How to sort a dataframe based on multi-column and drop rows that do not sort

Question

For example, if I had a dataframe that looks like

A	B
1	8
4	9
2	5
6	2

Is there a easier way to sort the two rows where Col-A is in ascending order and B in descending order and any row that B cannot be ordered in descending be dropped? The second row should be dropped because 9 is greater than 5.

I am expecting a table that looks like

A	B
1	8
2	5
6	2

Answer 1

You can sort_values, then compute a cummin, drop the rows that violate the cummin:

tmp = df.sort_values(['A','B'], ascending=[True, False])

m = tmp['B'].le(tmp['B'].cummin())

out = tmp[m]

# or to keep the original DataFrame
out = df.loc[m[m].index]

Output:

   A  B
0  1  8
2  2  5
3  6  2