Manipulating a large comma separated file with python

I have a large .csv file that contains headers in the first row, these header (column) names are not unique, each column with the same header is a sample replicate. The current .csv file is 5 GBs, I need it to be downsampled to under 1 GB. The way I would like this done is to take at random, 25% of each of the unique header replicates. As an example, consider an array: cell_type1 cell_type1 cell_type1 cell_type1 cell_type2 cell_type2 cell_type2 cell_type2 ... gene1 gene2 gene3 gene4 .... I would like this to be downsampled to: cell_type1 cell_type2 gene1 gene2 gene3 gene4 .... I would like the code that does this, and I would like the output of the code.