Tutorial: Using Haystack with REST API

Download

_{Last Updated:
April 14, 2023}

Level: Advanced
Time to complete: 30 minutes
Prerequisites: Basic understanding of Docker and basic knowledge of Haystack pipelines.
Nodes Used: ElasticsearchDocumentStore, EmbeddingRetriever
Goal: After you complete this tutorial, you will have learned how to interact with Haystack through REST API.

Overview

Learn how you can interact with Haystack through REST API. This tutorial introduces you to all the concepts needed to build an end-to-end document search application.

With Haystack, you can apply the latest NLP technology to your own data and create production-ready applications. Building an end-to-end NLP application requires the combination of multiple concepts:

DocumentStore is the component in Haystack responsible for loading and storing text data in the form of Documents. In this tutorial, the DocumentStore uses Elasticsearch behind the scene.
Haystack pipelines convert files into Documents, index them to the DocumentStore, and run NLP tasks such as question answering and document search.
REST API, as a concept, makes it possible for applications to interact with each other by handling their queries and returning responses. There is rest_api application within Haystack that exposes Haystack’s functionalities through a RESTful API.
Docker simplifies the environment setup needed to run Elasticsearch and Haystack API.

Preparing the Environment

Install Docker Compose and launch Docker. If you installed Docker Desktop, just start the application. Run docker info to see if Docker is up and running:
```
docker info
```
Download the docker-compose.yml file. Haystack provides a docker-compose.yml file that defines services for Haystack API and Elasticsearch.
1. Create a new folder called doc-search in a directory where you want to keep all tutorial related files.
2. Save the latest docker-compose.yml file from GitHub into the folder. To save the docker-compose.yml file into the directory directly, run:
```
curl --output docker-compose.yml https://raw.githubusercontent.com/deepset-ai/haystack/main/docker-compose.yml
```
Here’s what the /doc-search folder should look like:
```
/doc-search
└── docker-compose.yml
```

Now that your environment’s ready, you can start creating your indexing and query pipelines.

Creating the Pipeline YAML File

You can define components and pipelines using YAML code that Haystack translates into Python objects. In a pipeline YAML file, the components section lists all pipeline nodes and the pipelines section defines how these nodes are connected to each other. Let’s start with defining two different pipelines, one to index your documents and another one to query them. We’ll use one YAML file to define both pipelines.

Create a document search pipeline. This will be your query pipeline:
1. In the newly created doc-search folder, create a file named document-search.haystack-pipeline.yml. The docker-compose.yml file and the new pipeline YAML file should be on the same level in the directory:
```
/doc-search
├── docker-compose.yml
└── document-search.haystack-pipeline.yml
```
2. Provide the path to document-search.haystack-pipeline.yml as the volume source value in the docker-compose.yml file. The path must be relative to docker-compose.yml. As both files are in the same directory, the source value will be ./.
```
haystack-api:
  ...
  volumes:
    - ./:/opt/pipelines
```
3. Update the PIPELINE_YAML_PATH variable in docker-compose.yml with the name of the pipeline YAML file. The PIPELINE_YAML_PATH variable tells rest_api which YAML file to run.
```
environment:
  ...
  - PIPELINE_YAML_PATH=/opt/pipelines/document-search.haystack-pipeline.yml
  ...
```
4. Define the pipeline nodes in the components section of the file. A document search pipeline requires a DocumentStore and a Retriever. Our pipeline will use ElasticsearchDocumentStore and EmbeddingRetriever:
```
components:
  - name: DocumentStore # How you want to call this node here
    type: ElasticsearchDocumentStore # This is the Haystack node class
    params: # The node parameters
      embedding_dim: 384 # This parameter is required for the embedding_model
  - name: Retriever
    type: EmbeddingRetriever
    params:
      document_store: DocumentStore
      top_k: 10
      embedding_model: sentence-transformers/all-MiniLM-L6-v2
```
5. Create a query pipeline in the pipelines section. Here, name refers to the name of the pipeline, and nodes defines the order of the nodes in the pipeline:
```
pipelines:
  - name: query 
    nodes:
      - name: Retriever
        inputs: [Query]
```

In the same YAML file, create an indexing pipeline. This pipeline will index your documents to Elasticsearch through rest_api.

Define FileTypeClassifier, TextConverter, and PreProcessor nodes for the pipeline:

components:
  ...
  - name: FileTypeClassifier
    type: FileTypeClassifier
  - name: TextFileConverter
    type: TextConverter
  - name: Preprocessor
    type: PreProcessor
    params: # These parameters define how you want to split your documents
      split_by: word
      split_length: 250
      split_overlap: 30 
      split_respect_sentence_boundary: True

In the pipelines section of the YAML file, create a new pipeline called indexing. In this pipeline, indicate how the nodes you just defined are connected to each other, Retriever, and DocumentStore. This indexing pipeline supports .TXT files and preprocesses them before loading to Elasticsearch.

pipelines:
  ...
  - name: indexing
    nodes:
      - name: FileTypeClassifier
        inputs: [File]
      - name: TextFileConverter
        inputs: [FileTypeClassifier.output_1]
      - name: Preprocessor
        inputs: [TextFileConverter]
      - name: Retriever
        inputs: [Preprocessor]
      - name: DocumentStore
        inputs: [Retriever]

After creating query and indexing pipelines, add version: 1.12.1 to the top of the file. This is the Haystack version that comes with the Docker image in the docker-compose.yml. Now, the pipeline YAML is ready.

version: 1.12.1

components:
  - name: DocumentStore
    type: ElasticsearchDocumentStore
    params:
      embedding_dim: 384
  - name: Retriever
    type: EmbeddingRetriever
    params:
      document_store: DocumentStore
      top_k: 10 
      embedding_model: sentence-transformers/all-MiniLM-L6-v2
  - name: FileTypeClassifier
    type: FileTypeClassifier
  - name: TextFileConverter
    type: TextConverter
  - name: Preprocessor
    type: PreProcessor
    params:
      split_by: word
      split_length: 250
      split_overlap: 30 
      split_respect_sentence_boundary: True

pipelines:
  - name: query 
    nodes:
      - name: Retriever
        inputs: [Query]
  - name: indexing
    nodes:
      - name: FileTypeClassifier
        inputs: [File]
      - name: TextFileConverter
        inputs: [FileTypeClassifier.output_1]
      - name: Preprocessor
        inputs: [TextFileConverter]
      - name: Retriever
        inputs: [Preprocessor]
      - name: DocumentStore
        inputs: [Retriever]

Feel free to play with the pipeline setup later on. Add or remove some nodes, change the parameters, or add new ones. For more options for nodes and parameters, check out Haystack API Reference.

Launching Haystack API and Elasticsearch

Pipelines are ready. Now it’s time to start Elasticsearch and Haystack API.

Run docker-compose up to start the elasticsearch and haystack-api containers. This command installs all the necessary packages, sets up the environment, and launches both Elasticsearch and Haystack API. Launching may take 2-3 minutes.
```
docker-compose up
```
Test if everything is OK with the Haystack API by sending a cURL request to the /initialized endpoint. If everything works fine, you will get true as a response.
```
curl --request GET http://127.0.0.1:8000/initialized
```

Both containers are initialized. Time to fill your DocumentStore with files.

Indexing Files to Elasticsearch

Right now, your Elasticsearch instance is empty. Haystack API provides a /file-upload endpoint to upload files to Elasticsearch. This endpoint uses the indexing pipeline you defined in the pipeline YAML. After indexing files to Elasticsearch, you can perform document search.

Download the example files to the doc-search folder. The .zip file contains text files about countries and capitals crawled from Wikipedia.

/doc-search
├── docker-compose.yml
├── document-search.haystack-pipeline.yml
└── /article_txt_countries_and_capitals
    ├── 0_Minsk.txt
    └── ...

Index files to Elasticsearch. You can send cURL requests to the /file-upload endpoint to upload files to the Elasticsearch instance. If the file is successfully uploaded, you will get null as a response.

curl --request POST \
     --url http://127.0.0.1:8000/file-upload \
     --header 'accept: application/json' \
     --header 'content-type: multipart/form-data' \
     --form files=@article_txt_countries_and_capitals/0_Minsk.txt \
     --form meta=null

This method is not the best one if you have multiple files to upload. That’s because you need to replace file names in the request by hand. Instead, you can run a command that takes all .TXT files in the article_txt_countries_and_capitals folder and sends a POST request to index each file:

find ./article_txt_countries_and_capitals -name '*.txt' -exec \
     curl --request POST \
          --url http://127.0.0.1:8000/file-upload \
          --header 'accept: application/json' \
          --header 'content-type: multipart/form-data' \
          --form files="@{}" \
          --form meta=null \;

Querying Your Pipeline

That’s it, the application is ready! Send another POST request to retrieve documents about “climate in Scandinavia”:

curl --request POST \
     --url http://127.0.0.1:8000/query \
     --header 'accept: application/json' \
     --header 'content-type: application/json' \
     --data '{
     "query": "climate in Scandinavia"
     }'

As a response, you will get a QueryResponse object consisting of query, answers, and documents. Documents related to your query will be under the documents attribute of the object.

{
  "query": "climate in Scandinavia",
  "answers": [],
  "documents": [
    {
      "id": "24904f783ea4b90a47c33434a3e9df7a",
      "content": "Because of Sweden's high latitude, the length of daylight varies greatly. North of the Arctic Circle, the sun never sets for part of each summer, and it never rises for part of each winter. In the capital, Stockholm, daylight lasts for more than 18 hours in late June but only around 6 hours in late December. Sweden receives between 1,100 and 1,900 hours of sunshine annually...",
      "content_type": "text",
      "meta": {
        "_split_id": 33,
        "name": "43_Sweden.txt"
      },
      "score": 0.5017639926813274
    },
    ...
  ]
}

Congratulations! You have created a proper search system that runs using Haystack REST API.

Training Your Own Dense Passage Retrieval Model

Question Answering on a Knowledge Graph