wiki.ziemers.de

ziemer's informatik Wiki

Benutzer-Werkzeuge

Webseiten-Werkzeuge

Zuletzt angesehen:
wiki:software:beuthbot:deconcentrator

Unterschiede

Hier werden die Unterschiede zwischen zwei Versionen angezeigt.

Link zu dieser Vergleichsansicht

Nächste Überarbeitung 		Vorhergehende Überarbeitung
wiki:software:beuthbot:deconcentrator [23.01.2020 10:28]
Timo Bruns angelegt 		wiki:software:beuthbot:deconcentrator [19.05.2020 21:52] (aktuell)
Jan Fromme
Zeile 1: Zeile 1:
 ===== Deconcentrator  ===== ===== Deconcentrator  =====
 +let deconcentrator do the "hard" work of checking multiple natural language understanding processing providers. 
 +
 +
 +==== Table Of Content ====
 +   - [[wiki:software:beuthbot:deconcentrator#deconcentrator|Deconcentrator]]
 +   - [[wiki:software:beuthbot:deconcentrator#table_of_content|Table of content]]
 +   - [[wiki:software:beuthbot:deconcentrator#scope|Scope]]
 +   - [[wiki:software:beuthbot:deconcentrator#how-to|How To]]
 +   - [[wiki:software:beuthbot:deconcentrator#authors|Authors]]
 +
 +
 +==== scope ==== 
 +be a common endpoint for various nlu providers:
 +
 +- [[https://rasa.com/docs/rasa/|RASA]]
 +
 +others aren't implemented yet, but implementation should be trivial: 
 +
 +- [[https://azure.microsoft.com/de-de/services/cognitive-services/language-understanding-intelligent-service/|Microsoft LUIS]]\\
 +- [[https://cloud.google.com/natural-language/|Google Cloud NLU]]\\
 +- [[https://www.ibm.com/watson/services/natural-language-understanding/|IBM Watson NLU]]\\
 +
 +==== principles ==== 
 +
 +=== important frameworks/software pieces ===
 +- [[https://nginx.org|nginx]]: reverse proxy (static files) and uwsgi gateway
 +- [[https://pypi.org/project/uWSGI/|uWSGI]]: wsgi implementation
 +- [[https://www.django-rest-framework.org/|Django REST framework]]: REST interfaces, viewsets, generic serialization 
 +    logic etc.
 +- [[https://www.djangoproject.com/|Django]]: "The web framework for perfectionists with deadlines." 
 +- [[https://docs.celeryproject.org/|Celery]]: Distributed task queue for delegating I/O tasks (like doing web requests)
 +- [[https://www.rabbitmq.com/|RabbitMQ]]: async task queue itself
 +- [[https://redis.io/|redis]]: django cache, session cache, celery result backend
 +- [[https://memcached.org/|memcached]]: django cache.
 +- [[https://www.postgresql.org/|PostgreSQL]]: database backend.
 +=== important models ===
 +
 +- `Method`: the abstraction of a function to retrieve an actual NLU processing result.
 +- `Provider`: the actual provider, which is doing some kind of NLU processing. 
 +- `Strategy`: how to select a `Provider` for a specific `Objective`
 +- `Objective`: kind of a task that has to be done. It's the main entry-point, user-supplied. It contains the actual 
 +    payload which has to be NLU processed and selects an strategy.
 +- `Job`: the `Strategy` creates jobs from an `Objective`. Each job then has a specific `Provider` to use for processing.
 +- `Result`: the outcome of asking a `Provider`.
 +
 +=== implementation details ===
 +
 +- `Objective`, `Job` and `Result` make use of non-abc-dispatching (i. e. dispatching without a common abstract base 
 +    class). That means:
 +    - they have a common method with equal signature called `execute()` and are connected to the same `post_save` 
 +        handler.
 +    - once an object of one of these classes is `save`d, the `post_save` hook will call that common method. 
 +    - that method, then, calls the `Strategy` model method for further handling.
 +    - to avoid infinite recursion, one has to avoid calling `save()` within the `Strategy` method, instead using the 
 +        `<Model>.objects.filter(...).update(...)`-approach
 +- the `ViewSet`s design:
 +    - `Method`s, `Strategy`s, `Job`s and `Result`s can only be read
 +    - `Provider`s can be CRUD on demand. 
 +    - `Objective`s can be created and retrieved, but not updated or deleted.
 +
 +==== how-to ====
 +
 +The project contains multiple `docker-compose` files; therefore, basically you only have to `docker-compose up` the 
 +project. 
 +
 +- You need to build the deconcentrator image in the directory that holds the Dockerfile.
 +    docker build -t deconcentrator:latest .
 +- You can use the `management.sh` script to create the required secret files: `./management.sh prepare`; they are required to start the containers.
 +
 +- You will probably want to create a sym-link named `docker-compose.override.yml` to export the nginx port to the public 
 +    ln -s docker-compose.production.yml docker-compose.override.yml
 +- You will probably want to run the migrations; basically it's a `./manage.py migrate` call, which can be done via 
 +    ./management.sh exec uwsgi /mnt/deconcentrator/manage.py migrate
 +- To access the admin interface, you'll have to create a superuser account first: 
 +    ./management.sh exec uwsgi /mnt/deconcentrator/manage.py createsuperuser
 +
 +
 +
 +
 +==== Authors ====
 +
 +
 + **Kai Nessig** - //Initial work// - [[https://github.com/tiberius|GitHub]]
 +
 +See also the list of [[https://github.com/beuthbot/deconcentrator/graphs/contributors]] who participated in this project.
 +<WRAP pagebreak></WRAP>
wiki/software/beuthbot/deconcentrator.1579771739.txt.gz · Zuletzt geändert: 23.01.2020 10:28 von Timo Bruns

Seiten-Werkzeuge