?

If Facebook changes its user interface to mimic Instagram's, which adverse effect should they anticipate and monitor?

Network Experiment Design III

How would you make a control group and test group to account for network effects

Network Experiment Design

Let’s say we’re comparing two machine learning algorithms. In which case would you use a bagging algorithm versus a boosting algorithm? 

Give an example of the tradeoffs between the two.

In machine learning, when would you use a bagging algorithm over a boosting algorithm?

In machine learning, when would you use a bagging algorithm over a boosting algorithm?

Bagging vs. Boosting

Let’s say you’re a Product Data Scientist at Instagram. How would you measure the success of the Instagram TV product?

What is the primary goal of Instagram TV as a product feature?

What is the primary goal of Instagram TV as a product feature?

How would you measure the success of Facebook Groups?

Considering the purpose of Facebook Groups as a platform for connecting users through shared interests or real-life relationships, which metrics would be appropriate to measure the success of these groups?

Considering the purpose of Facebook Groups as a platform for connecting users through shared interests or real-life relationships, which metrics would be appropriate to measure the success of these groups?

You’re given a string that may contain the characters <code>{</code>, <code>}</code>, <code>[</code>, <code>]</code>, <code>(</code>, and <code>)</code>.

Task: Verify that the string is balanced. A balanced string is one where every opening character, <code>{</code>, <code>[</code>, or <code>(</code>, has a corresponding closing character, <code>}</code>, <code>]</code>, or <code>)</code>.

Write a function called <code>is_balanced(string: str) -&gt; bool</code> which verifies the balance of a string.

Example:

<pre tabindex="0" class="chroma"><code>is_balanced(&#39;(())[]{}&#39;) -&gt; True
</code></pre>

<pre tabindex="0" class="chroma"><code>is_balanced(&#39;{([(){}])()}&#39;) -&gt; True
</code></pre>

<pre tabindex="0" class="chroma"><code>is_balanced(&#39;{}[]())&#39;) -&gt; False
</code></pre>

<hr/>

How would you explain what a p-value is to someone who is not technical?

What does a p-value in a statistical test represent?

What does a p-value in a statistical test represent?

P-value to a Layman I

In a statistical test, how does a low p-value (less than 0.05) influence our decision about the null hypothesis?

In a statistical test, how does a low p-value (less than 0.05) influence our decision about the null hypothesis?

P-value to a Layman II

Explain what a p-value is to someone who is not technical

P-value to a Layman

Statistics

You are a data scientist at YouTube focused on creators.

A PM comes to you worried that amateur video creators could do well before but now it seems like only “superstars” do well.

What data points and metrics would you look at to decide if this is true or not?

You work at a ride-sharing company and are given a table containing information about the rides of users.

Write a query to get the duration (in minutes) of each ride longer than two hours. Sort the results by duration in descending order.

<code>rides</code> table:

<table>
<thead>
<tr>
<th>Columns</th>
<th>Type</th>
</tr>
</thead>

<tbody>
<tr>
<td><code>id</code></td>
<td>INTEGER</td>
</tr>

<tr>
<td><code>passenger_user_id</code></td>
<td>INTEGER</td>
</tr>

<tr>
<td><code>start_dt</code></td>
<td>DATETIME</td>
</tr>

<tr>
<td><code>end_dt</code></td>
<td>DATETIME</td>
</tr>

<tr>
<td><code>distance</code></td>
<td>FLOAT</td>
</tr>

<tr>
<td><code>ride_region</code></td>
<td>VARCHAR</td>
</tr>

<tr>
<td><code>is_completed</code>	</td>
<td>INTEGER</td>
</tr>
</tbody>
</table>
Output:

<table>
<thead>
<tr>
<th>Column	</th>
<th>Type</th>
</tr>
</thead>

<tbody>
<tr>
<td><code>duration_minutes</code></td>
<td>INTEGER</td>
</tr>
</tbody>
</table>

Write a query to get the duration in minutes of each ride longer than two hours, sorted descending

Time Difference

You are given a binary tree of unique positive numbers. Each node in the binary tree is represented as a dictionary with the following keys:

<ul>
<li><code>data</code>: integer value stored in the node</li>
<li><code>left</code>: reference to the left child (or <code>None</code>)</li>
<li><code>right</code>: reference to the right child (or <code>None</code>)</li>
</ul>

<pre tabindex="0" class="chroma"><code>node = {
 &#34;data&#34;: 6,
 &#34;left&#34;: {
 &#34;data&#34;: 3,
 &#34;left&#34;: {...},
 &#34;right&#34;: {...}
 },
 &#34;right&#34;: {
 &#34;data&#34;: 9,
 &#34;left&#34;: {...},
 &#34;right&#34;: {...}
 }
}
</code></pre>

Given two node values as input (<code>value1</code> and <code>value2</code>), write a function to return the value of the nearest common ancestor (lowest node in the tree that has both nodes as descendants).

Note: If one of the nodes doesn’t exist in the tree, return <code>-1</code>.

Example:

Input:

<pre tabindex="0" class="chroma"><code># Diagram of the binary tree
&#39;&#39;&#39;
 6 
 / \ 
 3 9 
 / \
 2 11
 / \
 5 8
&#39;&#39;&#39;
value1 = 8
value2 = 2
</code></pre>

Output:

<pre tabindex="0" class="chroma"><code>common_ancestor(root,value1,value2) -&gt; 3
</code></pre>

Explanation:

<ul>
<li>Ancestors of <code>8</code>: <code>11 → 3 → 6</code></li>
<li>Ancestors of <code>2</code>: <code>3 → 6</code></li>
<li>Common ancestors: <code>3</code> and <code>6</code></li>
<li>The nearest common ancestor is <code>3</code>.</li>
</ul>

You’re a data scientist at YouTube. Leadership wants to revisit how the platform measures success, not for a specific new feature, but for the product as a whole given its current state.

How would you define and measure success for YouTube? Identify your primary and secondary metrics, explain your rationale for each, and discuss how you would choose the right thresholds for any engagement-based metrics you propose.

The role of A/B testing in measuring the success rate of an analytics experiment

Success Measurement

Given a list of integers, identify all the duplicate values in the list. Assume that the list can contain both positive and negative numbers, and the order of the list does not matter. A number is considered a duplicate if it appears more than once in the list. Return a list of the duplicate numbers.

Example 1:

Input:

<pre tabindex="0" class="chroma"><code>nums = [1, 2, 3, 1, 2, 3]
</code></pre>

Output:

<pre tabindex="0" class="chroma"><code>find_duplicates(nums) -&gt; [1, 2, 3]
</code></pre>

The numbers 1, 2, and 3 all appear more than once in the list, so they are considered duplicates.

Example 2:

Input:

<pre tabindex="0" class="chroma"><code>nums = [1, -1, 2, 3, 3, -1]
</code></pre>

Output:

<pre tabindex="0" class="chroma"><code>find_duplicates(nums) -&gt; [-1, 3]
</code></pre>

The numbers -1 and 3 both appear more than once in the list, so they are considered duplicates. Note that the order of the output does not matter.

Example 3:

Input:

<pre tabindex="0" class="chroma"><code>nums = [1, 2, 3, 4, 5]
</code></pre>

Output:

<pre tabindex="0" class="chroma"><code>find_duplicates(nums) -&gt; []
</code></pre>

None of the numbers in the list appear more than once, so there are no duplicates.

Suppose there exists a new airline named Jetco that flies domestically across North America. Jetco recently had a study commissioned that tested the boarding time of every airline, and it came out that Jetco had the fastest average boarding times of any airline.

What factors could have biased this result, and what would you look into?

What could potentially bias the average boarding time in favor of an airline that only flies domestically?

What could potentially bias the average boarding time in favor of an airline that only flies domestically?

Boarding Times Bias

A new airline came out as the fastest average boarding times compared to other airlines.  What factors could have biased this result and what would you look into?

You’re AI engineer working on an energy forecasting system at Siemens that helps utilities predict short-term electricity demand to balance grid load and reduce outages. You train a recurrent neural network on long historical time series (weather, calendar effects, local demand signals) to forecast energy demand.

Early in training, the model appears to learn normally, but after several epochs the loss becomes highly unstable, oscillating wildly before turning into NaNs. At the same time, training slows dramatically and the model fails to recover even after restarting from recent checkpoints.

You’ve verified that the data pipeline is correct and labels are well-aligned with inputs. No changes were made to the dataset, but increasing the sequence length or learning rate makes the problem significantly worse.

How would you investigate the cause of this training instability? What signals would you look for during debugging, and how would those findings guide the changes you make to the model or training setup?

How would you investigate the cause of this training instability?

Training Instability in Neural Networks

Given a table of <code>cars</code> with columns <code>id</code> and <code>make</code>, write a query that outputs a random manufacturer’s name with an equal probability of selecting any name.

Input:

<code>cars</code> table

<table>
<thead>
<tr>
<th><code>id</code></th>
<th><code>make</code></th>
</tr>
</thead>

<tbody>
<tr>
<td>1</td>
<td>Ford</td>
</tr>

<tr>
<td>2</td>
<td>Toyota</td>
</tr>

<tr>
<td>3</td>
<td>Toyota</td>
</tr>

<tr>
<td>4</td>
<td>Honda</td>
</tr>

<tr>
<td>5</td>
<td>Honda</td>
</tr>

<tr>
<td>6</td>
<td>Honda</td>
</tr>
</tbody>
</table>
Output:

<table>
<thead>
<tr>
<th>Column</th>
<th>Type</th>
</tr>
</thead>

<tbody>
<tr>
<td><code>make</code></td>
<td>Text</td>
</tr>
</tbody>
</table>

Write a query that outputs a random manufacturer's name with an equal probability of selecting any name.

Uniform Car Maker

Let’s say that you work as a data scientist at a food delivery company.

You are tasked with building a model to predict food preparation times at a restaurant from the time when a customer sends in the order to when the meal is ready for the driver to pick up.

What are some reasons why measuring bias would be important in building this particular model?

What are the assumptions of linear regression?

Which assumption of the residuals of the standard linear regression model can not be overcome by increasing the sample size?

Regression assumptions

What are the assumptions of linear regression?

Assumptions of Linear Regression

How would you interpret coefficients of logistic regression for categorical and boolean variables?

Why is one-hot encoding recommended for categorical variables in logistic regression models?

Why is one-hot encoding recommended for categorical variables in logistic regression models?

You are given a <code>string</code> that represents some floating-point number. Write a function, <code>digit_accumulator</code>, that returns the sum of every digit in the <code>string</code>.

Example:

Input:

<pre tabindex="0" class="chroma"><code>s = &#34;123.0045&#34;
</code></pre>

Output:

<pre tabindex="0" class="chroma"><code>def digit_accumulator(s) -&gt; 15

Since 
1 + 2 + 3 + 0 + 0 + 4 + 5 = 15
</code></pre>

How would you handle the data preparation for building a machine learning model using imbalanced data?

Addressing imbalanced data in machine learning through carefully prepared techniques.

Data Preparation for Imbalanced Data

You’ve been tasked with building a classification machine-learning model to predict whether a transaction is either fraud or not fraud for a credit card company. You have ten years of historical data on transactions, including a flag for whether a transaction was manually identified as fraud.

Describe how you might go about building a fraud detection model for credit card transactions. Be sure to mention the possible model types, discuss the bias-variance tradeoff in model development, and address any complexities that arise from the class imbalance.

Input features in the data include:

<ul>
<li>Transaction amount</li>
<li>Merchant category for the transaction</li>
<li>Zip code for the merchant</li>
<li>Zip code for the billing address</li>
<li>Average transaction amount for the account over the past six months</li>
</ul>

Output feature:

<ul>
<li>0/1 flag for fraud (0 = not fraud, 1 = fraud)</li>
</ul>

Note: Fraudulent transactions are (thankfully) a very small percentage of all historical transactions. Assume fraudulent transactions are 0.01% of historical data.While building the model to perform the classification, you need to consider the bias/variance tradeoff, and take into account the fact that there is a class imbalance (very few of the observations are “fraud”).

Bias variance tradeoff and class imbalance in finance

Bias - Variance Tradeoff and Class Imbalance in Finance

A robot has been designed to navigate a two-dimensional 4x4 matrix by only moving forward or turning right when blocked by a wall of the matrix. Its starting position is in the top left corner of the matrix, denoted by (0,0), and the robot’s final destination is the bottom right corner.

Determine the full path of the robot before it hits the final destination or starts repeating the path.

The ‘walls’ of the matrix are either one of the four borders of the map or any block found within.

A ‘Block’ is donated by having 1 in the cell corresponding to it where an empty cell is having ‘0’.

Example 1:

<pre tabindex="0" class="chroma"><code>#Input
robot_map = [
 [0,0,0,0],[0,0,1,0],[0,1,0,0],[0,0,0,0]]

#Output
robot_path(robot_map) -&gt; [(0,0),(0,1),(0,2),(0,3),(1,3),(2,3),(3,3)]
</code></pre>

As seen in these diagrams, the robot’s path starts with this sequence of steps:

<ol>
<li>Goes forward three times and hits the wall. Turns right.</li>
<li>Goes forward three times (orientation has changed), 3. Reaches the final destination, and stops.</li>
</ol>

<img src="/s3/questions/a251a8da-2906-40de-9fe8-7f15f367fa34.png" alt="image"/>

Example 2:

<pre tabindex="0" class="chroma"><code>#Input
robot_map = [
 [0,0,1,0],[0,0,0,0],[0,1,0,0],[0,0,0,0]]

#Output
robot_path(robot_map) -&gt; [(0,0),(0,1),(1,1),(1,0)]

</code></pre>

Here, the robot moves forward one space, but hits the wall. After turning right, another wall is hit, starting a loop.

<img src="/s3/questions/34e81379-4514-472c-84a0-f31a4721f5b5.png" alt="image"/>

Determine the full path of the robot before it hits the final destination or starts repeating the path.

Walking Robot

The Fibonacci sequence is a series of numbers in which each number is the sum of the two preceding ones, usually starting with 0 and 1. It is often used in algorithm examples, and is defined by the following formula: F(n) = F(n-1) + F(n-2), with F(0) = 0 and F(1) = 1.

Your task is to implement the Fibonacci algorithm in three different methods:
1. Recursively
2. Iteratively
3. Using Memoization

Example 1:

Input:

<pre tabindex="0" class="chroma"><code>n = 5
</code></pre>

Output:

<pre tabindex="0" class="chroma"><code>fibonacci(n) -&gt; 5
</code></pre>

Example 2:

Input:

<pre tabindex="0" class="chroma"><code>n = 10
</code></pre>

Output:

<pre tabindex="0" class="chroma"><code>fibonacci(n) -&gt; 55
</code></pre>

The Fibonacci sequence starts as follows: 0, 1, 1, 2, 3, 5, 8, 13, 21, 34, 55…

This question requires the implementation of the Fibonacci sequence using three different methods: recursively, iteratively, and using memoization.

Implementing the Fibonacci Sequence in Three Different Methods

Imagine you are asked to build a machine learning model to decide new loan approvals for a
financial firm. You ask the data department in the company for a subset of data to get started
working on the problem. The data includes different features about applicants such as age,
occupation, zip code, height, number of children, favorite color, etc. You decide to build
multiple machine learning models to test out different ideas before settling on the best one.

How would you explain the bias-variance tradeoff with regards to
building and choosing a model to use?

Bias vs. Variance Tradeoff

Let’s say we’re managing a production line for assembling electric scooters, with 10 stations where each has a different average processing time per unit (for example, Station 1 takes 5 minutes, Station 2 takes 3 minutes, and so on).

How would you determine the right metric to use for balancing worker allocation across the stations, and how would you use it to figure out how many workers each station needs?

How would you decide on a metric and approach for worker allocation across an uneven production line?

Calculate Moving Average	SQL	Easy
Predict Customer Churn	Machine Learning	Medium
A/B Test Significance	Statistics	Medium
Optimize Query Performance	SQL	Hard
Feature Importance Analysis	Machine Learning	Medium
Clean Missing Data	Python	Easy
Neural Network Architecture	Deep Learning	Hard
Calculate Cohort Retention	SQL	Medium
Bayesian Probability	Statistics	Easy
Recommend Similar Products	Machine Learning	Hard

Lawrence Berkeley Lab Interview Questions

Lawrence Berkeley Lab Interview Guides

Lawrence Berkeley Lab Interview Questions

Challenge

Lawrence Berkeley Lab Salaries by Position

Discussion & Interview Experiences

Discussion & Interview Experiences