makaleChatbotu / database.py
yonkasoft's picture
Upload 2 files
7044ef9 verified
raw
history blame
6.98 kB
from datasets import load_dataset
import pandas as pd
import torch
from torch.utils.data import TensorDataset, DataLoader, RandomSampler, SequentialSampler
from transformers import BertTokenizer, BertForQuestionAnswering, BertConfig,AutoModelForCausalLM
from pymongo import MongoClient
import torchtext
torchtext.disable_torchtext_deprecation_warning()
from torchtext.data import get_tokenizer
from yeni_tokenize import TokenizerProcessor
class Database:
# MongoDB connection settings
def get_mongodb(database_name='yeniDatabase', collection_name='test', host='localhost', port=27017):
"""
MongoDB connection and collection selection
"""
client = MongoClient(f'mongodb://{host}:{port}/')
db = client[database_name]
collection = db[collection_name]
return collection
@staticmethod
def get_mongodb():
# MongoDB bağlantı bilgilerini döndürecek şekilde tanımlanmalıdır.
return 'mongodb://localhost:27017/', 'yeniDatabase', 'train'
@staticmethod
def get_input_texts():
# MongoDB bağlantı bilgilerini alma
mongo_url, db_name, collection_name = Database.get_mongodb()
# MongoDB'ye bağlanma
client = MongoClient(mongo_url)
db = client[db_name]
collection = db[collection_name]
# Sorguyu tanımlama
query = {"Prompt": {"$exists": True}}
# Sorguyu çalıştırma ve dökümanları çekme
cursor = collection.find(query, {"Prompt": 1, "_id": 0})
# Cursor'ı döküman listesine dönüştürme
input_texts_from_db = [doc['Prompt'] for doc in cursor]
# Input text'leri döndürme
# Düz metin listesine dönüştürme
return input_texts_from_db
input_text= get_input_texts()
print("metinler yazılıyor:")
for text in input_text:
print(text)
@staticmethod
def get_output_texts():
# MongoDB bağlantı bilgilerini alma
mongo_url, db_name, collection_name = Database.get_mongodb()
# MongoDB'ye bağlanma
client = MongoClient(mongo_url)
db = client[db_name]
collection = db[collection_name]
# Sorguyu tanımlama
query = {"Response": {"$exists": True}}
# Sorguyu çalıştırma ve dökümanları çekme
cursor = collection.find(query, {"Response": 1, "_id": 0})
# Cursor'ı döküman listesine dönüştürme
output_texts_from_db = [doc['Response'] for doc in cursor]
#output metin listesine çevirme
return output_texts_from_db
@staticmethod
def get_average_prompt_token_length():
# MongoDB bağlantı bilgilerini alma
mongo_url, db_name, collection_name = Database.get_mongodb()
# MongoDB'ye bağlanma
client = MongoClient(mongo_url)
db = client[db_name]
collection = db[collection_name]
# Tüm dökümanları çekme ve 'prompt_token_length' alanını alma
docs = collection.find({}, {'Prompt_token_length': 1})
# 'prompt_token_length' değerlerini toplama ve sayma
total_length = 0
count = 0
for doc in docs:
if 'Prompt_token_length' in doc:
total_length += doc['Prompt_token_length']
count += 1
# Ortalama hesaplama
average_length = total_length / count if count > 0 else 0
return int(average_length)
# Tokenizer ve Modeli yükleme
"""
class TokenizerProcessor:
def __init__(self, tokenizer_name='bert-base-uncased'):
self.tokenizer = BertTokenizer.from_pretrained(tokenizer_name)
def tokenize_and_encode(self, input_texts, output_texts, max_length=100):
encoded = self.tokenizer.batch_encode_plus(
text_pair=list(zip(input_texts, output_texts)),
padding='max_length',
truncation=True,
max_length=max_length,
return_attention_mask=True,
return_tensors='pt'
)
return encoded
paraphrase = tokenizer.encode_plus(sequence_0, sequence_2, return_tensors="pt")
not_paraphrase = tokenizer.encode_plus(sequence_0, sequence_1, return_tensors="pt")
paraphrase_classification_logits = model(**paraphrase)[0]
not_paraphrase_classification_logits = model(**not_paraphrase)[0]
def custom_padding(self, input_ids_list, max_length=100, pad_token_id=0):
padded_inputs = []
for ids in input_ids_list:
if len(ids) < max_length:
padded_ids = ids + [pad_token_id] * (max_length - len(ids))
else:
padded_ids = ids[:max_length]
padded_inputs.append(padded_ids)
return padded_inputs
def pad_and_truncate_pairs(self, input_texts, output_texts, max_length=100):
#input ve output verilerinin uzunluğunu eşitleme
inputs = self.tokenizer(input_texts, padding=False, truncation=False, return_tensors=None)
outputs = self.tokenizer(output_texts, padding=False, truncation=False, return_tensors=None)
input_ids = self.custom_padding(inputs['input_ids'], max_length, self.tokenizer.pad_token_id)
output_ids = self.custom_padding(outputs['input_ids'], max_length, self.tokenizer.pad_token_id)
input_ids_tensor = torch.tensor(input_ids)
output_ids_tensor = torch.tensor(output_ids)
input_attention_mask = (input_ids_tensor != self.tokenizer.pad_token_id).long()
output_attention_mask = (output_ids_tensor != self.tokenizer.pad_token_id).long()
return {
'input_ids': input_ids_tensor,
'input_attention_mask': input_attention_mask,
'output_ids': output_ids_tensor,
'output_attention_mask': output_attention_mask
}
"""
#cümleleri teker teker input ve output verilerinden çekmem gerekiyor
#def tokenize_and_pad_sequences(sequence_1,sequence2,)
"""class DataPipeline:
def __init__(self, tokenizer_name='bert-base-uncased', max_length=100):
self.tokenizer_processor = TokenizerProcessor(tokenizer_name)
self.max_length = max_length
def prepare_data(self):
input_texts = Database.get_input_texts()
output_texts = Database.get_output_texts()
encoded_data = self.tokenizer_processor.pad_and_truncate_pairs(input_texts, output_texts, self.max_length)
return encoded_data
def tokenize_texts(self, texts):
return [self.tokenize(text) for text in texts]
def encode_texts(self, texts):
return [self.encode(text, self.max_length) for text in texts]
# Example Usage
if __name__ == "__main__":
data_pipeline = DataPipeline()
encoded_data = data_pipeline.prepare_data()
print(encoded_data)
"""