2021-07-07

阿尔茨海默病分类技术挑战赛批量部署准确率计算

1.将批量部署的结果下载到本地。
结果保存在obs://yzz/output/infer-result-2193b707-9738-46e4-8c30-390812cf59ab

发现不能在OBS跟notebook上载文件夹。单个下载文件比较麻烦。(目前没有找到“全选”)。通过下述代码，把文件夹压缩成.zip文件，再下载到本地。

def zipDir(dirpath,outFullName):
    """
    压缩指定文件夹
    :param dirpath: 目标文件夹路径
    :param outFullName: 压缩文件保存路径+xxxx.zip
    :return: 无
    """
    zip = zipfile.ZipFile(outFullName,"w",zipfile.ZIP_DEFLATED)
    for path,dirnames,filenames in os.walk(dirpath):
        # 去掉目标跟路径，只对目标文件夹下边的文件及文件夹进行压缩
        fpath = path.replace(dirpath,'')

        for filename in filenames:
            zip.write(os.path.join(path,filename),os.path.join(fpath,filename))
    zip.close()
zipDir("obs://yzz/output/infer-result-2193b707-9738-46e4-8c30-390812cf59ab", "./result.zip")

得到result.zip,在本地解压

通过下述代码计算准确率

import re
import numpy as np
import pandas as pd
import os


def cal(sub_path):
    with open("%s" % sub_path, "r") as f:
        data = f.readline()
    str = data[13:]
    a = re.split(',', str)
    a[2] = re.split(']', a[2])[0]
    if 'e' in a[0]:
        f = a[0].find('e')
        c, b = a[0][:f].split('.')
        j = int(b) / (pow(10, len(b)))
        a0 = (int(c) + j) * (pow(10, int(a[0][f + 1:])))
    elif '.' in a[0]:
        c, b = a[0].split('.')
        j = int(b) * (pow(10, (len(b) * (-1))))
        a0 = int(c) + j
    else:
        a0 = int(a[0])
    if 'e' in a[1]:
        f = a[1].find('e')
        c, b = a[1][:f].split('.')
        j = int(b) / (pow(10, len(b)))
        a1 = (int(c) + j) * (pow(10, int(a[1][f + 1:])))
    elif '.' in a[1]:
        c, b = a[1].split('.')
        j = int(b) * (pow(10, (len(b) * (-1))))
        a1 = int(c) + j
    else:
        a1 = int(a[1])
    if 'e' in a[2]:
        f = a[2].find('e')
        c, b = a[2][:f].split('.')
        j = int(b) / (pow(10, len(b)))
        a2 = (int(c) + j) * (pow(10, int(a[2][f + 1:])))
    elif '.' in a[2]:
        c, b = a[2].split('.')
        j = int(b) * (pow(10, (len(b) * (-1))))
        a2 = int(c) + j
    else:
        a2 = int(a[2])
    a = [a0, a1, a2]
    return a.index(max(a))


# cal(r"./result\Subject_0001.npy_result.txt")


def cal_train(subject_dir, csv_path):
    df = pd.read_csv(csv_path, index_col=0)
    subjects = os.listdir(subject_dir)
    # print(subjects)

    x = []
    y = []
    for subject in subjects:
        features_path = os.path.join(subject_dir, subject)
        if not os.path.exists(features_path) or not features_path.endswith('txt'):
            continue
        else:
            row = df.loc[subject.split('.')[0][:13]]
            label = int(row['Label'])
            test = cal(features_path)
            x.append(test)
            y.append(label)

    x = np.array(x)
    y = np.array(y)
    return x, y


train_label, original_label = cal_train(r'./result', r'./train_open.csv')
print(train_label)
print(original_label)

#
# def acc(test_label, original_label):
#     num_correct = (test_label == original_label).sum().item()
#     return num_correct / len(test_label)
#
#
# print(acc(train_x, train_y))

num_correct = (train_label == original_label).sum().item()
# print(num_correct)
print(num_correct / len(train_label))
# print(len(train_label), len(original_label))

本文例子结果如下