__init__.py 文件源码

def five_group_stats(group):
    sales = np.array(group['Demanda_uni_equil'].values)
    samana = group['Semana'].values
    max_index = np.argmax(samana)
    returns = group['Dev_proxima'].mean()
    #this is signature on when slaes happens

    sorted_samana_index = np.argsort(samana)
    sorted_sales = sales[sorted_samana_index]

    signature = np.sum([ math.pow(2,s-3) for s in samana])
    kurtosis = fillna_and_inf(scipy.stats.kurtosis(sorted_sales))
    hmean = fillna_and_inf(scipy.stats.hmean(np.where(sales <0, 0.1, sales)))
    entropy = fillna_and_inf(scipy.stats.entropy(sales))
    std = fillna_and_inf(np.std(sales))
    N = len(sales)
    ci = fillna_and_inf(calculate_ci(std, N))
    corr = fillna_and_inf(scipy.stats.pearsonr(range(N), sorted_sales)[0])

    autocorr_list = np.correlate(sorted_sales, sorted_sales, mode='same')
    mean_autocorr = fillna_and_inf(np.mean(autocorr_list))

    mean = np.mean(sales)

    mean_corss_points_count = 0
    if N > 1:
        high_than_mean = mean < sorted_sales[0]
        for i in range(1,N):
            if (high_than_mean and mean > sorted_sales[i]) or (not high_than_mean and mean > sorted_sales[i]):
                mean_corss_points_count += mean_corss_points_count
            high_than_mean = mean < sorted_sales[i]

    return mean, N, std, np.median(sales), sales[max_index], samana[max_index], \
           returns, signature, kurtosis, hmean, entropy, ci, corr, mean_autocorr, mean_corss_points_count