stats.py 文件源码

def compute_percentile_from_distribution(counter, percentile):
    """ Takes a Counter object (or value:frequency dict) and computes a single percentile.
    Uses Type 7 interpolation from:
      Hyndman, R.J.; Fan, Y. (1996). "Sample Quantiles in Statistical Packages".
    """
    assert 0 <= percentile <= 100

    n = np.sum(counter.values())
    h = (n-1)*(percentile/100.0)
    lower_value = None

    cum_sum = 0
    for value, freq in sorted(counter.items()):
        cum_sum += freq
        if cum_sum > np.floor(h) and lower_value is None:
            lower_value = value
        if cum_sum > np.ceil(h):
            return lower_value + (h-np.floor(h)) * (value-lower_value)

# Test for compute_percentile_from_distribution()
#def test_percentile(x, p):
#    c = Counter()
#    for xi in x:
#        c[xi] += 1
#    my_res = np.array([compute_percentile_from_distribution(c, p_i) for p_i in p], dtype=float)
#    numpy_res = np.percentile(x, p)
#    print np.sum(np.abs(numpy_res - my_res))