Two Sigma data scientist reports vary in structure. Several describe an online assessment and multiple technical rounds, while one describes a shorter four-round loop. Statistics, probability and open-ended data analysis come up repeatedly.
$190K
Avg. Base Comp
$300K
Avg. Total Comp
Not reported
Typical Rounds
Not reported
Process Length
Candidates describe the Two Sigma data scientist interview as more statistical than a typical coding screen, though the format differs from report to report. Three candidates mention an online assessment, and the problems they recall include linear interpolation and linear regression. One of them had to code a regression loss function from scratch and said that practicing LeetCode left them underprepared for it.
The live rounds also vary. One candidate described three separate technical interviews on data analysis, coding and algorithms, and statistics or domain knowledge, and found the probability section harder than the coding. That candidate also had a question that came straight from the green book. Another candidate described a four-round loop: a recruiter screen, coding, statistics and probability, and a behavioral round with a hiring manager.
The open-ended prompts are where several reports focus on communication. Candidates were asked how they would predict house rental prices, judge whether a house price is reasonable, handle time-series data, and work through a Citi Bike problem. One report included an A/B test where click-through rate rose while revenue fell. Interviewers seemed to care about stated assumptions, model choices and validation as much as the final answer.
Five of the six reports ended without an offer, and one candidate was still waiting for a reply. If you prepare, practice explaining regression from first principles, work through probability problems out loud, and ask clarifying questions when a prompt is vague.
Synthesized from 6 candidate reports by our editorial team.
Had an interview recently?
Share your experience. Unlock the full guide.
Real interview reports from people who went through the Two Sigma process.
The interview had multiple rounds, mixing technical data science questions with problem-solving. I felt confident in coding but started sweating when questions became more open-ended and required explaining my assumptions.
Questions asked: I had four rounds: a recruiter screen, a technical coding interview, a statistics/probability round, and a behavioral interview with a hiring manager.
The coding round used Python and focused on data manipulation. One question was to find the top three most frequent values in a large dataset without sorting the entire array, and another involved writing a function to detect duplicate events in a stream.
The statistics round was the hardest. I was asked how I would evaluate an imbalanced fraud-detection model, when I would choose precision-recall over ROC-AUC, and to calculate a conditional probability problem involving drawing cards without replacement.
The most interesting case prompt was about an A/B test where click-through rate improved but revenue fell. They asked me to explain what could cause the discrepancy, what additional metrics I'd check, and whether I'd ship the experiment.
The behavioral round focused on explaining a past project, defending modeling choices, and describing a time I changed my mind after seeing new data. The biggest surprise was how much they cared about clearly communicating assumptions, not just getting the math right.
Share your own interview experience to unlock all reports, or subscribe for full access.
Sourced from candidate reports and verified by our team.
Topics based on recent interview experiences.
Featured question at Two Sigma
Given two sorted lists, write a function to merge them into one sorted list.
| Question | |
|---|---|
| Random SQL Sample | |
| Maximum Profit | |
| Get Top N Frequent Words | |
| Poker Pair | |
| Bicycle Rental Data Pipeline | |
| Summing Numeric Strings | |
| Find Square Root | |
| Data Stream Median | |
| Bernoulli Sample | |
| Why Do You Want to Work With Us | |
| LRU Cache 1 | |
| Sum Numbers As Strings | |
| Using APIs for Downstream Tasks | |
| Empty Neighborhoods | |
| 2nd Highest Salary | |
| Rolling Bank Transactions | |
| Closest SAT Scores | |
| Top Three Salaries | |
| Subscription Overlap | |
| Employee Salaries | |
| Experiment Validity | |
| First to Six | |
| 500 Cards | |
| First Touch Attribution | |
| Top 5 Turnover Risk | |
| String Shift | |
| Compute Deviation | |
| Prime to N | |
| Last Transaction |
Synthesized from candidate reports. Individual experiences may vary.
One candidate began with a recruiter screen as the first of four rounds. Another described a separate HR or recruiter round that was purely behavioral, and a third said an HR round came after the technical interviews. Where this conversation falls appears to vary, so be ready to discuss your background and motivation early or late.
Three candidates describe an online assessment. One recalled a linear interpolator, a regression of daily temperature by town, and efficient regression fitting. Another recalled Python and pandas problems with some statistics. A third had three hours on HackerRank and coded an ML regression loss function from scratch, which was not the pure algorithm practice they expected.
Two candidates describe three technical rounds covering data analysis, coding and algorithms, and statistics or domain knowledge. One said each ran an hour, that probability was harder than the coding, and that a question came from the green book. The other recalled probability brain teasers with few hints. A different candidate's four-round loop included Python questions on top-three frequent values and duplicate events in a stream, and a statistics round on imbalanced fraud models and card-drawing conditional probability.
Several candidates describe open-ended analysis. Prompts included predicting house rental prices, judging whether a house price is reasonable, processing time-series data, correlation questions and a Citi Bike problem. One candidate discussed a housing prediction problem with an interviewer for about an hour, and another was given a dataset to analyze in code. A third was asked about an A/B test where click-through rate rose but revenue fell. Reports stress explaining assumptions, variable selection and validation.
Candidates describe a late conversation with hiring managers. One said it involved two or three managers from matched teams plus a senior manager, mixing open-ended technical questions with behavioral discussion. Another recalled senior team members and hiring managers. The candidate with the four-round loop was asked to explain a past project, defend modeling choices and describe a time they changed their mind after seeing new data.
