147 lines
5.4 KiB
Python
147 lines
5.4 KiB
Python
from abc import abstractmethod
|
|
from typing import Union, List, Optional
|
|
|
|
from sentence_transformers import SentenceTransformer
|
|
from transformers import AutoTokenizer, AutoModel
|
|
|
|
import numpy as np
|
|
import torch
|
|
import re
|
|
|
|
from code_search.model.unixcoder import UniXcoder
|
|
|
|
|
|
class BaseEmbeddingsProvider:
|
|
|
|
@abstractmethod
|
|
def embed_code(
|
|
self, code: Optional[str] = None, docstring: Optional[str] = None
|
|
) -> np.array:
|
|
"""Converts code and/or docstring to vector"""
|
|
|
|
|
|
class AutoModelEmbeddingsProvider(BaseEmbeddingsProvider):
|
|
def __init__(
|
|
self,
|
|
model_name: str = "microsoft/codebert-base",
|
|
tokenizer_name: Optional[str] = None,
|
|
max_tokens: int = 512,
|
|
):
|
|
if tokenizer_name is None:
|
|
tokenizer_name = model_name
|
|
self.tokenizer = AutoTokenizer.from_pretrained(tokenizer_name)
|
|
self.model = AutoModel.from_pretrained(model_name)
|
|
self.max_tokens = max_tokens
|
|
self.model_name = model_name
|
|
|
|
def embed_code(
|
|
self, code: Optional[str] = None, docstring: Optional[str] = None
|
|
) -> np.array:
|
|
token_ids = self.get_token_ids(code, docstring)[:512]
|
|
context_embeddings = (
|
|
self.model(torch.tensor(token_ids)[None, :])[0].squeeze().detach().numpy()
|
|
)
|
|
if 1 == len(context_embeddings.shape):
|
|
return context_embeddings
|
|
return np.mean(context_embeddings, axis=0)
|
|
|
|
def get_token_ids(
|
|
self, code: Optional[str] = None, docstring: Optional[str] = None
|
|
) -> Union[int, List[int]]:
|
|
tokens = self.get_tokens(code, docstring)
|
|
tokens_ids = self.tokenizer.convert_tokens_to_ids(tokens)
|
|
return tokens_ids
|
|
|
|
def get_tokens(
|
|
self, code: Optional[str] = None, docstring: Optional[str] = None
|
|
) -> List[str]:
|
|
# Maximum number of tokens has to include the separators used by CodeBERT
|
|
max_tokens = self.max_tokens - 3
|
|
|
|
code_tokens = []
|
|
if code is not None:
|
|
code_tokens = self.tokenizer.tokenize(
|
|
code, max_length=max_tokens, truncation=True
|
|
)
|
|
docstring_tokens = []
|
|
if docstring is not None:
|
|
docstring_tokens = self.tokenizer.tokenize(
|
|
docstring, max_length=max_tokens, truncation=True
|
|
)
|
|
|
|
# If both code and docstring is provided, we need to cut off some
|
|
# tokens above the limit. Here there preference is to remove the code
|
|
# as it should be longer in general.
|
|
n_code_tokens, n_doc_tokens = (
|
|
min(len(code_tokens), max_tokens),
|
|
min(len(docstring_tokens), max_tokens),
|
|
)
|
|
n_code_tokens -= n_doc_tokens
|
|
|
|
# Build all the tokens using some possible separators. The separators
|
|
# are aligned to CodeBERT model, but if a selected transformer does not
|
|
# have them, everything should also work.
|
|
tokens = []
|
|
if hasattr(self.tokenizer, "cls_token"):
|
|
tokens.append(self.tokenizer.cls_token)
|
|
tokens.extend(code_tokens[:n_code_tokens])
|
|
if hasattr(self.tokenizer, "sep_token"):
|
|
tokens.append(self.tokenizer.sep_token)
|
|
tokens.extend(docstring_tokens[:n_doc_tokens])
|
|
if hasattr(self.tokenizer, "eos_token"):
|
|
tokens.append(self.tokenizer.eos_token)
|
|
return tokens
|
|
|
|
def __str__(self):
|
|
return self.model_name
|
|
|
|
|
|
class UniXcoderEmbeddingsProvider(BaseEmbeddingsProvider):
|
|
def __init__(self, device: Optional[str] = None):
|
|
default_device = "cuda" if torch.cuda.is_available() else "cpu"
|
|
self.device = torch.device(default_device if device is None else device)
|
|
self.model = UniXcoder("microsoft/unixcoder-base")
|
|
self.model.to(self.device)
|
|
self.model_name = "microsoft/unixcoder-base"
|
|
|
|
def embed_code(
|
|
self, code: Optional[str] = None, docstring: Optional[str] = None
|
|
) -> np.array:
|
|
tokens_ids = self.model.tokenize(
|
|
[f"{docstring or ''} {code or ''}"], max_length=512, mode="<encoder-only>"
|
|
)
|
|
source_ids = torch.tensor(tokens_ids).to(self.device)
|
|
_, func_embedding = self.model(source_ids)
|
|
vector = func_embedding.detach().cpu().numpy()[0]
|
|
return vector
|
|
|
|
|
|
class SentenceTransformerEmbeddingsProvider(BaseEmbeddingsProvider):
|
|
camel_case_regex = re.compile(r"([a-z\s])([A-Z])")
|
|
underscore_regex = re.compile(r"([a-z])_([a-z])")
|
|
special_chars_regex = re.compile(r"\(|\)|\{|\}|\<|\>|\[|\]|\&|::|;")
|
|
method_call_regex = re.compile(r"\.([a-z])")
|
|
multiple_white_char_regex = re.compile(r"\s{2,}")
|
|
|
|
def __init__(self, sentence_transformer_name: str):
|
|
self.model = SentenceTransformer(sentence_transformer_name)
|
|
self.model_name = sentence_transformer_name
|
|
|
|
def embed_code(
|
|
self, code: Optional[str] = None, docstring: Optional[str] = None
|
|
) -> np.array:
|
|
inputs = []
|
|
if docstring is not None:
|
|
inputs.append(docstring)
|
|
if code is not None:
|
|
inputs.append(code)
|
|
return self.model.encode(self._preprocess_text(" ".join(inputs)))
|
|
|
|
def _preprocess_text(self, text: str) -> str:
|
|
text = self.camel_case_regex.sub("\\1 \\2", text)
|
|
text = self.underscore_regex.sub("\\1 \\2", text)
|
|
text = self.special_chars_regex.sub(" ", text)
|
|
text = self.method_call_regex.sub(" \\1", text)
|
|
text = self.multiple_white_char_regex.sub(" ", text)
|
|
return text
|