How can I match length of DataFrame when comparing indexes?

df[df.index > df[df.a == 5].index] (shape of df is dynamic)

Example:

df = pd.DataFrame([[0, 10], [5, 10], [0, 10], [5, 10], [0, 10], [0, 10]], columns=["a", "b"])

m = df.index > df[df.a == 5].index
df.loc[m, 'b'] -= np.arange(1, m.sum()   1)

Desired result:

   a   b
0  0  10
1  5  9
2  0  9
3  5  8
4  0  8
5  0  8

CodePudding user response：

The error happens because the length of df[df.a == 5] does not match the length of df, so following expression:

df.index > df[df.a == 5].index

is invalid. Both sides has to match in length or one side must be broadcastable, but if the lengths are 5 and 2, that's not possible.

It seems you want group rows according to the position "a" equals 5 and subtract the group number from "b". In that case, you could use eq cumsum instead:

df['b'] -= df['a'].eq(5).cumsum()

Output:

   a   b
0  0  10
1  5   9
2  0   9
3  5   8
4  0   8
5  0   8