import json
import argparse
import sys
import numpy as np
import jieba
from nltk.translate.bleu_score import sentence_bleu, SmoothingFunction
from nltk import ngrams



def load_file(filename):
    data = []
    with open(filename, "r") as f:
        for line in f.readlines():
            data.append(json.loads(line))
        f.close()
    return data

def proline(line):
    return " ".join([w for w in jieba.cut("".join(line.strip().split()))])


def compute(golden_file, pred_file, return_dict=True):
    golden_data = load_file(golden_file)
    pred_data = load_file(pred_file)

    if len(golden_data) != len(pred_data):
        raise RuntimeError("Wrong Predictions")

    num = 0
    for g, p in zip(golden_data, pred_data):
        if isinstance(g["label"], str):
            l = int(g["label"].strip())
        elif isinstance(g["label"], int):
            l = g["label"]
        else:
            raise Exception("Data type error")

        if isinstance(p["label"], str):
            p = int(p["label"].strip())            
        elif isinstance(p["label"], int):
            p = p["label"]            
        else:
            raise Exception("Data type error")
        if l == p:
            num += 1

    return {'accuracy': float(num)/len(golden_data)}

def main():
    argv = sys.argv
    print("预测结果：{}, 测试集: {}".format(argv[1], argv[2]))
    print(compute(argv[2], argv[1]))


if __name__ == '__main__':
    main()