Graph Recipe

Default Recipe or Graph Recipe?

You will probably only need graph recipes if you're running ML experiments or ablation studies on an existing model. We recommend starting with the default recipe and for many applications that will be all that's needed.

We now support graph recipes in addition to the default recipe. Graph recipes provide more granular control over how execution graph schemas are built.

New in 3.1

This feature is experimental. We introduce experimental features to get feedback from our community, so we encourage you to try it out! However, the functionality might be changed or removed in the future. If you have feedback (positive or negative) please share it with us on the Rasa Forum.

Differences with Default Recipe

There are some differences between the default recipe and the new graph recipe. Main differences are:

• Default recipe is named default.v1 in the config file whereas graph recipes are named graph.v1.
• Default recipes provide an easy to use recipe structure whereas graph recipes are more advanced and powerful.
• Default recipes are very opinionated and provide various defaults whereas graph recipes are more explicit.
• Default recipes can auto-configure themselves and dump the defaults used to the file if some sections in config.yml are missing, whereas graph recipes do none of this and assume what you see is what you get. There are no surprises with graph recipes.
• Default recipe divides graph configuration into mainly two parts: pipeline and policies. These can also be described as NLU and core (dialogue management) parts. For graph recipe on the other hand, the separation is between training (ie. train_schema) and prediction (ie. predict_schema).

Starting from scratch?

If you don't know which recipe to choose, use the default recipe to bootstrap your project fast. If later you find that you need more fine-grained control, you can always change your recipe to be a graph recipe.

Graph Configuration File Structure

Graph recipes share recipe and language keys with the same meaning. Similarities end there as graph recipes do not have pipeline or policies keys but they do have train_schema and predict_schema keys for determining the graph nodes during train and predict runs respectively. In addition to this, target nodes for NLU and core can be specified explicitly with graph recipes, these can be declared with nlu_target and core_target. If targets are omitted, node names used by default recipe will take over, and these are run_RegexMessageHandler and select_prediction for nlu and core respectively.

Here's an example graph recipe:

# The config recipe.
recipe: graph.v1

language: en

core_target: custom_core_target

nlu_target: custom_nlu_target

train_schema:
  nodes:
    # We skip schema_validator node (we only have this for DefaultV1Recipe
    # since we don't do validation for the GraphV1Recipe)
    finetuning_validator:
      needs:
        importer: __importer__
      uses: rasa.graph_components.validators.finetuning_validator.FinetuningValidator
      constructor_name: create
      fn: validate
      config:
        validate_core: true
        validate_nlu: true
      eager: false
      is_target: false
      is_input: true
      resource: null
    nlu_training_data_provider:
      needs:
        importer: finetuning_validator
      uses: rasa.graph_components.providers.nlu_training_data_provider.NLUTrainingDataProvider
      constructor_name: create
      fn: provide
      config:
        language: en
        persist: false
      eager: false
      is_target: false
      is_input: true
      resource: null
    domain_provider:
      needs:
        importer: finetuning_validator
      uses: rasa.graph_components.providers.domain_provider.DomainProvider
      constructor_name: create
      fn: provide_train
      config: { }
      eager: false
      is_target: true
      is_input: true
      resource: null
    domain_for_core_training_provider:
      needs:
        domain: domain_provider
      uses: rasa.graph_components.providers.domain_for_core_training_provider.DomainForCoreTrainingProvider
      constructor_name: create
      fn: provide
      config: { }
      eager: false
      is_target: false
      is_input: true
      resource: null
    story_graph_provider:
      needs:
        importer: finetuning_validator
      uses: rasa.graph_components.providers.story_graph_provider.StoryGraphProvider
      constructor_name: create
      fn: provide
      config:
        exclusion_percentage: null
      eager: false
      is_target: false
      is_input: true
      resource: null
    training_tracker_provider:
      needs:
        story_graph: story_graph_provider
        domain: domain_for_core_training_provider
      uses: rasa.graph_components.providers.training_tracker_provider.TrainingTrackerProvider
      constructor_name: create
      fn: provide
      config: { }
      eager: false
      is_target: false
      is_input: false
      resource: null
    train_MemoizationPolicy0:
      needs:
        training_trackers: training_tracker_provider
        domain: domain_for_core_training_provider
      uses: rasa.core.policies.memoization.MemoizationPolicy
      constructor_name: create
      fn: train
      config: { }
      eager: false
      is_target: true
      is_input: false
      resource: null

predict_schema:
  nodes:
    nlu_message_converter:
      needs:
        messages: __message__
      uses: rasa.graph_components.converters.nlu_message_converter.NLUMessageConverter
      constructor_name: load
      fn: convert_user_message
      config: {}
      eager: true
      is_target: false
      is_input: false
      resource: null
    custom_nlu_target:
      needs:
        messages: nlu_message_converter
        domain: domain_provider
      uses: rasa.nlu.classifiers.regex_message_handler.RegexMessageHandler
      constructor_name: load
      fn: process
      config: {}
      eager: true
      is_target: false
      is_input: false
      resource: null
    domain_provider:
      needs: {}
      uses: rasa.graph_components.providers.domain_provider.DomainProvider
      constructor_name: load
      fn: provide_inference
      config: {}
      eager: true
      is_target: false
      is_input: false
      resource:
        name: domain_provider
    run_MemoizationPolicy0:
      needs:
        domain: domain_provider
        tracker: __tracker__
        rule_only_data: rule_only_data_provider
      uses: rasa.core.policies.memoization.MemoizationPolicy
      constructor_name: load
      fn: predict_action_probabilities
      config: {}
      eager: true
      is_target: false
      is_input: false
      resource:
        name: train_MemoizationPolicy0
    rule_only_data_provider:
      needs: {}
      uses: rasa.graph_components.providers.rule_only_provider.RuleOnlyDataProvider
      constructor_name: load
      fn: provide
      config: {}
      eager: true
      is_target: false
      is_input: false
      resource:
        name: train_RulePolicy1
    custom_core_target:
      needs:
        policy0: run_MemoizationPolicy0
        domain: domain_provider
        tracker: __tracker__
      uses: rasa.core.policies.ensemble.DefaultPolicyPredictionEnsemble
      constructor_name: load
      fn: combine_predictions_from_kwargs
      config: {}
      eager: true
      is_target: false
      is_input: false
      resource: null

graph targets

For NLU, default target name of run_RegexMessageHandler will be used, while for core (dialogue management) the target will be called select_prediction if omitted. Make sure you have graph nodes with relevant names in your schema definitions.

In a similar fashion, note that the default resource needed by the first graph node is fixed to be __importer__ (representing configuration, training data etc.) for training task and it is __message__ (representing the message received) for prediction task. Make sure your first nodes make use of these dependencies.

Graph Node Configuration

As you can see in the example above, graph recipes are very much explicit and you can configure each graph node as you would like. Here is an explanation of what some of the keys mean:

• needs: You can define here what data your graph node requires and from which parent node. Key is the data name, whereas the value would refer to the node name.

needs:
    messages: nlu_message_converter

Current graph node needs messages which is provided by nlu_message_converter node.

• uses: You can provide the class used to instantiate this node with this key. Please provide the full path in Python path syntax, eg.

uses: rasa.graph_components.converters.nlu_message_converter.NLUMessageConverter

You are not required to use Rasa internal graph component classes and you can use your own components here.

• constructor_name: This is the constructor used to instantiate your component. Example:

constructor_name: load

• fn: This is the function used in executing the graph component. Example:

fn: combine_predictions_from_kwargs

• config: You can provide any configuration parameters for your components using this key.

config:
    language: en
    persist: false

• eager: This determines if your component should be eagerly loaded when the graph is constructed or if it should wait until the runtime (this is called lazy instantiation). Usually we always instantiate lazily during training and eagerly during inference (to avoid slow first prediction).

eager: true

• resource: If given, graph node is loaded from this resource instead of instantiated from scratch. This is e.g. used to load a trained component for predictions.

resource:
    name: train_RulePolicy1

• is_target: Boolean value, if True then this node can't be pruned during fingerprinting (it might be replaced with a cached value though). This is e.g. used for all components which train as their result always needs to be added to the model archive so that the data is available during inference.

is_target: false

• is_input: Boolean value; nodes with is_input are always run (also during the fingerprint run). This makes sure that we e.g. detect changes in file contents.

 is_input: false