Question

import pickle

def compare_pred_with_true_values(weights):
    for vect in weights:
        if vect[1] <= 0.5:
            vect[1] = 0
        else:
            vect[1] = 1
    return weights

def counter(weights): 
    count_0_knowing_0 = 0
    count_1_knowing_1 = 0
    count_0_knowing_1 = 0
    count_1_knowing_0 = 0

    for vect in weights:
        if int(vect[0])==0 and vect[0]==vect[1]:
            count_0_knowing_0 += 1
        elif int(vect[0])==1 and vect[0]==vect[1]:
            count_1_knowing_1 += 1
        elif int(vect[0])==1 and vect[0]!=vect[1]:
            count_0_knowing_1 += 1
        else: 
            count_1_knowing_0 +=1

    json = {"count_0_knowing_0": count_0_knowing_0,
            "count_1_knowing_1": count_1_knowing_1,
            "count_0_knowing_1": count_0_knowing_1,
            "count_1_knowing_0": count_1_knowing_0}
    return json


if __name__ == "__main__":
    weights = pickle.load(open("weights_extension.pkl", "rb"))
    weights = [[vect[0], vect[1]] for vect in weights]

    weights_copy = compare_pred_with_true_values(weights)
    json = counter(weights_copy)
    print(json)

weights只是形式为[[0, 0.0013], [1, 0.578], ..., [0, 0.0012]]的列表，输出为{"count_0_knowing_0": 4283, "count_1_knowing_1": 39717, "count_0_knowing_1": 1283, "count_1_knowing_0": 320}。该代码用于查看“相对于真实值的预测有多好？”

该代码最初是用于测试的，但是现在我需要将其插入我的主代码中，但这远非最佳。我不知道我们是否可以找到可以完成相同工作的python库。使用Scikit学习还是scipy？

我们如何扩展该代码，使其可以与多种类型的标签一起使用？此处适用于标签0和1，但是我们可以扩展它使其适用于标签{-n, .., -2, -1, 0, 1, 2, 3, 4, ..., m}吗？

Answer 1

counter()函数可以这样编写：

首先，像这样创建json变量：

json = {'count_{!s}_knowing_{!s}'.format(a, b): 0
        for a in range(2) for b in range(2)}

然后，引用这样的变量：

count_0_knowing_0

成为...

 json['count_0_knowing_0']

然后最后return json，而无需再次创建json变量。

相对于真实值的预测有多好？

1 个答案: