我遵循以下文档:https://github.com/deepmipt/DeepPavlov/blob/master/docs/components/classifiers.rst#id53
我的代码如下:
import os
from deeppavlov import build_model, configs
os.environ["KERAS_BACKEND"] = "tensorflow"
CONFIG_PATH = configs.classifiers.intents_dstc2_big
model = build_model(CONFIG_PATH, download=True)
print(model(["Hello"]))我期望得到如下输出:
"goals": {"pricerange": "cheap"},
"db_result": null,
"dialog-acts": [{"slots": [["pricerange", "cheap"]], "act": "inform"}]}但是,我得到的数字数组如下所示:
[[0.004440320190042257, 0.0035526982974261045, 0.003814868861809373, 0.004386670421808958, 0.0026496422942727804, 0.004122086800634861, 0.004859328735619783, 0.005762884858995676, 0.006169301923364401, 0.9743947386741638, 0.005218957085162401, 0.004720163065940142, 0.006856555584818125, 0.0047727120108902454, 0.008368589915335178, 0.011183635331690311, 0.007578883320093155, 0.005414197687059641, 0.008248056285083294, 0.005105976946651936, 0.005934832151979208, 0.005890967790037394, 0.005130860488861799, 0.005532102193683386, 0.005490032024681568, 0.0046647703275084496, 0.004590084310621023, 0.004707065410912037]]如何正确显示或使用输出?
发布于 2019-01-08 13:07:18
intents_dstc2_big模型并没有为您提供正确的DSTC2输出,相反,它根据原始数据集中的act和slot值来识别发声的意图。
例如,
"goals": {"food": "dontcare", "pricerange": "cheap", "area": "south"},
"db_result": null,
"dialog-acts": [{"slots": [], "act": "thankyou"}, {"slots": [], "act": "bye"}]}此消息包含两个意图(thankyou, bye)
为了获得意图方面的输出,您应该稍微更改一下配置。
import os
from deeppavlov import build_model, configs, train_model
from deeppavlov.core.common.file import read_json
os.environ["KERAS_BACKEND"] = "tensorflow"
model_config = read_json(configs.classifiers.intents_dstc2_big)
model_config['chainer']['out'] = ['y_pred_labels']
model = build_model(model_config, download=True)
print(model(["thank you good bye"]))更多输出选项,您可以在in the configuration file中找到。
请让我知道这是否足够有帮助。
https://stackoverflow.com/questions/53983616
复制相似问题