Snips-nlu intent在快速启动后返回空的intent json

Question

我使用的是windows10机器上的snips-nlu库。

我能够使用pip install snips-nlu成功安装它，然后使用snips-nlu download en安装该语言。我的example.py文件如下所示：

engine = SnipsNLUEngine(resources=load_resources('snips_nlu_en')) # this works for me
# engine = SnipsNLUEngine(resources=load_resources('en')) # this throws errors about en not being installed even after installing them

with io.open("dataset.json") as f:
    dataset = json.load(f)

engine.fit(dataset)
parsing = engine.parse("Are there holidays next week in botswana?")
print(json.dumps(parsing, indent=2))

执行时，我得到以下输出：

{
  "input": "Are there holidays next week in botswana?",
  "intent": {
    "intentName": null,
    "probability": 0.4607009138432084
  },
  "slots": []
}

我的yaml文件如下所示：

# ask for country holidays
---
type: intent
name: holidayQuestion
slots:
  - name: country
    entity: snips/country
  - name: date
    entity: snips/datetime
utterances:
  - What holidays is it [date] in [country]?
  - Is it a holiday in [country]?
  - Is it a holiday [date] in [country]?
  - Is it a [country] holiday [date]?
  - Holiday [date] in [country]?
  - Holiday [date] [country]?
  - Holiday [country]?

我用snips-nlu generate-dataset en dataset.yaml > dataset.json把它转换成json文件。

当我执行我的python文件时，它会报错：

FileNotFoundError: No data found for the 'snips/country' builtin entity in language 'en'. You must download the corresponding resources by running 'python -m snips_nlu download-entity snips/country en' before you can use this builtin entity.

我在windows cmd上使用管理员权限执行它，但它仍然返回空插槽和没有intentName。有人能帮我解释一下发生了什么事吗？我已经尝试卸载pip、python和所有包，重新启动计算机并清除与python/pip相关的环境变量。但没有骰子。

Answer 1

首先，为了避免使用管理员权限运行命令，您应该在virtual environment中运行所有内容。

在virtualenv中运行后，在您的用例中使用snips-nlu之前，您必须安装英文资源和snips/country内置实体：

python -m snips_nlu download en 
python -m snips_nlu download-entity snips/country en

这通常会下载并链接相应的资源，以便snips-nlu库可以在默认位置访问它们。如果您由于权限错误而未在virtualenv中运行，则这两个命令可能会失败。

在此之后，您可以尝试使用CLI处理数据集：

> python -m snips_nlu train dataset.json nlu_engine
Create and train the engine...
Persisting the engine...
Saved the trained engine to nlu_engine

然后：

> python -m snips_nlu parse nlu_engine
Enter a query (type 'q' to quit): Are there holidays next week in botswana?
{
  "input": "Are there holidays next week in botswana?",
  "intent": {
    "intentName": "holidayQuestion",
    "probability": 0.7427372869733698
  },
  "slots": [
    {
      "entity": "snips/datetime",
      "range": {
        "end": 28,
        "start": 19
      },
      "rawValue": "next week",
      "slotName": "date",
      "value": {
        "grain": "Week",
        "kind": "InstantTime",
        "precision": "Exact",
        "value": "2019-12-02 00:00:00 +01:00"
      }
    },
    {
      "entity": "snips/country",
      "range": {
        "end": 40,
        "start": 32
      },
      "rawValue": "botswana",
      "slotName": "country",
      "value": {
        "kind": "Country",
        "value": "Botswana"
      }
    }
  ]
}

你能分享你正在使用的snips-nlu版本吗？如果可能，您应该更新到最新的可用版本。