(
use_llm_func,community_report,maybe_edge,relations,generate_relations,\
cluster_cluster_relation_prompt,layer,tokenizer,max_depth
)
| 446 | 'index':indices |
| 447 | } |
| 448 | def process_relation( |
| 449 | use_llm_func,community_report,maybe_edge,relations,generate_relations,\ |
| 450 | cluster_cluster_relation_prompt,layer,tokenizer,max_depth |
| 451 | |
| 452 | ): |
| 453 | cluster1_nodes=community_report[maybe_edge[0]]['children'] |
| 454 | cluster2_nodes=community_report[maybe_edge[1]]['children'] |
| 455 | |
| 456 | threshold=min(len(cluster1_nodes)*0.2,len(cluster2_nodes)*0.2) |
| 457 | # threshold=1 |
| 458 | exists_relation={**get_direct_relations(cluster1_nodes,cluster2_nodes,relations), |
| 459 | **get_direct_relations(cluster1_nodes,cluster2_nodes,generate_relations)} |
| 460 | if exists_relation=={}: |
| 461 | return None |
| 462 | |
| 463 | cluster1_description=community_report[maybe_edge[0]]['findings'] |
| 464 | cluster2_description=community_report[maybe_edge[1]]['findings'] |
| 465 | relation_infromation=[ |
| 466 | f"relationship<|>{v['src_tgt']}<|>{v['tgt_src']}<|>{v['description']} " |
| 467 | for k,v in exists_relation.items() |
| 468 | ] |
| 469 | temp_relations={} |
| 470 | tokens=len(tokenizer.encode("\n".join(relation_infromation))) |
| 471 | gene_tokens=(layer+1)*40 |
| 472 | allowed_tokens=(max_depth-layer)*40*2 |
| 473 | # allowed_tokens=100000 |
| 474 | # allowed_tokens=1 |
| 475 | if tokens>allowed_tokens: |
| 476 | print(f"{tokens}大于{allowed_tokens}，进行llm生成\n{maybe_edge[0]}和{maybe_edge[1]} in processing") |
| 477 | exact_prompt=cluster_cluster_relation_prompt.format(entity_a=maybe_edge[0],entity_b=maybe_edge[1],\ |
| 478 | entity_a_description=cluster1_description,entity_b_description=cluster2_description,\ |
| 479 | relation_information="\n".join(relation_infromation),tokens=gene_tokens) |
| 480 | |
| 481 | response = use_llm_func(exact_prompt) |
| 482 | temp_relations[maybe_edge]={ |
| 483 | 'src_tgt':maybe_edge[0], |
| 484 | 'tgt_src':maybe_edge[1], |
| 485 | 'description':response, |
| 486 | 'weight':1, |
| 487 | 'level':layer+1 |
| 488 | } |
| 489 | else: |
| 490 | print(f"{tokens}小于{allowed_tokens}，不进行llm生成") |
| 491 | temp_relations[maybe_edge]={ |
| 492 | 'src_tgt':maybe_edge[0], |
| 493 | 'tgt_src':maybe_edge[1], |
| 494 | 'description':"\n".join(relation_infromation), |
| 495 | 'weight':1, |
| 496 | 'level':layer+1 |
| 497 | } |
| 498 | return temp_relations |
| 499 | |
| 500 | class Hierarchical_Clustering(ClusteringAlgorithm): |
| 501 | def perform_clustering( |
nothing calls this directly
no test coverage detected