merge master

CHANGELOG.rst

@@ -11,6 +11,7 @@ Added
 -----
 - environment variables specified with ``${env_variable}`` in a yaml
   configuration file are now replaced with the value of the environment variable
+- more documentation on how to run NLU with Docker
 - ``analyzer`` parameter to ``intent_featurizer_count_vectors`` featurizer to
   configure whether to use word or character n-grams
 

docs/choosing_pipeline.rst

@@ -201,6 +201,8 @@ a full list of components. For example, these two configurations are equivalent:
 
 Below is a list of all the pre-configured pipeline templates.
 
+.. _section_spacy_pipeline:
+
 spacy_sklearn
 ~~~~~~~~~~~~~
 
@@ -225,6 +227,8 @@ the components and configure them separately:
     - name: "ner_synonyms"
     - name: "intent_classifier_sklearn"
 
+.. _section_tensorflow_embedding_pipeline:
+
 tensorflow_embedding
 ~~~~~~~~~~~~~~~~~~~~
 
@@ -253,6 +257,7 @@ default is to use a simple whitespace tokenizer:
 If you have a custom tokenizer for your language, you can replace the whitespace
 tokenizer with something more accurate. 
 
+.. _section_mitie_pipeline:
 
 mitie
 ~~~~~

docs/docker.rst

@@ -1,18 +1,87 @@
 :desc: Using Rasa NLU with Docker
+
 .. _section_docker:
 
 Running in Docker
 =================
 
-Rasa NLU docker images are provided for different backends:
+.. contents::
+
+Images
+------
+
+`Rasa NLU docker images <https://hub.docker.com/r/rasa/rasa_nlu/tags/>`_ are provided for different backends:
+
+- ``spacy``: If you use the :ref:`section_spacy_pipeline` pipeline
+- ``tensorflow``: If you use the :ref:`section_tensorflow_embedding_pipeline`
+  pipeline
+- ``mitie``: If you use the :ref:`section_mitie_pipeline` pipeline
+- ``bare``: If you want to take a base image and enhance it with your custom
+  dependencies
+- ``full`` (default): If you use components from different pre-defined pipelines
+  and want to have everything included in the image.
+
+.. note::
+
+    For the ``tensorflow`` and the ``full`` image a x86_64 CPU with AVX support
+    is required.
+
+
+Training NLU
+------------
+
+To train a NLU model you need to mount two directories into the Docker container:
+
+- a directory containing your project which in turn includes your NLU
+  configuration and your NLU training data
+- a directory which will contain the trained NLU model
 
-The host must have a x86_64 CPU with AVX support.
+.. code-block:: shell
+
+    docker run \
+        -v <project_directory>:/app/project \
+        -v <model_output_directory>:/app/model \
+        rasa/rasa_nlu:latest \
+        run \
+            python -m rasa_nlu.train \
+                -c /app/project/<nlu configuration>.yml \
+                -d /app/project/<nlu data> \
+                -o /app/model \
+                --project <nlu project name>
+
+Running NLU with Rasa Core
+--------------------------
+
+See this `guide <https://rasa.com/docs/core/docker_walkthrough/>`_ which
+describes how to set up all Rasa components as Docker containers and how to
+connect them.
+
+
+Running NLU as Standalone Server
+---------------------
+
+To run NLU as server you have to
+
+- mount a directory with the trained NLU models
+- expose a port
 
 .. code-block:: bash
 
-    docker run -p 5000:5000 rasa/rasa_nlu:latest-full 
+    docker run \
+        -p 5000:5000 \
+        -v <directory with nlu models>:/app/projects \
+        rasa/rasa_nlu:latest \
+        start \
+            --path /app/projects
+            --port 5000
+
+You can then send requests to your NLU server as it is described in the
+:ref:`section_http`, e.g. if it is running on the localhost:
+
+.. code-block:: bash
 
+    curl --request GET \
+         --url 'http://localhost:5000/parse?q=Hello%20world!'
 
-.. include:: feedback.inc	
-
 
+.. include:: feedback.inc