pandas：快速自定义聚合

Question

我有一个必须重新采样的时间索引数据：

interval = pd.Timedelta(1/8, "s")
resampled_df = df[["T", "N"]].resample(interval).max()

它工作得非常快，但我需要自定义聚合函数（极端）而不是max

def extreme_agg(array_like):
    # return max or min - which absolute value is greater
    return max(array_like.max(), array_like.min(), key=abs)

interval = pd.Timedelta(1/8, "s")
resampled_df = df[["T", "N"]].resample(interval).apply(extreme_agg)

我也试过了

resampled_df = df[["T", "N"]].resample(interval).agg(extreme_agg)

但是这两种方式都非常缓慢。 你知道如何让它更快吗？

或者有没有我的extreme_agg 的快速等价物？

Answer 1

您可以使用更改this function 处理DataFrame.xs 选择的最小值和最大值，首先是min 和max 的聚合值：

np.random.seed(2021)

N = 10000
df = pd.DataFrame({'T':np.random.randint(100, size=N),
                   'N':np.random.randint(100, size=N)},
                  index=pd.timedelta_range(0, freq='100ms', periods=N)).sub(50)
# print (df)
    

def npwhere(df):
    interval = pd.Timedelta(1/8, "s")
    resampled_df = df[["T", "N"]].resample(interval).agg(['max','min'])
    amax = resampled_df.xs('max', axis=1, level=1)
    amin = resampled_df.xs('min', axis=1, level=1)
    return pd.DataFrame(np.where(-amin > amax, amin, amax), 
                                index=resampled_df.index,
                                columns=['T','N'])

resampled_df = npwhere(df)
print (resampled_df.head(10))

def extreme_agg(array_like):
    # return max or min - which absolute value is greater
    return max(array_like.max(), array_like.min(), key=abs)

interval = pd.Timedelta(1/8, "s")
resampled_df1 = df[["T", "N"]].resample(interval).agg(extreme_agg)
print (resampled_df1.head(10))

print (resampled_df.equals(resampled_df1))
True

---

In [206]: %timeit npwhere(df)
12.4 ms ± 46.7 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)

In [207]: %timeit df[["T", "N"]].resample(interval).agg(lambda x: max(x, key = abs))
306 ms ± 4.47 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

In [208]: %timeit df[["T", "N"]].resample(interval).agg(extreme_agg)
2.29 s ± 14 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

Answer 2

这应该可行：

resampled_df = df[["T", "N"]].resample(interval).agg(lambda x: max(x, key = abs))