ここだ1つのnumpyのベースのソリューション -
def grouby_max(df):
arr = df[['id','series']].values
n = arr.shape[0]-1
idx = (arr[:,0]*(arr[:,1].max()+1) + arr[:,1]).argsort()
sidx = np.append(np.nonzero(arr[idx[1:],0] > arr[idx[:-1],0])[0],n)
return df.iloc[idx[sidx]]
ランタイムテスト -
In [201]: # Setup input
...: N = 100 # Number of groups
...: data = np.random.randint(11,999999,(10000,5))
...: data[:,0] = np.sort(np.random.randint(1,N+1,(data.shape[0])))
...: df = pd.DataFrame(data, columns=[['id','series','s1','s2','s3']])
...:
In [202]: %timeit df.loc[df.groupby('id')['series'].idxmax()]
100 loops, best of 3: 15.8 ms per loop #@EdChum's soln
In [203]: %timeit df.sort_values(by="series", ascending=False).groupby("id", as_index=False).first()
100 loops, best of 3: 4.52 ms per loop #@jezrael's soln
In [204]: %timeit grouby_max(df)
100 loops, best of 3: 1.96 ms per loop