Python PandasNumPy：当行值是行值的倍数时重置滚动平均值，最小尺寸

您可以try :

def rolling_mean(col, start=1, multiple=2, min_count=4):
    curr = start
    num_obs = 0
    acc = 0

    for v in col:
        acc += v
        num_obs += 1

        if num_obs < min_count:
            yield curr
            continue

        m = acc / num_obs
        if v >= multiple * m:
            curr *= -1
            num_obs = 0
            acc = 0

        yield curr


df["new_A"] = list(rolling_mean(df["A"]))
df["new_B"] = list(rolling_mean(df["B"]))

print(df)

打印:

       A    B  new_A  new_B
0   0.10  0.1      1      1
1   0.10  0.1      1      1
2   0.15  0.4      1      1
3   0.10  0.1      1      1
4   0.10  0.8      1     -1
5   0.70  0.1     -1     -1
6   0.10  0.1     -1     -1
7   0.10  0.1     -1     -1
8   0.50  0.1     -1     -1
9   1.00  0.1      1     -1
10  0.10  0.9      1      1
11  0.10  0.1      1      1

You can speed-up the computation using numba:

from numba import njit


@njit
def rolling_mean_numba(col, start=1, multiple=2, min_count=4):
    curr = start
    num_obs = 0
    acc = 0

    out = np.empty_like(col, dtype=np.int8)

    for i, v in enumerate(col):
        acc += v
        num_obs += 1

        if num_obs < min_count:
            out[i] = curr
            continue

        m = acc / num_obs
        if v >= multiple * m:
            curr *= -1
            num_obs = 0
            acc = 0

        out[i] = curr

    return out

df["new_A"] = rolling_mean_numba(df["A"].values)
df["new_B"] = rolling_mean_numba(df["B"].values)

print(df)