Metadata-Version: 2.1
Name: hermes-python
Version: 0.4.0
Summary: Python bindings for Snips Hermes Protocol
Home-page: https://github.com/snipsco/hermes-protocol/tree/develop/platforms/hermes-python
Author: Anthony Reinette
Author-email: anthony.reinette@snips.ai
License: MIT
Keywords: snips
Platform: UNKNOWN
Classifier: Programming Language :: Python :: 2
Classifier: Programming Language :: Python :: 2.7
Classifier: Programming Language :: Python :: 3.5
Description-Content-Type: text/markdown
Requires-Dist: six
Requires-Dist: future
Provides-Extra: test
Requires-Dist: mock ; extra == 'test'
Requires-Dist: pytest ; extra == 'test'
Requires-Dist: coverage ; extra == 'test'
Requires-Dist: pytest-cov ; extra == 'test'
Requires-Dist: setuptools-rust ; extra == 'test'

Hermes Python
*************

.. image:: https://travis-ci.org/snipsco/hermes-protocol.svg
   :target: https://travis-ci.org/snipsco/hermes-protocol

.. image:: https://badge.fury.io/py/hermes-python.svg
   :target: https://badge.fury.io/py/hermes-python


About
*****

The ``hermes-python`` library provides python bindings for the Hermes
protocol that snips components use to communicate together over MQTT.
``hermes-python`` allows you to interface seamlessly with the Snips
platform and kickstart development of Voice applications.

``hermes-python`` abstracts away the connection to the MQTT bus and
the parsing of incoming and outcoming messages from and to the
components of the snips platform.


Requirements
************

Pre-compiled wheels are available for Python 2.7+ and Python 3.5

The pre-compiled wheels supports the following platform tags :

* ``manylinux1_x86_64``

* ``armv7l``, ``armv6``

* ``macos``

If you want to install ``hermes-python`` on another platform, you have
to build it from source.


Installation
************

The library is packaged as a pre-compiled platform wheel, available on
`PyPi <https://pypi.org/project/hermes-python/>`_.

It can be installed with : ``pip install hermes-python``.

Or you can add it to your *requirements.txt* file.


Building from source
********************

If you want to use ``hermes-python`` on platforms that are not
supported, you have to manually compile the wheel.

You need to have *rust* installed :

``curl https://sh.rustup.rs -sSf``

Clone, the ``hermes-protocol`` repository :

::

   git clone git@github.com:snipsco/hermes-protocol.git
   cd hermes-protocol

You need to compile the dynamically linked shared object library :

::

   mkdir -p platforms/hermes-python/target
   CARGO_TARGET_DIR=platforms/hermes-python/target cargo rustc --lib --manifest-path hermes-mqtt-ffi/Cargo.toml --release -- --crate-type cdylib
   mv platforms/hermes-python/target/release/libhermes_mqtt_ffi.dylib platforms/hermes-python/hermes_python/dylib/

You can then build the wheel :

::

   virtualenv env
   source env/bin/activate
   python setup.py bdist_wheel

The built wheels should be in ``platforms/hermes-python/dist``

You can install those with pip : ``pip install
platforms/hermes-python/dist/<your_wheel>.whl``


Tutorial
********

The lifecycle of a script using ``hermes-python`` has the following
steps :

* Initiating a connection to the MQTT broker

* Registering callback functions to handle incoming intent parsed by
   the snips platform

* Listening to incoming intents

* Closing the connection

Let’s quickly dive into an example :

Let’s write an app for a Weather Assistant ! This code implies that
you created a weather assistant using the `Snips Console
<https://console.snips.ai/>`_, and that it has a
``searchWeatherForecast`` intent.

::

   from hermes_python.hermes import Hermes

   MQTT_ADDR = "localhost:1883"        # Specify host and port for the MQTT broker

   def subscribe_weather_forecast_callback(hermes, intent_message):    # Defining callback functions to handle an intent that asks for the weather.
       print("Parsed intent : {}".format(intent_message.intent.intent_name))

   with Hermes(MQTT_ADDR) as h: # Initialization of a connection to the MQTT broker
       h.subscribe_intent("searchWeatherForecast", subscribe_weather_forecast_callback) \  # Registering callback functions to handle the searchWeatherForecast intent
            .start()
       # We get out of the with block, which closes and releases the connection.

This app is a bit limited as it only prints out which intent was
detected by our assistant. Let’s add more features.


Handling the ``IntentMessage`` object
=====================================

In the previous example, we registered a callback that had this
signature.

::

   subscribe_intent_callback(hermes, intent_message)

The ``intent_message`` object contains information that was extracted
from the spoken sentence.

For instance, in the previous code snippet, we extracted the name of
the recognized intent with

::

   intent_message.intent.intent_name

We could also retrieve the associated confidence score the NLU engine
had when classifying this intent with

::

   intent_message.intent.confidence_score


Extracting slots
----------------

Here are some best practices when dealing with slots. The
``IntentMessage`` object has a ``slots`` attribute.

This ``slots`` attributes is a **container** that is empty when the
intent message doesn’t have slots :

::

   assert len(intent_message.slots) == 0

This container is a dictionary where the key is the name of the slot,
and the value is a list of all the slot values for this slot name.

You can access these values in two ways :

::

   assert len(intent_message.slots.slot1) == 0
   assert len(intent_message.slots["slot1"]) == 0

The slot values are of type ``NluSlot`` which is a deeply nested
object, we offer convenience methods to rapidly access the
*slot_value* attribute of the *NluSlot*.

To access the first ``slot_value`` of a slot called ``myslot``, you
can use :

::

   intent_message.slots.myslot.first()

You can also access all the ``slot_value`` of a slot called ``myslot``
:

::

   intent_message.slots.myslot.all()

Let’s add to our Weather assistant example.

We assume that the ``searchWeatherForecast`` has one slot called
``forecast_location``, that indicates which location the user would
like to know the weather at.

Let’s print all the ``forecast_location`` slots :

::

   for slot in intent_message.slots.forecast_location:
       name = slot.slot_name
       confidence = slot.confidence_score
       print("For slot : {}, the confidence is : {}".format(name, confidence))

The *dot* notation was used, but we can also use the dictionary
notation :

::

   for slot in intent_message.slots.forecast_location:
       name = slot["slot_name"]
       print(name)

Some convenience methods are available to easily retrieve slot values
:

*Retrieving the first slot value for a given slot name*

::

   slot_value = intent_message.slots.forecast_location.first()

*Retrieving all slot values for a given slot name*

::

   slot_values = intent_message.slots.forecast_location.all()

Coming back to our example, we can now have the app print the
``forecast_location`` slot value back to the user :

::

   def subscribe_weather_forecast_callback(hermes, intent_message):
       slot_value = intent_message.slots.forecast_location.first().value
       print("The slot was : {}".format(slot_value)


Managing sessions
=================

Snips platform includes support for conversations with back and forth
communication between the Dialogue Manager and the client code. For a
conversation, the dialogue manager creates **sessions**.


Starting a session
------------------

A session can be started in two manners :

* with a notification

* with an action


Ending a session
----------------


Slot Filling
------------


Configuring MQTT options
========================

The connection to your MQTT broker can be configured with the
``hermes_python.ffi.utils.MqttOptions`` class.

The ``Hermes`` client uses the options specified in the
``MqttOptions`` class when establishing the connection to the MQTT
broker.

Here is a code example :

::

   from hermes_python.hermes import Hermes
   from hermes_python.ffi.utils import MqttOptions

   mqtt_opts = MqttOptions()

   def simple_intent_callback(hermes, intent_message):
       print("I received an intent !")

   with Hermes(mqtt_options=mqtt_opts) as h:
       h.subscribe_intents().loop_forever()

Here are the options you can specify in the MqttOptions class :

* ``broker_address``: The address of the MQTT broker. It should be
   formatted as ``ip:port``.

* ``username``: Username to use on the broker. Nullable

* ``password``: Password to use on the broker. Nullable

* ``tls_hostname``: Hostname to use for the TLS configuration.
   Nullable, setting a value enables TLS

* ``tls_ca_file``: CA files to use if TLS is enabled. Nullable

* ``tls_ca_path``: CA path to use if TLS is enabled. Nullable

* ``tls_client_key``: Client key to use if TLS is enabled. Nullable

* ``tls_client_cert``: Client cert to use if TLS is enabled. Nullable

* ``tls_disable_root_store``: Boolean indicating if the root store
   should be disabled if TLS is enabled.

Let’s connect to an external MQTT broker that requires a username and
a password :

::

   from hermes_python.hermes import Hermes
   from hermes_python.ffi.utils import MqttOptions

   mqtt_opts = MqttOptions(username="user1", password="password", broker_address="my-mqtt-broker.com:18852")

   def simple_intent_callback(hermes, intent_message):
       print("I received an intent !")

   with Hermes(mqtt_options=mqtt_opts) as h:
       h.subscribe_intents().loop_forever()


Configuring Dialogue
====================

``hermes-python`` offers the possibility to configure different
aspects of the Dialogue system.


Enabling and disabling intents on the fly
-----------------------------------------

It is possible to enable and disable intents of your assistant on the
fly. Once an intent is disabled, it will not be recognized by the NLU.

Note that intents in the intent filters of started or continued
session will take precedence over intents that are enabled/disabled in
the configuration of the Dialogue.

You can disable/enable intents with the following methods :

::

   from hermes_python.ontology.dialogue.session import DialogueConfiguration

   dialogue_conf = DialogueConfiguration()                          \
                           .disable_intent("intent1")               \
                           .enable_intent("intent2")                \
                           .enable_intents(["intent1", "intent2"])  \
                           .disable_intents(["intent2", "intent1"])

   hermes.configure_dialogue(dialogue_conf)


Release Checklist
*****************

Everytime you need to perform a release, do the following steps : - [
] Commit all changes to the project for said release - [ ] Write all
the changes introduced in the Changelog (source/HISTORY.rst file) and
commit it - [ ] Run tests - [ ] Build the documentation and commit the
README.rst - [ ] Bump the version and commit it - [ ] Upload to PyPI


Build details
*************


Creating macOS wheels
=====================

The build script : ``build_scripts/build_macos_wheels.sh`` uses
``pyenv`` to generate ``hermes-python`` wheels for different versions
of python.

To be able to run it, you need to :

* install pyenv : brew install pyenv. Then follow the additional
   steps detailled

* you then have to install python at different versions:  ``pyenv
   install --list`` to list the available version to install

* Before installing and building the different python version from
   sources, install the required dependencies : `Link here
   <https://github.com/pyenv/pyenv/wiki/>`_

That’s it !
History
==========

0.4.4 (2019-03-20)
^^^^^^^^^^^^^^^^^^
* Adds support to configure the Dialogue Mananger : enabling and disabling intents on the fly.
* Adds slot filling API : You can ask for a specific slot when continuing a session
* adding support for `OrdinalSlot`

0.3.3 (2019-03-06)
^^^^^^^^^^^^^^^^^^
* Fixes a bug with `publish_start_session_notification` that didn't take the `text` parameter into account.

0.3.2 (2019-02-25)
^^^^^^^^^^^^^^^^^^
* Fixes an important bug that gave the argument `hermes` the wrong type for every registered callback. 
* Fixes an important bug that caused the program to crash when parsing intentMessage that had no slots. 

0.3.1 (2019-02-25)
^^^^^^^^^^^^^^^^^^
* Fixes import bug with templates, the `hermes_python.ffi.utils` module now re-exports `MqttOptions`

0.3.0 (2019-02-25)
^^^^^^^^^^^^^^^^^^
* `IntentClassifierResult`'s `probability` field has been renamed to `confidence_score`.
* Introduces support for snips-platform `1.1.0 - 0.61.1`.

0.2.0 (2019-02-04)
^^^^^^^^^^^^^^^^^^
* Introduces options to connect to the MQTT broker (auth + TLS are now supported).

0.1.29 (2019-01-29)
^^^^^^^^^^^^^^^^^^^
* Fixes bug when deserializing `TimeIntervalValue` that used wrong `encode` method instead of `decode`.

0.1.28 (2019-01-14)
^^^^^^^^^^^^^^^^^^^
* Fixes bug when the `__exit__` method was called twice on the `Hermes` class.
* Introduces two methods to the public api : `connect` and `disconnect` that should bring more flexibility

0.1.27 (2019-01-07)
^^^^^^^^^^^^^^^^^^^
* Fixed broken API introduced in `0.1.26` with the publish_continue_session method of the Hermes class. 
* Cast any string that goes in the mqtt_server_adress parameter in the constructor of the Hermes class to be a 8-bit string.

0.1.26 (2019-01-02)
^^^^^^^^^^^^^^^^^^^^^
* LICENSING : This wheel now has the same licenses as the parent project : APACHE-MIT. 
* Subscription to not recognized intent messages is added to the API. You can now write your own callbacks to handle unrecognized intents.  
* Adds send_intent_not_recognized flag to continue session : indicate whether the dialogue manager should handle non recognized intents by itself or sent them as an `IntentNotRecognizedMessage` for the client to handle.

0.1.25 (2018-12-13)
^^^^^^^^^^^^^^^^^^^^^
* Better error handling : Errors from wrapped C library throw a LibException with detailled errors. 




