Aller au contenu principal
Version: 1.0.0-beta.6

Packages & Modules

Modules are the skills of Leon, this is thanks to them Leon is able to do things according to what you say. In this section you will have a deeper look into modules.

Packages#

Packages contain an infinity of modules. A package is nothing more than a folder containing modules, you can consider them as a category of modules.

E.g. the Checker package contains modules such as the Is It Down one because this package includes modules related to the "checking" purposes.

The full packages list is available here.

Directory Structure#

Packages are listed in the packages directory. Let's take the Video Downloader package as example.

Note the package name must be lowercase and in English.

videodownloader
├── config
│ ├── config.json
│ └── config.sample.json
├── data
│ ├── answers
│ │ ├── en.json
│ │ └── fr.json
│ ├── db
│ └── expressions
│ ├── en.json
│ └── fr.json
├── test
│ └── youtube.spec.js
├── README.md
├── __init__.py
├── version.txt
└── youtube.py

Each package:

  • Has its own version.
  • Has its own configuration (for each module).
  • Has its own dataset (for each module).
  • Has its own tests (for each module).

Versioning#

  • Leon's packages follow the SemVer.
  • Each new module increases the MINOR version number of a package (e.g. 1.0.0 -> 1.1.0).
  • Each time a MAJOR or MINOR version number of a package is increased, then increase the MINOR one of the whole project (e.g. 1.0.0 -> 2.0.0 | 1.1.0 -> 1.1.0).
  • Each time a PATCH version number of a package is increased, then increase the PATCH version number of the whole project (e.g. 1.0.0 -> 1.0.1 -> 1.0.1).

Modules#

Modules are the skills of Leon. They contain one or several action(s) to be able to accomplish specific job(s).

When Leon understands what you told him, he:

  1. Triggers a module action.
  2. Do the job.
  3. Returns you the output of that execution.

Each module has its own purpose and its own configuration. Do not hesitate to browse the packages list to understand their goals.

Today, modules are written in Python but in the future they could also support other languages thanks to the bridges.

Configuration#

Let's take the Video Downloader package again as example.

{
"youtube": {
"api_key": "YOUR_GOOGLE_API_KEY",
"playlist_id": "PLAYLIST_ID",
"options": {
"synchronization": {
"enabled": true,
"method": "direct",
"email": ""
}
}
}
}

packages/videodownloader/config/config.json

In each package configuration file, you can add new keys/value as much as you want while you are creating a module.

In this example, the keys api_key and playlist_id have been added to the YouTube module configuration. It allows the module to pick the values to request the YouTube API.

To have access to these properties, you can use the utils.config(key) function.

Options are used when it needs interaction between a module and the core. They can be used for the synchronizer for example.

Tip

Do not hesitate to take a look at the other modules to have a better understanding.

Dataset & Translations#

To reply and understand you, Leon needs his dataset and translations. Indeed, his dataset are divided into two parts: expressions and answers.

  • Every module has their own expressions and answers.
  • Each of these dataset has their own translations.
  • Translations are represented by the filename of these dataset, such as en.json, fr.json, etc.

Expressions#

Expressions are the data used to train the Leon's understanding. When you execute the training script, all of the expressions of each module are browsed to generate the classifier.

Expressions are wrapped inside a module action. This is how Leon understands which action he needs to do.

Note that each expression of each module action has its own confidence.

{
"meaningoflife": {
"run": {
"expressions": [
"What is the meaning of life?",
"Tell me what is the meaning of life"
]
}
}
}

E.g. Who Am I module English expressions belonging to the Leon package. These expressions are wrapped inside the run action.

Fallbacks#

Despite the expressions you wrote, it might be possible Leon still does not understand some of them. This is where fallbacks jump in the game.

In the core/langs.json file, you can find the list of the supported languages with several properties:

  • short: the short language code.
  • min_confidence: the minimum confidence of the Leon's comprehension. If the confidence is smaller than the given one, Leon replies you he is not sure about what you said.
  • fallbacks: force the module selection. Use the words key to determine on which words you want Leon pick up a module. And use the package, module and action keys to define which module action should be executed on the given words.

Answers#

Answers are the data used by Leon to provide you results binded with the modules outputs. In addition, answers contains sub properties to have different kind of answers per module.

{
"greeting": {
"default": [
"Hello there!"
],
"morning_good_day": [
"Good morning, I hope your day will be full of joy and productivity!"
],
"morning": [
"Good morning!"
],
"too_late": [
"Hello! I'm feeling grateful you still talk to me, but you should get some sleep now."
]
}
}

E.g. part of the Greeting module English answers belonging to the Leon package.

HTML#

It is possible to use HTML in your answers.

{
"github": {
"list_element": [
"<li>#%rank%. <a href=\"%repository_url%\" target=\"_blank\">%repository_name%</a> created by <a href=\"%author_url%\" target=\"_blank\">%author_username%</a> with %stars_nb% new stars.</li>"
]
}
}

E.g. part of the GitHub module English answers belonging to the Trend package.

Create a Module#

Tip
  • Creating a module is one of the best way to contribute in Leon! Before doing that, please make sure you review this document ❤️
  • For example, you could think of creating a To-Do List module (although this one already exists). Check out the roadmap to see what is in the pipeline.
  • Don't hesitate to open an issue if you have any questions.

Each module is included in a package (e.g. packages/{PACKAGE NAME}/{MODULE NAME}.py).

Steps#

Here are the basics steps to create a module. For those steps, we will take a tweets grabber module as example.

1. Define the Purpose(s)#

  • I want to create a tweets grabber module. When I say or write:
Grab my latest tweets
  • I want Leon tells me my 5 latest tweets with the stats of each.
  • It seems this module does not correspond to any existing package (category). So I create the Twitter package by creating the packages/twitter folder.
  • To do so, I make sure it follows the package directory structure and contains the required files mentioned in that structure.
Tip

If your module is more advanced and must contain multiple purposes, do not hesitate to create several actions.

2. Name Your Module#

  • I choose to name my module Tweets Grabber.

3. Write the Code#

  • To request the Twitter API, I need API credentials. So I set the Twitter API key(s) in the packages/twitter/config/config.json file I previously created in the step 1.
  • In addition, I create the packages/twitter/tweetsgrabber.py file, define my action(s) and I write the code of my module.
  • While I'm writing the code, I edit server/src/query-object.sample.json and from the project root directory I use the following command:
PIPENV_PIPFILE=bridges/python/Pipfile pipenv run python bridges/python/main.py server/src/query-object.sample.json
# It executes my module on the fly

4. Write the Tests#

  • Now that I'm satisfied with my module, I create the packages/twitter/test/tweetsgrabber.spec.js file.
  • I write my module tests in that file.

5. Shortly Explain How To Use#

  • In the packages/twitter/README.md file, I add a short description of the purpose of my module.
  • I briefly explain how to use the module (which kind of sentences can we say, if there is any configuration to do, etc.).

6. Share#

Actions (Module Functions)#

In the module file, you must add an action (function) that will be the entry point of the execution. An action takes the input string (query) and the entities as parameters.

When you have only one action in your module, the usual action name is run:

# Query the "run" action
What is the meaning of life?
#!/usr/bin/env python
# -*- coding:utf-8 -*-
import utils
def run(string, entities):
"""Leon says what's the meaning of life"""
# string: What is the meaning of life?
# entities: There is none here
return utils.output('end', 'meaning_of_life', utils.translate('meaning_of_life'))

When you have several actions in your modules:

# Query the "create_list" action
Create the shopping list
# Query the "add_todo" action
Add potatoes to my shopping list
#!/usr/bin/env python
# -*- coding:utf-8 -*-
import utils
def create_list(string, entities):
"""Leon creates a to-do list"""
# Your awesome code here...
return utils.output('end', 'list_created', utils.translate('list_created'))
def add_todo(string, entities):
"""Leon add todos to a list"""
# Your awesome code here...
return utils.output('end', 'todo_added', utils.translate('todo_added'))
Tip

Don't forget that Leon knows which action he must execute thanks to the expressions you define.

Naming Convention#

  • The module filename must contain only lowercase alphabetic characters and must use the English language.

    E.g. Meaning of Life module filename: meaningoflife.py

  • Actions names must use snakecase (lowercase alphabetic characters and `` only) and must use the English language.

    E.g. To-Do List module actions: create_list, add_todo, complete_todo, etc.

Query Object#

Every time you communicate to Leon, he will creates a temporary query object JSON file with the following properties:

  • lang: short code of the used language.
  • package: used package name.
  • module: used module name.
  • action: used action name.
  • query: your sentence.
  • entities: an array of the entities Leon has extracted from your sentence. An entity can be whatever you define as an entity (custom entity) or it can be a built-in entity such as a date duration, a number, a domain name, etc.

The server/src/query-object.sample.json file is here to let you execute and test the behavior of your module code on the fly during its creation. Edit it according to your module properties.

Entities#

Entities are chunks that Leon extracts from your sentences. These entities are shared to your actions so you can manipulate them to give more sense to your modules.

There are different types of entities that are listed below:

Built-In Entities#

Built-in entities are the ones already included in Leon. Leon extracts entities from queries automatically.

The full list is available here.

Tip

Feel free to see some examples to understand how built-in entities are used. Those are perfect examples:

As you can see, you can iterate over the entities to grab the information you need (domain names, dates, etc.).

Custom Entities#

Custom entities are the ones you define yourself according to specific use cases. You can create your own entities in the translations files located in: packages/{PACKAGE NAME}/data/expressions/{LANG FILE}.json. In that file, custom entities are included in the actions properties at the same level as the expressions.

They are represented by an array of objects:

"entities": [
{
"type": "",
"name": "",
...
}
]

As you can see, a custom entity is made of a type, a name and more depending of its type.

E.g. see the create_list action entities of the To-Do List module.

Custom entities have two types listed below:

Trim Entities#

Trim entities allow you to cut/trim parts of the query to extract only the text you want. This is done thanks to these conditions:

  • { "type": "between", "from": [], "to": [] }
  • { "type": "after", "from": "" }
  • { "type": "after_first", "from": "" }
  • { "type": "after_last", "from": "" }
  • { "type": "before", "to": "" }
  • { "type": "before_first", "to": "" }
  • { "type": "before_last", "to": "" }

To illustrate that, let's say we are creating a To-Do List module. To do so, we will define a custom entity list.

When we have the following queries:

Create a shopping list
Create my shopping list

We want to extract the text shopping to associate it as a list entity. We use the between condition to catch what is between a or my and list:

"entities": [
{
"type": "trim",
"name": "list",
"conditions": [
{
"type": "between",
"from": ["a", "my"],
"to": ["list", "list"]
}
]
}
]

In the module file packages/{PACKAGE NAME}/todolist.py:

def create_list(string, entities):
# string: "Create a shopping list"
# entities: [{'type': 'between', 'start': 9, 'end': 16, 'len': 8, 'accuracy': 1, 'sourceText': 'shopping', 'utteranceText': 'shopping', 'entity': 'list'}]
print('entity name:', entities[0]['entity']) # entity name: list
print('entity value:', entities[0]['sourceText']) # entity value: shopping
Tip

You can take a look at the real To-Do List module of the Calendar package.

Regex Entities#

Regex entities allow you to grab parts of the query via a regular expression.

Let's say we create a Color Picker module. To do so, we will define a regex entity color.

When we have the following query:

I like red and blue colors

We want to extract the strings red and blue to associate it as color entities. We use a regex to catch these colors:

"entities": [
{
"type": "regex",
"name": "color",
"regex": "blue|yellow|red|pink|green"
}
]

In the module file (packages/{PACKAGE NAME}/colorpicker.py):

def run(string, entities):
# string: "I like red and blue colors"
# entities: [{'start': 7, 'end': 10, 'accuracy': 1, 'sourceText': 'red', 'utteranceText': 'red', 'entity': 'color'}, {'start': 15, 'end': 19, 'accuracy': 1, 'sourceText': 'blue', 'utteranceText': 'blue', 'entity': 'color'}]
print('entity name:', entities[0]['entity']) # entity name: color
print('entity value:', entities[0]['sourceText']) # entity value: red
print('entity name:', entities[1]['entity']) # entity name: color
print('entity value:', entities[1]['sourceText']) # entity value: blue

Persistent Data#

Leon uses TinyDB to deal with packages databases. Each package can have its own database and the database is managed by modules.

For more information, you can refer to the:

Installing Third Party Python Packages#

Leon runs in a virtual environment to ensure that the project's packages/dependencies doesn't conflict with the ones installed system wide.

To install third party packages, kindly follow these steps;

  1. Open a terminal window at the bridges/python directory.
  2. Perform pipenv install {PACKAGE NAME}=={PACKAGE VERSION}. Note that it must use a specific package version.
  3. Import the newly installed package in the required module file with import {PACKAGE NAME}.

Kindly note that {PACKAGE NAME} and {PACKAGE VERSION} are placeholders. They should be replaced with the name and version of the actual package you wish to install.

Outputs#

Every module does something, and the outputs allow the core to understand what the module did and what is the state of its execution. This is thanks to the outputs that Leon knows what to do next.

The core understands two types of outputs:

  • inter, which are the intermediate outputs. You can have as many intermediate outputs as you want.
  • end, which is the final output. You must only have one final output, that allows Leon to know that the module execution is done.

Outputs are represented by the utils.output() function.

Test a Module#

On The Fly#

To test the behavior of your module while you are creating it, you can run this following command from the project root:

PIPENV_PIPFILE=bridges/python/Pipfile pipenv run python bridges/python/main.py server/src/query-object.sample.json

For example, for the Is It Down module the query object file could look like that:

{
"lang": "en",
"package": "checker",
"module": "isitdown",
"action": "run",
"query": "Check if github.com and mozilla.org are up",
"entities": [
{
"sourceText": "github.com",
"utteranceText": "github.com",
"entity": "url",
"resolution": {
"value": "github.com"
}
},
{
"sourceText": "mozilla.org",
"utteranceText": "mozilla.org",
"entity": "url",
"resolution": {
"value": "mozilla.org"
}
}
]
}
Tip

Don't forget to take a look at this list to see how entities are formatted.

End-to-End#

Modules come with their own tests. They are represented by a unique file for every module that can be found at: packages/{PACKAGE NAME}/test/{MODULE NAME}.spec.js.

As you may noticed, JavaScript is used to test modules because the core is written in JavaScript and we do end-to-end tests by processing a query to the NLU. Then the Leon's brain is executed and returns the output. This is the output, especially the codes that have been interpreted by your module that you need to consider.

Leon uses Jest as a testing framework.

Here are two tests examples of the Is It Down module:

// You must use describe() with the following name syntax: {PACKAGE NAME}:{MODULE NAME}
describe('checker:isitdown', async () => {
// Each test must be represented by the test()
test('detects invalid domain name', async () => {
global.nlu.brain.execute = jest.fn()
// The string you want to test
await global.nlu.process('Check if github is up')
const [obj] = global.nlu.brain.execute.mock.calls
await global.brain.execute(obj[0])
// Grab the accumulated codes and verify if they are the expected ones
expect(global.brain.finalOutput.codes).toIncludeSameMembers(['invalid_domain_name'])
})
test('detects down domain name', async () => {
global.nlu.brain.execute = jest.fn()
await global.nlu.process('Check if fakedomainnametotestleon.fr is up')
const [obj] = global.nlu.brain.execute.mock.calls
await global.brain.execute(obj[0])
expect(global.brain.finalOutput.codes).toIncludeSameMembers([
'checking',
'down',
'done'
])
})
})

These tests can be found in packages/checker/test/isitdown.spec.js

Once you finished to write your tests, you can execute the following command to run them:

npm run test:module {PACKAGE NAME}:{MODULE NAME}
# E.g.
npm run test:module checker:isitdown

Utils Functions#

Utils functions are available in bridges/python/utils.py.

To use the following functions, do not forget to import the Python utils module at the beginning of your Leon's module:

import utils
# utils.myFunc()
Tip

You can also contribute by improving these functions or by adding new ones to make the modules creation even better.

getqueryobj()#

Returns the query object.

utils.getqueryobj()
# >> { 'id': '1560037280351-8cbc', 'lang': 'en', 'package': 'leon', 'module': 'whoami', 'action': 'run', 'query': 'Who are you?', 'entities': [] }

translate(key, d = { })#

Randomly pick up a module answer from the packages/{PACKAGE NAME}/data/answers/{LANG}.json file via the given key, do string interpolation via the given data object and return the plain string answer.

  • key: module answer key to pick up the right string.
  • d: data object for string interpolation.
{
"greeting": {
"morning_good_day": [
"Good morning, I wish you a very pleasant day!",
"Good morning, I hope your day will be full of joy and productivity!"
],
"funny_hello": [
"%fun_hello%, yes I try to be fun here."
]
}
}
utils.translate('morning_good_day')
# >> Good morning, I wish you a very pleasant day!
# OR >> Good morning, I hope your day will be full of joy and productivity!
fun_hello = 'Heeey'
utils.translate('funny_hello', { 'fun_hello': fun_hello })
# >> Heeey, yes I try to be fun here.

output(type, code, speech = '')#

Communicate the module data to the core.

  • type (inter|end): output type to inform the core if the module execution is done or not. The end type must appears one time.
  • code: output code to provide an additional information about the type of output. Usually used by the modules tests.
  • speech: plain string answer.
utils.output('inter', 'just_a_code', 'This is an intermediate answer.')
# >> <object output>
utils.output('end', 'done', utils.translate('done_answer'))
# >> <object output>

http(method, url, headers = None)#

Send HTTP request with the user-agent Leon/{VERSION NUMBER}. It uses the Request Python library.

  • method: HTTP method.
  • url: URL to request.
  • headers: HTTP headers.
utils.http('GET', 'https://getleon.ai')
# >> cf. https://docs.python-requests.org/en/master/user/advanced/#request-and-response-objects
# Request with custom headers
utils.http('POST', 'https://an-awesome-api.com', { 'Authorization': 'Bearer xxx' })
# >> cf. https://docs.python-requests.org/en/master/user/advanced/#request-and-response-objects

config(key)#

Get a module configuration from the packages/{PACKAGE NAME}/config/config.json file.

  • key: module configuration key.
{
"anawesomemodule": {
"api_key": "my-super-api-key",
"options": {}
}
}
utils.config('api_key')
# >> my-super-api-key

createdldir()#

Create the downloads folder of the current module. When Leon needs to download something, it is saved into: downloads/{PACKAGE NAME}/{MODULE NAME}.

utils.createdldir()
# >> <download module folder path>

db(dbtype = 'tinydb')#

Create a new dedicated database for a specific package.

  • dbtype (tinydb): database type. Only supports TinyDB today.
utils.db()
# >> { 'db': db_instance, 'query': Query, 'operations': operations }
Éditer cette page