Choosing a Pipeline

In Rasa Open Source, incoming messages are processed by a sequence of components. These components are executed one after another in a so-called processing pipeline defined in your config.yml. Choosing an NLU pipeline allows you to customize your model and finetune it on your dataset.

How to Choose a Pipeline

The Short Answer

If your training data is in English, a good starting point is the following pipeline:

language: "en"

pipeline:
  - name: ConveRTTokenizer
  - name: ConveRTFeaturizer
  - name: RegexFeaturizer
  - name: LexicalSyntacticFeaturizer
  - name: CountVectorsFeaturizer
  - name: CountVectorsFeaturizer
    analyzer: "char_wb"
    min_ngram: 1
    max_ngram: 4
  - name: DIETClassifier
    epochs: 100
  - name: EntitySynonymMapper
  - name: ResponseSelector
    epochs: 100

If your training data is not in English, start with the following pipeline:

language: "fr"  # your two-letter language code

pipeline:
  - name: WhitespaceTokenizer
  - name: RegexFeaturizer
  - name: LexicalSyntacticFeaturizer
  - name: CountVectorsFeaturizer
  - name: CountVectorsFeaturizer
    analyzer: "char_wb"
    min_ngram: 1
    max_ngram: 4
  - name: DIETClassifier
    epochs: 100
  - name: EntitySynonymMapper
  - name: ResponseSelector
    epochs: 100

A Longer Answer

We recommend using following pipeline, if your training data is in English:

language: "en"

The pipeline contains the [ConveRTFeaturizer](https://legacy-docs-v1.rasa.com/1.10.23/nlu/components/#convertfeaturizer) that provides pre-trained word embeddings of the user utterance.

### [Choosing the Right Components](https://legacy-docs-v1.rasa.com/1.10.23/nlu/choosing-a-pipeline/#id9)

There are components for entity extraction, for intent classification, response selection, pre-processing, and others. You can learn more about any specific component on the [Components](https://legacy-docs-v1.rasa.com/1.10.23/nlu/components/#components) page.

A pipeline usually consists of three main parts:

- [Tokenization](https://legacy-docs-v1.rasa.com/1.10.23/nlu/choosing-a-pipeline/#tokenization)
- [Featurization](https://legacy-docs-v1.rasa.com/1.10.23/nlu/choosing-a-pipeline/#featurization)
- [Entity Recognition / Intent Classification / Response Selectors](https://legacy-docs-v1.rasa.com/1.10.23/nlu/choosing-a-pipeline/#entity-recognition-intent-classification-response-selectors)

#### [Tokenization](https://legacy-docs-v1.rasa.com/1.10.23/nlu/choosing-a-pipeline/#id15)

For tokenization of English input, we recommend the [ConveRTTokenizer](https://legacy-docs-v1.rasa.com/1.10.23/nlu/components/#converttokenizer).

#### [Featurization](https://legacy-docs-v1.rasa.com/1.10.23/nlu/choosing-a-pipeline/#id16)

You need to decide whether to use components that provide pre-trained word embeddings or not. We recommend in cases of small amounts of training data to start with pre-trained word embeddings.

##### [Pre-trained Embeddings](https://legacy-docs-v1.rasa.com/1.10.23/nlu/choosing-a-pipeline/#id18)

The advantage of using pre-trained word embeddings in your pipeline is that if you have a training example like: "I want to buy apples", and Rasa is asked to predict the intent for "get pears", your model already knows that the words "apples" and "pears" are very similar.

##### [Supervised Embeddings](https://legacy-docs-v1.rasa.com/1.10.23/nlu/choosing-a-pipeline/#id19)

If you don’t use any pre-trained word embeddings inside your pipeline, you are not bound to a specific language and can train your model to be more domain specific.

### [Multi-Intent Classification](https://legacy-docs-v1.rasa.com/1.10.23/nlu/choosing-a-pipeline/#id10)

You can use Rasa Open Source components to split intents into multiple labels. For example, you can predict multiple intents (`thank+goodbye`) or model hierarchical intent structure (`feedback+positive` being more similar to `feedback+negative` than `chitchat`).

```yaml
language: "en"

pipeline:
- name: "WhitespaceTokenizer"
  intent_tokenization_flag: True
  intent_split_symbol: "_"
- name: "CountVectorsFeaturizer"
- name: "DIETClassifier"

Handling Class Imbalance

Classification algorithms often do not perform well if there is a large class imbalance, for example if you have a lot of training data for some intents and very little training data for others. To mitigate this problem, you can use a balanced batching strategy.

language: "en"

pipeline:
# - ... other components
- name: "DIETClassifier"
  batch_strategy: sequence

Component Lifecycle

Each component processes an input and/or creates an output. The order of the components is determined by the order they are listed in the config.yml.

For example, for the sentence "I am looking for Chinese food", the output is:

{
    "text": "I am looking for Chinese food",
    "entities": [
        {
            "start": 8,
            "end": 15,
            "value": "chinese",
            "entity": "cuisine",
            "extractor": "DIETClassifier",
            "confidence": 0.864
        }
    ],
    "intent": {"confidence": 0.6485910906220309, "name": "restaurant_search"},
    "intent_ranking": [
        {"confidence": 0.6485910906220309, "name": "restaurant_search"},
        {"confidence": 0.1416153159565678, "name": "affirm"}
    ]
}