haystack/test/test_retriever.py

import time

import numpy as np
import pandas as pd
import pytest
from pathlib import Path
from elasticsearch import Elasticsearch

from haystack.document_stores import WeaviateDocumentStore
from haystack.schema import Document
from haystack.document_stores.elasticsearch import ElasticsearchDocumentStore
from haystack.document_stores.faiss import FAISSDocumentStore
from haystack.document_stores import MilvusDocumentStore
from haystack.nodes.retriever.dense import DensePassageRetriever, TableTextRetriever
from haystack.nodes.retriever.sparse import ElasticsearchRetriever, ElasticsearchFilterOnlyRetriever, TfidfRetriever
from transformers import DPRContextEncoderTokenizerFast, DPRQuestionEncoderTokenizerFast

from conftest import SAMPLES_PATH


@pytest.fixture()
def docs():
    documents = [
        Document(
            content="""Aaron Aaron ( or ; ""Ahärôn"") is a prophet, high priest, and the brother of Moses in the Abrahamic religions. Knowledge of Aaron, along with his brother Moses, comes exclusively from religious texts, such as the Bible and Quran. The Hebrew Bible relates that, unlike Moses, who grew up in the Egyptian royal court, Aaron and his elder sister Miriam remained with their kinsmen in the eastern border-land of Egypt (Goshen). When Moses first confronted the Egyptian king about the Israelites, Aaron served as his brother's spokesman (""prophet"") to the Pharaoh. Part of the Law (Torah) that Moses received from""",
            meta={"name": "0"},
            id="1",
        ),
        Document(
            content="""Democratic Republic of the Congo to the south. Angola's capital, Luanda, lies on the Atlantic coast in the northwest of the country. Angola, although located in a tropical zone, has a climate that is not characterized for this region, due to the confluence of three factors: As a result, Angola's climate is characterized by two seasons: rainfall from October to April and drought, known as ""Cacimbo"", from May to August, drier, as the name implies, and with lower temperatures. On the other hand, while the coastline has high rainfall rates, decreasing from North to South and from to , with""",
            id="2",
        ),
        Document(
            content="""Schopenhauer, describing him as an ultimately shallow thinker: ""Schopenhauer has quite a crude mind ... where real depth starts, his comes to an end."" His friend Bertrand Russell had a low opinion on the philosopher, and attacked him in his famous ""History of Western Philosophy"" for hypocritically praising asceticism yet not acting upon it. On the opposite isle of Russell on the foundations of mathematics, the Dutch mathematician L. E. J. Brouwer incorporated the ideas of Kant and Schopenhauer in intuitionism, where mathematics is considered a purely mental activity, instead of an analytic activity wherein objective properties of reality are""",
            meta={"name": "1"},
            id="3",
        ),
        Document(
            content="""The Dothraki vocabulary was created by David J. Peterson well in advance of the adaptation. HBO hired the Language Creatio""",
            meta={"name": "2"},
            id="4",
        ),
        Document(
            content="""The title of the episode refers to the Great Sept of Baelor, the main religious building in King's Landing, where the episode's pivotal scene takes place. In the world created by George R. R. Martin""",
            meta={},
            id="5",
        ),
    ]
    return documents


# TODO check if we this works with only "memory" arg
@pytest.mark.parametrize(
    "retriever_with_docs,document_store_with_docs",
    [
        ("dpr", "elasticsearch"),
        ("dpr", "faiss"),
        ("dpr", "memory"),
        ("dpr", "milvus1"),
        ("embedding", "elasticsearch"),
        ("embedding", "faiss"),
        ("embedding", "memory"),
        ("embedding", "milvus1"),
        ("elasticsearch", "elasticsearch"),
        ("es_filter_only", "elasticsearch"),
        ("tfidf", "memory"),
    ],
    indirect=True,
)
def test_retrieval(retriever_with_docs, document_store_with_docs):
    if not isinstance(retriever_with_docs, (ElasticsearchRetriever, ElasticsearchFilterOnlyRetriever, TfidfRetriever)):
        document_store_with_docs.update_embeddings(retriever_with_docs)

    # test without filters
    res = retriever_with_docs.retrieve(query="Who lives in Berlin?")
    assert res[0].content == "My name is Carla and I live in Berlin"
    assert len(res) == 5
    assert res[0].meta["name"] == "filename1"

    # test with filters
    if not isinstance(document_store_with_docs, (FAISSDocumentStore, MilvusDocumentStore)) and not isinstance(
        retriever_with_docs, TfidfRetriever
    ):
        # single filter
        result = retriever_with_docs.retrieve(query="godzilla", filters={"name": ["filename3"]}, top_k=5)
        assert len(result) == 1
        assert type(result[0]) == Document
        assert result[0].content == "My name is Christelle and I live in Paris"
        assert result[0].meta["name"] == "filename3"

        # multiple filters
        result = retriever_with_docs.retrieve(
            query="godzilla", filters={"name": ["filename2"], "meta_field": ["test2", "test3"]}, top_k=5
        )
        assert len(result) == 1
        assert type(result[0]) == Document
        assert result[0].meta["name"] == "filename2"

        result = retriever_with_docs.retrieve(
            query="godzilla", filters={"name": ["filename1"], "meta_field": ["test2", "test3"]}, top_k=5
        )
        assert len(result) == 0


@pytest.mark.elasticsearch
def test_elasticsearch_custom_query():
    client = Elasticsearch()
    client.indices.delete(index="haystack_test_custom", ignore=[404])
    document_store = ElasticsearchDocumentStore(
        index="haystack_test_custom", content_field="custom_text_field", embedding_field="custom_embedding_field"
    )
    documents = [
        {"content": "test_1", "meta": {"year": "2019"}},
        {"content": "test_2", "meta": {"year": "2020"}},
        {"content": "test_3", "meta": {"year": "2021"}},
        {"content": "test_4", "meta": {"year": "2021"}},
        {"content": "test_5", "meta": {"year": "2021"}},
    ]
    document_store.write_documents(documents)

    # test custom "terms" query
    retriever = ElasticsearchRetriever(
        document_store=document_store,
        custom_query="""
            {
                "size": 10, 
                "query": {
                    "bool": {
                        "should": [{
                            "multi_match": {"query": ${query}, "type": "most_fields", "fields": ["content"]}}],
                            "filter": [{"terms": {"year": ${years}}}]}}}""",
    )
    results = retriever.retrieve(query="test", filters={"years": ["2020", "2021"]})
    assert len(results) == 4

    # test custom "term" query
    retriever = ElasticsearchRetriever(
        document_store=document_store,
        custom_query="""
                {
                    "size": 10, 
                    "query": {
                        "bool": {
                            "should": [{
                                "multi_match": {"query": ${query}, "type": "most_fields", "fields": ["content"]}}],
                                "filter": [{"term": {"year": ${years}}}]}}}""",
    )
    results = retriever.retrieve(query="test", filters={"years": "2021"})
    assert len(results) == 3


@pytest.mark.slow
@pytest.mark.parametrize("retriever", ["dpr"], indirect=True)
def test_dpr_embedding(document_store, retriever, docs):

    document_store.return_embedding = True
    document_store.write_documents(docs)
    document_store.update_embeddings(retriever=retriever)
    time.sleep(1)

    # always normalize vector as faiss returns normalized vectors and other document stores do not
    doc_1 = document_store.get_document_by_id("1").embedding
    doc_1 /= np.linalg.norm(doc_1)
    assert len(doc_1) == 768
    assert abs(doc_1[0] - (-0.0250)) < 0.001
    doc_2 = document_store.get_document_by_id("2").embedding
    doc_2 /= np.linalg.norm(doc_2)
    assert abs(doc_2[0] - (-0.0314)) < 0.001
    doc_3 = document_store.get_document_by_id("3").embedding
    doc_3 /= np.linalg.norm(doc_3)
    assert abs(doc_3[0] - (-0.0200)) < 0.001
    doc_4 = document_store.get_document_by_id("4").embedding
    doc_4 /= np.linalg.norm(doc_4)
    assert abs(doc_4[0] - (-0.0070)) < 0.001
    doc_5 = document_store.get_document_by_id("5").embedding
    doc_5 /= np.linalg.norm(doc_5)
    assert abs(doc_5[0] - (-0.0049)) < 0.001


@pytest.mark.slow
@pytest.mark.parametrize("retriever", ["retribert"], indirect=True)
@pytest.mark.embedding_dim(128)
def test_retribert_embedding(document_store, retriever, docs):
    if isinstance(document_store, WeaviateDocumentStore):
        # Weaviate sets the embedding dimension to 768 as soon as it is initialized.
        # We need 128 here and therefore initialize a new WeaviateDocumentStore.
        document_store = WeaviateDocumentStore(
            weaviate_url="http://localhost:8080", index="haystack_test", embedding_dim=128
        )
        document_store.weaviate_client.schema.delete_all()
        document_store._create_schema_and_index_if_not_exist()
    document_store.return_embedding = True
    document_store.write_documents(docs)
    document_store.update_embeddings(retriever=retriever)
    time.sleep(1)

    assert len(document_store.get_document_by_id("1").embedding) == 128
    assert abs(document_store.get_document_by_id("1").embedding[0]) < 0.6
    assert abs(document_store.get_document_by_id("2").embedding[0]) < 0.03
    assert abs(document_store.get_document_by_id("3").embedding[0]) < 0.095
    assert abs(document_store.get_document_by_id("4").embedding[0]) < 0.3
    assert abs(document_store.get_document_by_id("5").embedding[0]) < 0.32


@pytest.mark.slow
@pytest.mark.parametrize("retriever", ["table_text_retriever"], indirect=True)
@pytest.mark.parametrize("document_store", ["elasticsearch"], indirect=True)
@pytest.mark.embedding_dim(512)
def test_table_text_retriever_embedding(document_store, retriever, docs):

    document_store.return_embedding = True
    document_store.write_documents(docs)
    table_data = {
        "Mountain": ["Mount Everest", "K2", "Kangchenjunga", "Lhotse", "Makalu"],
        "Height": ["8848m", "8,611 m", "8 586m", "8 516 m", "8,485m"],
    }
    table = pd.DataFrame(table_data)
    table_doc = Document(content=table, content_type="table", id="6")
    document_store.write_documents([table_doc])
    document_store.update_embeddings(retriever=retriever)
    time.sleep(1)

    doc_1 = document_store.get_document_by_id("1")
    assert len(doc_1.embedding) == 512
    assert abs(doc_1.embedding[0] - (0.0593)) < 0.001
    doc_2 = document_store.get_document_by_id("2")
    assert abs(doc_2.embedding[0] - (0.9031)) < 0.001
    doc_3 = document_store.get_document_by_id("3")
    assert abs(doc_3.embedding[0] - (0.1366)) < 0.001
    doc_4 = document_store.get_document_by_id("4")
    assert abs(doc_4.embedding[0] - (0.0575)) < 0.001
    doc_5 = document_store.get_document_by_id("5")
    assert abs(doc_5.embedding[0] - (0.1486)) < 0.001
    doc_6 = document_store.get_document_by_id("6")
    assert len(doc_6.embedding) == 512
    assert abs(doc_6.embedding[0] - (0.2745)) < 0.001


@pytest.mark.parametrize("retriever", ["dpr"], indirect=True)
@pytest.mark.parametrize("document_store", ["memory"], indirect=True)
def test_dpr_saving_and_loading(retriever, document_store):
    retriever.save("test_dpr_save")

    def sum_params(model):
        s = []
        for p in model.parameters():
            n = p.cpu().data.numpy()
            s.append(np.sum(n))
        return sum(s)

    original_sum_query = sum_params(retriever.query_encoder)
    original_sum_passage = sum_params(retriever.passage_encoder)
    del retriever

    loaded_retriever = DensePassageRetriever.load("test_dpr_save", document_store)

    loaded_sum_query = sum_params(loaded_retriever.query_encoder)
    loaded_sum_passage = sum_params(loaded_retriever.passage_encoder)

    assert abs(original_sum_query - loaded_sum_query) < 0.1
    assert abs(original_sum_passage - loaded_sum_passage) < 0.1

    # comparison of weights (RAM intense!)
    # for p1, p2 in zip(retriever.query_encoder.parameters(), loaded_retriever.query_encoder.parameters()):
    #     assert (p1.data.ne(p2.data).sum() == 0)
    #
    # for p1, p2 in zip(retriever.passage_encoder.parameters(), loaded_retriever.passage_encoder.parameters()):
    #     assert (p1.data.ne(p2.data).sum() == 0)

    # attributes
    assert loaded_retriever.processor.embed_title == True
    assert loaded_retriever.batch_size == 16
    assert loaded_retriever.processor.max_seq_len_passage == 256
    assert loaded_retriever.processor.max_seq_len_query == 64

    # Tokenizer
    assert isinstance(loaded_retriever.passage_tokenizer, DPRContextEncoderTokenizerFast)
    assert isinstance(loaded_retriever.query_tokenizer, DPRQuestionEncoderTokenizerFast)
    assert loaded_retriever.passage_tokenizer.do_lower_case == True
    assert loaded_retriever.query_tokenizer.do_lower_case == True
    assert loaded_retriever.passage_tokenizer.vocab_size == 30522
    assert loaded_retriever.query_tokenizer.vocab_size == 30522
    assert loaded_retriever.passage_tokenizer.model_max_length == 512
    assert loaded_retriever.query_tokenizer.model_max_length == 512


@pytest.mark.parametrize("retriever", ["table_text_retriever"], indirect=True)
@pytest.mark.embedding_dim(512)
def test_table_text_retriever_saving_and_loading(retriever, document_store):
    retriever.save("test_table_text_retriever_save")

    def sum_params(model):
        s = []
        for p in model.parameters():
            n = p.cpu().data.numpy()
            s.append(np.sum(n))
        return sum(s)

    original_sum_query = sum_params(retriever.query_encoder)
    original_sum_passage = sum_params(retriever.passage_encoder)
    original_sum_table = sum_params(retriever.table_encoder)
    del retriever

    loaded_retriever = TableTextRetriever.load("test_table_text_retriever_save", document_store)

    loaded_sum_query = sum_params(loaded_retriever.query_encoder)
    loaded_sum_passage = sum_params(loaded_retriever.passage_encoder)
    loaded_sum_table = sum_params(loaded_retriever.table_encoder)

    assert abs(original_sum_query - loaded_sum_query) < 0.1
    assert abs(original_sum_passage - loaded_sum_passage) < 0.1
    assert abs(original_sum_table - loaded_sum_table) < 0.01

    # attributes
    assert loaded_retriever.processor.embed_meta_fields == ["name", "section_title", "caption"]
    assert loaded_retriever.batch_size == 16
    assert loaded_retriever.processor.max_seq_len_passage == 256
    assert loaded_retriever.processor.max_seq_len_table == 256
    assert loaded_retriever.processor.max_seq_len_query == 64

    # Tokenizer
    assert isinstance(loaded_retriever.passage_tokenizer, DPRContextEncoderTokenizerFast)
    assert isinstance(loaded_retriever.table_tokenizer, DPRContextEncoderTokenizerFast)
    assert isinstance(loaded_retriever.query_tokenizer, DPRQuestionEncoderTokenizerFast)
    assert loaded_retriever.passage_tokenizer.do_lower_case == True
    assert loaded_retriever.table_tokenizer.do_lower_case == True
    assert loaded_retriever.query_tokenizer.do_lower_case == True
    assert loaded_retriever.passage_tokenizer.vocab_size == 30522
    assert loaded_retriever.table_tokenizer.vocab_size == 30522
    assert loaded_retriever.query_tokenizer.vocab_size == 30522
    assert loaded_retriever.passage_tokenizer.model_max_length == 512
    assert loaded_retriever.table_tokenizer.model_max_length == 512
    assert loaded_retriever.query_tokenizer.model_max_length == 512


@pytest.mark.embedding_dim(128)
def test_table_text_retriever_training(document_store):
    retriever = TableTextRetriever(
        document_store=document_store,
        query_embedding_model="prajjwal1/bert-tiny",
        passage_embedding_model="prajjwal1/bert-tiny",
        table_embedding_model="prajjwal1/bert-tiny",
        use_gpu=False,
    )

    retriever.train(
        data_dir=SAMPLES_PATH / "mmr",
        train_filename="sample.json",
        n_epochs=1,
        n_gpu=0,
        save_dir="test_table_text_retriever_train",
    )

    # Load trained model
    retriever = TableTextRetriever.load(load_dir="test_table_text_retriever_train", document_store=document_store)


@pytest.mark.elasticsearch
def test_elasticsearch_highlight():
    client = Elasticsearch()
    client.indices.delete(index="haystack_hl_test", ignore=[404])

    # Mapping the content and title field as "text" perform search on these both fields.
    document_store = ElasticsearchDocumentStore(
        index="haystack_hl_test",
        content_field="title",
        custom_mapping={"mappings": {"properties": {"content": {"type": "text"}, "title": {"type": "text"}}}},
    )
    documents = [
        {
            "title": "Green tea components",
            "meta": {
                "content": "The green tea plant contains a range of healthy compounds that make it into the final drink"
            },
            "id": "1",
        },
        {
            "title": "Green tea catechin",
            "meta": {"content": "Green tea contains a catechin called epigallocatechin-3-gallate (EGCG)."},
            "id": "2",
        },
        {
            "title": "Minerals in Green tea",
            "meta": {"content": "Green tea also has small amounts of minerals that can benefit your health."},
            "id": "3",
        },
        {
            "title": "Green tea Benefits",
            "meta": {"content": "Green tea does more than just keep you alert, it may also help boost brain function."},
            "id": "4",
        },
    ]
    document_store.write_documents(documents)

    # Enabled highlighting on "title"&"content" field only using custom query
    retriever_1 = ElasticsearchRetriever(
        document_store=document_store,
        custom_query="""{
            "size": 20,
            "query": {
                "bool": {
                    "should": [
                        {
                            "multi_match": {
                                "query": ${query},
                                "fields": [
                                    "content^3",
                                    "title^5"
                                ]
                            }
                        }
                    ]
                }
            },
            "highlight": {
                "pre_tags": [
                    "**"
                ],
                "post_tags": [
                    "**"
                ],
                "number_of_fragments": 3,
                "fragment_size": 5,
                "fields": {
                    "content": {},
                    "title": {}
                }
            }
        }""",
    )
    results = retriever_1.retrieve(query="is green tea healthy")

    assert len(results[0].meta["highlighted"]) == 2
    assert results[0].meta["highlighted"]["title"] == ["**Green**", "**tea** components"]
    assert results[0].meta["highlighted"]["content"] == ["The **green**", "**tea** plant", "range of **healthy**"]

    # Enabled highlighting on "title" field only using custom query
    retriever_2 = ElasticsearchRetriever(
        document_store=document_store,
        custom_query="""{
            "size": 20,
            "query": {
                "bool": {
                    "should": [
                        {
                            "multi_match": {
                                "query": ${query},
                                "fields": [
                                    "content^3",
                                    "title^5"
                                ]
                            }
                        }
                    ]
                }
            },
            "highlight": {
                "pre_tags": [
                    "**"
                ],
                "post_tags": [
                    "**"
                ],
                "number_of_fragments": 3,
                "fragment_size": 5,
                "fields": {
                    "title": {}
                }
            }
        }""",
    )
    results = retriever_2.retrieve(query="is green tea healthy")

    assert len(results[0].meta["highlighted"]) == 1
    assert results[0].meta["highlighted"]["title"] == ["**Green**", "**tea** components"]