How to get a sample dataset from large datafram while keeping the proportion of categorical variables in a column

Question

I have a large DataFrame with 2 million observations. For my further analysis, I intend to use a relatively smaller sample (around 15-20% of the original DataFrame) drawn from t