sitingGZ/

german-bert-clinical-ner

import sys
sys.path.append('modules')

import torch
from transformers import AutoConfig, AutoTokenizer, AutoModelForMaskedLM, EncoderDecoderConfig
from BERT2span_semantic_disam import BERT2span
from helpers import load_config, set_seed
from inference import final_label_results_rescaled

base_name =  "bert-base-german-cased"
configs = load_config('configs/step3_gpu_span_semantic_group.yaml')
tokenizer = AutoTokenizer.from_pretrained(base_name)
bertMLM = AutoModelForMaskedLM.from_pretrained(base_name)
bert_sner = BERT2span(configs, bertMLM, tokenizer)

checkpoint_path = "checkpoints/german_bert_ex4cds_500_semantic_term.ckpt"
state_dict = torch.load(checkpoint_path, map_location=torch.device('cpu'))
bert_sner.load_state_dict(state_dict)
bert_sner.eval()

suggested_terms = {'Condition': 'Zeichen oder Symptom',
               'DiagLab': 'Diagnostisch und Laborverfahren',
                'LabValues': 'Klinisches Attribut',
                 'HealthState': 'Gesunder Zustand',
                 'Measure': 'Quantitatives Konzept',
                 'Medication': 'Pharmakologische Substanz',
                 'Process': 'Physiologische Funktion',
                 'TimeInfo': 'Zeitliches Konzept'}

words = "Aktuell Infekt mit Nachweis von E Coli und Pseudomonas im TBS- CRP 99mg/dl".split()
words_list = [words]
heatmaps, ner_results = final_label_results_rescaled(words_list, tokenizer, bert_sner, suggested_terms, threshold=0.5)