File size: 10,485 Bytes
4ff79c6 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 | # SPDX-FileCopyrightText: 2022-present deepset GmbH <info@deepset.ai>
#
# SPDX-License-Identifier: Apache-2.0
import itertools
from collections import defaultdict
from enum import Enum
from math import inf
from typing import Any, Dict, List, Optional, Union
from haystack import Document, component, default_from_dict, default_to_dict, logging
from haystack.core.component.types import Variadic
logger = logging.getLogger(__name__)
class JoinMode(Enum):
"""
Enum for join mode.
"""
CONCATENATE = "concatenate"
MERGE = "merge"
RECIPROCAL_RANK_FUSION = "reciprocal_rank_fusion"
DISTRIBUTION_BASED_RANK_FUSION = "distribution_based_rank_fusion"
def __str__(self):
return self.value
@staticmethod
def from_str(string: str) -> "JoinMode":
"""
Convert a string to a JoinMode enum.
"""
enum_map = {e.value: e for e in JoinMode}
mode = enum_map.get(string)
if mode is None:
msg = f"Unknown join mode '{string}'. Supported modes in DocumentJoiner are: {list(enum_map.keys())}"
raise ValueError(msg)
return mode
@component
class DocumentJoiner:
"""
Joins multiple lists of documents into a single list.
It supports different join modes:
- concatenate: Keeps the highest-scored document in case of duplicates.
- merge: Calculates a weighted sum of scores for duplicates and merges them.
- reciprocal_rank_fusion: Merges and assigns scores based on reciprocal rank fusion.
- distribution_based_rank_fusion: Merges and assigns scores based on scores distribution in each Retriever.
### Usage example:
```python
document_store = InMemoryDocumentStore()
p = Pipeline()
p.add_component(instance=InMemoryBM25Retriever(document_store=document_store), name="bm25_retriever")
p.add_component(
instance=SentenceTransformersTextEmbedder(model="sentence-transformers/all-MiniLM-L6-v2"),
name="text_embedder",
)
p.add_component(instance=InMemoryEmbeddingRetriever(document_store=document_store), name="embedding_retriever")
p.add_component(instance=DocumentJoiner(), name="joiner")
p.connect("bm25_retriever", "joiner")
p.connect("embedding_retriever", "joiner")
p.connect("text_embedder", "embedding_retriever")
query = "What is the capital of France?"
p.run(data={"query": query})
```
"""
def __init__(
self,
join_mode: Union[str, JoinMode] = JoinMode.CONCATENATE,
weights: Optional[List[float]] = None,
top_k: Optional[int] = None,
sort_by_score: bool = True,
):
"""
Creates a DocumentJoiner component.
:param join_mode:
Specifies the join mode to use. Available modes:
- `concatenate`: Keeps the highest-scored document in case of duplicates.
- `merge`: Calculates a weighted sum of scores for duplicates and merges them.
- `reciprocal_rank_fusion`: Merges and assigns scores based on reciprocal rank fusion.
- `distribution_based_rank_fusion`: Merges and assigns scores based on scores
distribution in each Retriever.
:param weights:
Assign importance to each list of documents to influence how they're joined.
This parameter is ignored for
`concatenate` or `distribution_based_rank_fusion` join modes.
Weight for each list of documents must match the number of inputs.
:param top_k:
The maximum number of documents to return.
:param sort_by_score:
If `True`, sorts the documents by score in descending order.
If a document has no score, it is handled as if its score is -infinity.
"""
if isinstance(join_mode, str):
join_mode = JoinMode.from_str(join_mode)
join_mode_functions = {
JoinMode.CONCATENATE: self._concatenate,
JoinMode.MERGE: self._merge,
JoinMode.RECIPROCAL_RANK_FUSION: self._reciprocal_rank_fusion,
JoinMode.DISTRIBUTION_BASED_RANK_FUSION: self._distribution_based_rank_fusion,
}
self.join_mode_function = join_mode_functions[join_mode]
self.join_mode = join_mode
self.weights = [float(i) / sum(weights) for i in weights] if weights else None
self.top_k = top_k
self.sort_by_score = sort_by_score
@component.output_types(documents=List[Document])
def run(self, documents: Variadic[List[Document]], top_k: Optional[int] = None):
"""
Joins multiple lists of Documents into a single list depending on the `join_mode` parameter.
:param documents:
List of list of documents to be merged.
:param top_k:
The maximum number of documents to return. Overrides the instance's `top_k` if provided.
:returns:
A dictionary with the following keys:
- `documents`: Merged list of Documents
"""
output_documents = []
documents = list(documents)
output_documents = self.join_mode_function(documents)
if self.sort_by_score:
output_documents = sorted(
output_documents, key=lambda doc: doc.score if doc.score is not None else -inf, reverse=True
)
if any(doc.score is None for doc in output_documents):
logger.info(
"Some of the Documents DocumentJoiner got have score=None. It was configured to sort Documents by "
"score, so those with score=None were sorted as if they had a score of -infinity."
)
if top_k:
output_documents = output_documents[:top_k]
elif self.top_k:
output_documents = output_documents[: self.top_k]
return {"documents": output_documents}
def _concatenate(self, document_lists: List[List[Document]]) -> List[Document]:
"""
Concatenate multiple lists of Documents and return only the Document with the highest score for duplicates.
"""
output = []
docs_per_id = defaultdict(list)
for doc in itertools.chain.from_iterable(document_lists):
docs_per_id[doc.id].append(doc)
for docs in docs_per_id.values():
doc_with_best_score = max(docs, key=lambda doc: doc.score if doc.score else -inf)
output.append(doc_with_best_score)
return output
def _merge(self, document_lists: List[List[Document]]) -> List[Document]:
"""
Merge multiple lists of Documents and calculate a weighted sum of the scores of duplicate Documents.
"""
scores_map: dict = defaultdict(int)
documents_map = {}
weights = self.weights if self.weights else [1 / len(document_lists)] * len(document_lists)
for documents, weight in zip(document_lists, weights):
for doc in documents:
scores_map[doc.id] += (doc.score if doc.score else 0) * weight
documents_map[doc.id] = doc
for doc in documents_map.values():
doc.score = scores_map[doc.id]
return list(documents_map.values())
def _reciprocal_rank_fusion(self, document_lists: List[List[Document]]) -> List[Document]:
"""
Merge multiple lists of Documents and assign scores based on reciprocal rank fusion.
The constant k is set to 61 (60 was suggested by the original paper,
plus 1 as python lists are 0-based and the paper used 1-based ranking).
"""
k = 61
scores_map: dict = defaultdict(int)
documents_map = {}
weights = self.weights if self.weights else [1 / len(document_lists)] * len(document_lists)
# Calculate weighted reciprocal rank fusion score
for documents, weight in zip(document_lists, weights):
for rank, doc in enumerate(documents):
scores_map[doc.id] += (weight * len(document_lists)) / (k + rank)
documents_map[doc.id] = doc
# Normalize scores. Note: len(results) / k is the maximum possible score,
# achieved by being ranked first in all doc lists with non-zero weight.
for _id in scores_map:
scores_map[_id] /= len(document_lists) / k
for doc in documents_map.values():
doc.score = scores_map[doc.id]
return list(documents_map.values())
def _distribution_based_rank_fusion(self, document_lists: List[List[Document]]) -> List[Document]:
"""
Merge multiple lists of Documents and assign scores based on Distribution-Based Score Fusion.
(https://medium.com/plain-simple-software/distribution-based-score-fusion-dbsf-a-new-approach-to-vector-search-ranking-f87c37488b18)
If a Document is in more than one retriever, the one with the highest score is used.
"""
for documents in document_lists:
scores_list = []
for doc in documents:
scores_list.append(doc.score if doc.score is not None else 0)
mean_score = sum(scores_list) / len(scores_list)
std_dev = (sum((x - mean_score) ** 2 for x in scores_list) / len(scores_list)) ** 0.5
min_score = mean_score - 3 * std_dev
max_score = mean_score + 3 * std_dev
delta_score = max_score - min_score
for doc in documents:
doc.score = (doc.score - min_score) / delta_score if delta_score != 0.0 else 0.0
# if all docs have the same score delta_score is 0, the docs are uninformative for the query
output = self._concatenate(document_lists=document_lists)
return output
def to_dict(self) -> Dict[str, Any]:
"""
Serializes the component to a dictionary.
:returns:
Dictionary with serialized data.
"""
return default_to_dict(
self,
join_mode=str(self.join_mode),
weights=self.weights,
top_k=self.top_k,
sort_by_score=self.sort_by_score,
)
@classmethod
def from_dict(cls, data: Dict[str, Any]) -> "DocumentJoiner":
"""
Deserializes the component from a dictionary.
:param data:
The dictionary to deserialize from.
:returns:
The deserialized component.
"""
return default_from_dict(cls, data)
|