Facebook Data Scientist | October 2020

Company: Facebook

Position: Data Scientist

Location: nan

Level: Mid-level

Outcome: Techncal Screen Rejected

How was the interview process? What was it like?

Strong DS and ML skills, strong SQL and python coding

What technical questions were asked?

Python / Pandas, AB Testing & Statistics, Probability, SQL / Analytics, Modeling and ML Knowledge

What was one of your solutions?

1. PCA is a dimensional reduction technique. Each PC explain certain % of variance. The way to choose top K PC's are by using scree plot and elbow method. Another method is by sorting PC's by variability being explained and select the top k which meets the variability threshold say 90% for example.

2. Bagging and Boosting ensembling algorithms. They both are combination of multiple models. In bagging (bootstrap aggregation) the output is going to be either average of the all the models (regression) or by majority (classification). The example is Random forest. 

3. Precision is about identifying TP/(TP+FP) and recall is nothing but sensitivity which is = TP/(TP+FN). The optimal trade-off between precision and recall is identified by measuring the F-score.

4. AR - autoregression - actually considering the previous lags of the data and predict the future timestamps. We take t-1, t-2, t-3 etc and perform regression to predict for t+1. AR is actually capturing the Trend. MA - moving average - we are modeling for seasonality. The order of AR and MA is based on the auto-correlation plot.

5. Law of large numbers says that as the sample size tends to infinity the mean of sample will be equal to the mean of population.

CLT says when the sample size tends to infinity the sampling mean will be normally distributed