Selecting different no. of random rows from pandas dataframe based on counts

I have a data frame with 10000 rows,2 columns(text and category) and 3 categories (pos,neg and neutral). I want to randomly select 5000 rows of positive class and 2500 rows each for negative class and neutral class. I am able to do 5000 rows of each class. But I am wondering how to select different no. of rows for different classes. Any help would be appreciated.