Domain, Slots, and Actions¶

Domain¶

The Domain defines the universe in which your bot operates. It specifies exactly:

which intents you are expecting to respond to
which slots you wish to track
which actions your bot can take

For example, the DefaultDomain has the following yaml definition:

# all hashtags are comments :)
intents:
 - greet
 - default
 - goodbye

entities:
 - name

slots:
  name:
    type: text

templates:
  utter_greet:
    - "hey there {name}!"    # variable will be filled by slot with the same name or by custom code
  utter_goodbye:
    - "goodbye :("
    - "bye bye"              # multiple templates will allow the bot to randomly pick from them
  utter_default:
    - "default message"

actions:
  - utter_default
  - utter_greet
  - utter_goodbye

What does this mean?

An intent is a string like "greet" or "restaurant_search". It describes what your user probably meant to say. For example, “show me Mexican restaurants”, and “I want to have lunch” could both be described as a restaurant_search intent.

slots are the things you want to keep track of during a conversation. For example, in the messages above you would want to store “Mexican” as a cuisine type. The tracker has an attribute like tracker.get_slot("cuisine") which will return "Mexican"

actions are the things your bot can actually do. They are invoked by calling the action.run() method. For example, an action can:

respond to a user
make an external API call
query a database

Note

For mor information about the utter template format (e.g. the use of variables like {name} or buttons) take a look at Utterance templates.

Defining Custom Actions¶

The easiest are UtterActions, which just send a message to the user. You define them by adding an entry to the action list that is named after the utterance. E.g. if there should be an action that utters the template called utter_greet you need to add utter_greet to the list of defined actions. In the above example yaml you can see that all three of the defined actions are just named after utter templates and hence just respond with a message to the user.

What about more complicated actions? To continue with the restaurant example, if the user says “show me a Mexican restaurant”, your bot would execute the action ActionCheckRestaurants, which might look like this:

from rasa_core.actions import Action
from rasa_core.events import SlotSet

class ActionCheckRestaurants(Action):
   def name(self):
      # type: () -> Text
      return "action_check_restaurants"

   def run(self, dispatcher, tracker, domain):
      # type: (Dispatcher, DialogueStateTracker, Domain) -> List[Event]

      cuisine = tracker.get_slot('cuisine')
      q = "select * from restaurants where cuisine='{0}' limit 1".format(cuisine)
      result = db.query(q)

      return [SlotSet("matches", result if result is not None else [])]

Note that actions do not mutate the tracker directly. Instead, an action can return events which are logged by the tracker and used to modify its own state.

Putting it all together¶

Let’s add just this one new action to a custom domain (assuming we stored the action in a module called restaurant.actions):

actions:
  - utter_default
  - utter_greet
  - utter_goodbye
  - restaurant.actions.ActionCheckRestaurants   # custom action

We only show the changed action list here, you also need to include the other parts from the original domain! The point of this is just to show how the pieces fit together. As you can see, in the actions section of your domain, you can list utter actions (which respond an utter template to the user) as well as custom actions using their module path.

For an example you can run, check the Building a Simple Bot.

Utterance templates¶

Utterance templates are messages the bot will send back to the user. Either automatically by an action with the same name as the utterance (e.g. in the above example the utter_default template and action) or by an action with custom code.

Images and Buttons¶

Templates defined in a domains yaml file can contain images and buttons as well:

templates:
  utter_greet:
  - text: "Hey! How are you?"
    buttons:
    - title: "great"
      payload: "great"
    - title: "super sad"
      payload: "super sad"
  utter_cheer_up:
  - text: "Here is something to cheer you up:"
    image: "https://cdn77.eatliver.com/wp-content/uploads/2017/10/trump-frog.jpg"

Note

Please keep in mind that it is up to the implementation of the output channel on how to display the defined buttons. E.g. the cmdline interface can not display buttons or images, but tries to mimic them in the command line.

Variables¶

You can also use variables in your templates to insert information collected during the dialogue. You can either do that in your custom python code or by using the automatic slot filling mechanism. E.g if you got a template like this:

templates:
  utter_greet:
  - text: "Hey, {name}. How are you?"

Rasa will automatically fill that variable with a value found in a slot called name.

In custom code, you can retrieve a template by using:

class ActionCustom(Action):
   def name(self):
      return "action_custom"

   def run(self, dispatcher, tracker, domain):
      # send utter default template to user
      dispatcher.utter_template("utter_default")
      # ... other code
      return []

If the template contains variables denoted with {my_variable} you can supply values for the fields by passing them as key word arguments to utter_template:

dispatcher.utter_template("utter_default", my_variable="my text")

Variations¶

If you want to randomly vary the response send to the user, you can list multiple responses and the bot will randomly pick one of them, e.g.:

templates:
  utter_greeting:
  - text: "Hey, {name}. How are you?"
  - text: "Hey, {name}. How is your day going?"

Slots¶

Most slots influence the prediction of the next action the bot should run. For the prediction, the slots value is not used directly, but rather it is featurized. E.g. for a slot of type text, the value is irrelevant, for the featurization the only thing that matters is if a text is set or not. In a slot of type unfeaturized any value can be stored, the slot never influences the prediction.

The choice of a slots type should be done with care. If a slots value should influence the dialogue flow (e.g. the users age influences which question follows next) you should choose a slot where the value influences the dialogue model.

These are all of the predefined slot classes and what they’re useful for:

text¶

Use For:	User preferences where you only care whether or not they’ve been specified.
Example:	slots: cuisine: type: text
Description:	Results in the feature of the slot being set to `1` if any value is set. Otherwise the feature will be set to `0` (no value is set).

bool¶

Use For:	True or False
Example:	slots: is_authenticated: type: bool
Description:	Checks if slot is set and if True

categorical¶

Use For:	Slots which can take one of N values
Example:	slots: risc_level: type: categorical values: - low - medium - high
Description:	Creates a one-hot encoding describing which of the `values` matched.

float¶

Use For:	Continuous values
Example:	slots: temperature: type: float min_value: -100.0 max_value: 100.0
Defaults:	`max_value=1.0`, `min_value=0.0`
Description:	All values below `min_value` will be treated as `min_value`, the same happens for values above `max_value`. Hence, if `max_value` is set to `1`, there is no difference between the slot values `2` and `3.5` in terms of featurization (e.g. both values will influence the dialogue in the same way and the model can not learn to differentiate between them).

list¶

Use For:	Lists of values
Example:	slots: shopping_items: type: list
Description:	The feature of this slot is set to `1` if a value with a list is set, where the list is not empty. If no value is set, or the empty list is the set value, the feature will be `0`. The length of the list stored in the slot does not influence the dialogue.

unfeaturized¶

Use For:	Data you want to store which shouldn’t influence the dialogue flow
Example:	slots: internal_user_id: type: unfeaturized
Description:	There will not be any featurization of this slot, hence its value does not influence the dialogue flow and is ignored when predicting the next action the bot should run.

data¶

Use For:	Base class for creating own slots
Example:	Warning This type should not be used directly, but rather be subclassed.
Description:	User has to subclass this and define the `as_feature` method containing any custom logic.