TikTok Data Science OA, 4 Questions Done in 28 Minutes: Metrics, Table Merges, Feature Preprocessing, Random Forest Threshold Tuning

I took the TikTok DS OA today. The four questions had some substance, but three were familiar, so I finished in about half an hour.

TikTok's OA isn't actually hard — most people just don't prepare enough and get a bit nervous. The questions are fairly standard; here are all four and my approaches.

Q1: Compute three metrics

Compute three metrics from the given data:

Metric	How to compute
Average rating	Mean of the rating column, rounded to two decimals
Second-language share	Share of records where second_language isn't no, as a percentage with two decimals
Success rate	Combine all rides_*.csv files and compute the share of status equal to success, as a percentage with two decimals

Return the results in the required format.

Q2: Combine multiple tables

Merge several tables into one. The main task is merging as described and making sure field values follow the required conventions. Do the merge and field extraction as specified, then save the result.

Q3: Driver data preprocessing

Preprocess the driver data — impute missing values, encode categorical variables, standardize numeric features and convert the label:

1. Missing values: fill age with the training set's mean.
2. Categorical variables: for sec_language, car_model and similar, build mapping dictionaries from the training set; unknown values get a new code.
3. Numeric features: standardize columns like net_xxx using the training set's mean and standard deviation.
4. Label: convert driver_class to numbers (A → 0, B → 1).

Avoid data leakage: every transformation must use statistics from the training set only, then be applied unchanged to the validation and test sets.

Q4: Binary classifier + threshold tuning

Train a binary classifier on the cleaned data, with class B as the positive class:

1. Fit a random forest on the training set.
2. Tune the decision threshold on the validation set: maximize recall while keeping precision high.
3. Apply the best threshold from validation to binarize the test set's predicted probabilities.
4. Save and return the predictions in the required format.

Good luck, and may all your tests pass!