from pinecone_text.sparse import SpladeEncoder import re import torch import torch.nn.functional as F from transformers import CLIPModel, CLIPProcessor, CLIPTokenizer import logging class EmbeddingModels: def __init__(self, device="cuda" if torch.cuda.is_available() else "cpu"): self.device = device logging.info(f'Using Device {self.device}') self.sparse_model = SpladeEncoder(device=self.device) self.img_model_ID = "openai/clip-vit-large-patch14" self.img_model, self.img_processor, self.img_tokenizer = self.get_image_model_info(self.img_model_ID) logging.info("Model Loaded") def get_image_model_info(self, model_ID): model = CLIPModel.from_pretrained(model_ID).to(self.device) processor = CLIPProcessor.from_pretrained(model_ID) tokenizer = CLIPTokenizer.from_pretrained(model_ID) return model, processor, tokenizer def get_single_image_embedding(self, my_image): image = self.img_processor( text=None, images=my_image, return_tensors="pt" )["pixel_values"].to(self.device) embedding = self.img_model.get_image_features(image) logging.info("Embeddings Created") embeddings = F.normalize(embedding, p=2, dim=1) logging.info("Embeddings Normalized") values = embeddings[0].tolist() return values def preprocessing_patent_data(self,text): # Removing Common tags in patent pattern0 = r'\b(SUBSTITUTE SHEET RULE 2 SUMMARY OF THE INVENTION|BRIEF DESCRIPTION OF PREFERRED EMBODIMENTS|BRIEF DESCRIPTION OF THE DRAWINGS/FIGURES|BEST MODE FOR CARRYING OUT THE INVENTION|BACKGROUND AND SUMMARY OF THE INVENTION|FIELD AND BACKGROUND OF THE INVENTION|BACKGROUND OF THE PRESENT INVENTION|FIELD AND BACKGROUND OF INVENTION|STAND DER TECHNIK- BACKGROUND ART|BRIEF DESCRIPTION OF THE DRAWINGS|DESCRIPTION OF THE RELATED ART|BRIEF SUMMARY OF THE INVENTION|UTILITY MODEL CLAIMS A CONTENT|DESCRIPTION OF BACKGROUND ART|BRIEF DESCRIPTION OF DRAWINGS|BACKGROUND OF THE INVENTION|BACKGROUND TO THE INVENTION|TÉCNICA ANTERIOR- PRIOR ART|DISCLOSURE OF THE INVENTION|BRIEF SUMMARY OF INVENTION|BACKGROUND OF RELATED ART|SUMMARY OF THE DISCLOSURE|SUMMARY OF THE INVENTIONS|SUMMARY OF THE INVENTION|OBJECTS OF THE INVENTION|THE CONTENT OF INVENTION|DISCLOSURE OF INVENTION|Disclosure of Invention|Complete Specification|RELATED BACKGROUND ART|BACKGROUND INFORMATION|BACKGROUND TECHNOLOGY|DETAILED DESCRIPTION|SUMMARY OF INVENTION|DETAILED DESCRIPTION|PROBLEM TO BE SOLVED|EFFECT OF INVENTION|WHAT IS CLAIMED IS|What is claimed is|What is Claim is|SUBSTITUTE SHEET|SELECTED DRAWING|BACK GROUND ART|BACKGROUND ART|Background Art|JPO&INPIT|CONSTITUTION|DEFINITIONS|Related Art|BACKGROUND|JPO&INPIT|JPO&NCIPI|COPYRIGHT|SOLUTION|SUMMARY)\b' text = re.sub(pattern0, '[SEP]', text, flags=re.IGNORECASE) text = ' '.join(text.split()) # Removing all tags between Heading to /Heading and id= regex = r'<\s*heading[^>]*>(.*?)<\s*/\s*heading>|<[^<]+>|id=\"p-\d+\"|:' result = re.sub(regex, '[SEP]', text, flags=re.IGNORECASE) # find_formula_names from pat text to exclude it from below logic regex chemical_list = [] pattern1 = r'\b((?:(?:H|He|Li|Be|B|C|N|O|F|Ne|Na|Mg|Al|Si|P|S|Cl|Ar|K|Ca|Sc|Ti|V|Cr|Mn|Fe|Co|Ni|Cu|Zn|Ga|Ge|As|Se|Br|Kr|Rb|Sr|Y|Zr|Nb|Mo|Tc|Ru|Rh|Pd|Ag|Cd|In|Sn|Sb|Te|I|Xe|Cs|Ba|La|Hf|Ta|W|Re|Os|Ir|Pt|Au|Hg|Tl|Pb|Bi|Po|At|Rn|Fr|Ra|Ac|Rf|Db|Sg|Bh|Hs|Mt|Ds|Rg|Cn|Nh|Fl|Mc|Lv|Ts|Og|Ce|Pr|Nd|Pm|Sm|Eu|Gd|Tb|Dy|Ho|Er|Tm|Yb|Lu|Th|Pa|U|Np|Pu|Am|Cm|Bk|Cf|Es|Fm|Md|No|Lr)\d*)+)\b' formula_names = re.findall(pattern1, result) for formula in formula_names: if len(formula)>=2: chemical_list.append(formula) # print("chemical_list:", chemical_list) # Remove numbers and alphanum inside brackets excluding chemical forms pattern2 = r"\((?![A-Za-z]+\))[\w\d\s,-]+\)|\([A-Za-z]\)" def keep_strings(text): matched = text.group(0) if any(item in matched for item in chemical_list): return matched return ' ' cleaned_text = re.sub(pattern2, keep_strings, result) cleaned_text = ' '.join(cleaned_text.split()) cleaned_text= re.sub("(\[SEP\]+\s*)+", ' ', cleaned_text, flags=re.IGNORECASE) # below new logic to remove chemical compounds (eg.chemical- polymerizable compounds) p_text2=re.sub('[\—\-\═\=]', ' ', cleaned_text) pattern1 = r'\b((?:(?:H|He|Li|Be|B|C|N|O|F|Ne|Na|Mg|Al|Si|P|S|Cl|Ar|K|Ca|Sc|Ti|V|Cr|Mn|Fe|Co|Ni|Cu|Zn|Ga|Ge|As|Se|Br|Kr|Rb|Sr|Y|Zr|Nb|Mo|Tc|Ru|Rh|Pd|Ag|Cd|In|Sn|Sb|Te|I|Xe|Cs|Ba|La|Hf|Ta|W|Re|Os|Ir|Pt|Au|Hg|Tl|Pb|Bi|Po|At|Rn|Fr|Ra|Ac|Rf|Db|Sg|Bh|Hs|Mt|Ds|Rg|Cn|Nh|Fl|Mc|Lv|Ts|Og|Ce|Pr|Nd|Pm|Sm|Eu|Gd|Tb|Dy|Ho|Er|Tm|Yb|Lu|Th|Pa|U|Np|Pu|Am|Cm|Bk|Cf|Es|Fm|Md|No|Lr)\d*)+)\b' cleaned_text = re.sub(pattern1, "", p_text2) cleaned_text = re.sub(' ,+|, +', ' ', cleaned_text) cleaned_text = re.sub(' +', ' ', cleaned_text) cleaned_text = re.sub('\.+', '.', cleaned_text) cleaned_text = re.sub('[0-9] [0-9] +', ' ', cleaned_text) cleaned_text = re.sub('( )', ' ', cleaned_text) cleaned_text=cleaned_text.strip() return cleaned_text def get_single_sparse_text_embedding(self, df_chunk): df_chunk = self.preprocessing_patent_data(df_chunk) txt_sp = self.sparse_model.encode_documents(df_chunk) # tensor = torch.tensor(txt_sp['values']) # normalized_tensor = F.normalize(tensor, p=2.0, dim=0, eps=1e-12) # values = normalized_tensor.tolist() # # Update the sparse_vector with normalized values # normalized_sparse_vector = { # 'indices': txt_sp['indices'], # 'values': values # } return txt_sp def normalize_sparse_vector_values(self,sparse_vector): """ Normalize the values of a sparse vector to a 0-1 range using min-max scaling, considering a known range of sparse scores. Args: sparse_vector: A dict representing a sparse vector with 'indices' and 'values' min_score: The minimum score in the range of sparse scores (default is 0) max_score: The maximum score in the range of sparse scores (default is 6000) Returns: A dict representing the sparse vector with normalized 'values'. """ # normalized_values = [(value - min_score) / (max_score - min_score) for value in sparse_vector['values']] self.tensor = torch.tensor(sparse_vector['values']) self.normalized_tensor = F.normalize(self.tensor, p=2.0, dim=0, eps=1e-12) values = self.normalized_tensor.tolist() # Update the sparse_vector with normalized values self.normalized_sparse_vector = { 'indices': sparse_vector['indices'], 'values': values } return self.normalized_sparse_vector