stats.py 文件源码

def despike(df, window=31, l=6):
    """
    Remove outliers from the columns of :class:`DataFrame` by
    comparing the absolute deviation from the windowed median to the
    windowed robust standard deviation (see :func:`robust_std`). Use a
    centered window of length *window* (must be odd). Replace values
    that are *l* robust standard deviations from the absolute
    difference from the median with the median.

    Reference: Hampel F. R., "The influence curve and its role in
    robust estimation," Journal of the American Statistical
    Association, 69, 382-393, 1974.
    """
    if window % 2 == 0:
        raise ValueError('window length must be odd')
    df_rolling = df.rolling(window, center=True)
    df_rolling_median = df_rolling.median()
    df_robust_std = df_rolling.apply(robust_std)
    I = (df - df_rolling_median).abs() > l * df_robust_std
    df_despike = df.copy()
    df_despike[I] = df_rolling_median
    return df_despike.iloc[(window-1):-(window-1)]