问题
I am trying to get, for each individual row, the name of the column with the max/min value up to N-values.
Given something like this:
a b c d e
1.2 2 0.1 0.8 0.01
2.1 1.1 3.2 4.6 3.4
0.2 1.9 8.8 0.3 1.3
3.3 7.8 0.12 3.2 1.4
I can get the max with idxmax(axis=1) and so on the min with idxmin(axis=1) but this only works for the top-max and bottom-min, not generalizable for N-values.
I want to get, if called with N=2:
a b c d e Max1 Max2 Min1 Min2
1.2 2.0 0.1 0.8 0.1 b a c e
2.1 1.1 3.2 4.6 3.4 d d b a
0.2 1.9 8.8 0.3 1.3 c b a d
3.3 7.8 0.1 3.2 1.4 b a c e
I know I can always get the row data, calculate the N-th value and map to a list of columns-names by index, just wondering a better, more elegant way if possible.
回答1:
You can use nlargest and nsmallest:
In [11]: res = df.apply(lambda x: pd.Series(np.concatenate([x.nlargest(2).index.values, x.nsmallest(2).index.values])), axis=1)
In [12]: res
Out[12]:
0 1 2 3
0 b a e c
1 d e b a
2 c b a d
3 b a c e
In [13]: df[["Max1", "Max2", "Min1", "Min2"]] = res
In [14]: df
Out[14]:
a b c d e Max1 Max2 Min1 Min2
0 1.2 2.0 0.10 0.8 0.01 b a e c
1 2.1 1.1 3.20 4.6 3.40 d e b a
2 0.2 1.9 8.80 0.3 1.30 c b a d
3 3.3 7.8 0.12 3.2 1.40 b a c e
回答2:
If the order of the largest/smallest and second largest/smallest values don't matter, then you can use np.argpartition.
N = 2 # Number of min/max values
u = np.argpartition(df, axis=1, kth=N).values
v = df.columns.values[u].reshape(u.shape)
maxdf = pd.DataFrame(v[:,-N:]).rename(columns=lambda x: f'Max{x+1}')
mindf = pd.DataFrame(v[:,:N]).rename(columns=lambda x: f'Min{x+1}')
pd.concat([df, maxdf, mindf], axis=1)
a b c d e Max1 Max2 Min1 Min2
0 1.2 2.0 0.10 0.8 0.01 b a e c
1 2.1 1.1 3.20 4.6 3.40 d e b a
2 0.2 1.9 8.80 0.3 1.30 b c a d
3 3.3 7.8 0.12 3.2 1.40 a b c e
来源：https://stackoverflow.com/questions/54895311/get-column-names-for-the-n-max-min-values-per-row-in-pandas