def embeddings(dbfile, vectors, maxsize):
"""
Builds a sentence embeddings index.
Args:
dbfile: input SQLite file
vectors: vector path
maxsize: maximum number of documents to process
Returns:
embeddings index
"""
embeddings = Embeddings({"path": vectors,
"scoring": "bm25",
"pca": 3,
"quantize": True})
# Build scoring index if scoring method provided
if embeddings.config.get("scoring"):
embeddings.score(Index.stream(dbfile, maxsize))
# Build embeddings index
embeddings.index(Index.stream(dbfile, maxsize))
return embeddings
def embeddings(dbfile):
"""
Builds a sentence embeddings index.
Args:
dbfile: input SQLite file
Returns:
embeddings index
"""
embeddings = Embeddings({
"path":
Models.vectorPath("stackexchange-300d.magnitude"),
"storevectors":
True,
"scoring":
"bm25",
"pca":
3,
"quantize":
True
})
# Build scoring index if scoring method provided
if embeddings.config.get("scoring"):
embeddings.score(Index.stream(dbfile))
# Build embeddings index
embeddings.index(Index.stream(dbfile))
return embeddings
def testWords(self):
"""
Test embeddings backed by word vectors
"""
# Initialize model path
path = os.path.join(tempfile.gettempdir(), "model")
os.makedirs(path, exist_ok=True)
# Build tokens file
with tempfile.NamedTemporaryFile(mode="w", delete=False) as output:
tokens = output.name
for x in self.data:
output.write(x + "\n")
# Word vectors path
vectors = os.path.join(path, "test-300d")
# Build word vectors, if they don't already exist
WordVectors.build(tokens, 300, 1, vectors)
# Create dataset
data = [(x, row, None) for x, row in enumerate(self.data)]
# Create embeddings model, backed by word vectors
embeddings = Embeddings({
"path": vectors + ".magnitude",
"storevectors": True,
"scoring": "bm25",
"pca": 3,
"quantize": True
})
# Call scoring and index methods
embeddings.score(data)
embeddings.index(data)
# Test search
self.assertIsNotNone(embeddings.search("win", 1))
# Generate temp file path
index = os.path.join(tempfile.gettempdir(), "wembeddings")
# Test save/load
embeddings.save(index)
embeddings.load(index)
# Test search
self.assertIsNotNone(embeddings.search("win", 1))
def train(vector, score):
"""
Trains an Embeddings model on STS dev + train data.
Args:
vector: word vector model path
score: scoring method (bm25, sif, tfidf or None for averaging)
Returns:
trained Embeddings model
"""
print("Building model")
embeddings = Embeddings({
"path": Models.vectorPath(vector),
"scoring": score,
"pca": 3
})
rows1 = STS.read(Models.testPath("stsbenchmark", "sts-dev.csv"))
rows2 = STS.read(Models.testPath("stsbenchmark", "sts-train.csv"))
rows = rows1 + rows2
documents = []
for row in rows:
tokens = Tokenizer.tokenize(row[2] + " " + row[3])
if tokens:
documents.append((row[0], tokens, None))
else:
print("Skipping all stop word string: ", row)
# Build scoring index if scoring method provided
if embeddings.config.get("scoring"):
embeddings.score(documents)
# Build embeddings index
embeddings.index(documents)
return embeddings
def embeddings(dbfile, vectors, maxsize):
"""
Builds a sentence embeddings index.
Args:
dbfile: input SQLite file
vectors: path to vectors file or configuration
maxsize: maximum number of documents to process
Returns:
embeddings index
"""
# Read config and create Embeddings instance
embeddings = Embeddings(Index.config(vectors))
# Build scoring index if scoring method provided
if embeddings.config.get("scoring"):
embeddings.score(Index.stream(dbfile, maxsize))
# Build embeddings index
embeddings.index(Index.stream(dbfile, maxsize))
return embeddings
