outliers.py 文件源码

def flag_outliers(series, iqr_multiplier=1.5):
    """Use Tukey's boxplot criterion for outlier identification.
    """
    top_quartile_cutoff = np.percentile(series.get_values(), 75)
    bottom_quartile_cutoff = np.percentile(series.get_values(), 25)
    # Compute interquartile range
    iqr = top_quartile_cutoff - bottom_quartile_cutoff
    top_outlier_cutoff = top_quartile_cutoff + iqr * iqr_multiplier
    bottom_outlier_cutoff = bottom_quartile_cutoff - iqr * iqr_multiplier
    return series[(series < bottom_outlier_cutoff) | (series > top_outlier_cutoff)]


# In[ ]: