## reading in feature label is in @GCN folder. too lazy to port this function out. LabelDescLoader = GCN_data_loader.LabelProcessor() if args.tree: # @label_in_ontology to get GO in the whole ontology, will be needed if we use tree method LabelSamples = LabelDescLoader.get_examples(args.data_dir, label_array=label_in_ontology) LabelSamples = GCN_data_loader.convert_labels_to_features(LabelSamples, MAX_SEQ_LEN_LABEL_DEF, Vocab, all_name_array=label_in_ontology, tokenize_style='space') else: ## only get vectors for labels we want. LabelSamples = LabelDescLoader.get_examples(args.data_dir, label_array=label_to_test) LabelSamples = GCN_data_loader.convert_labels_to_features(LabelSamples, MAX_SEQ_LEN_LABEL_DEF, Vocab, all_name_array=label_to_test, tokenize_style='space') GO_loader_for_biLSTM, GO_name_for_biLSTM = GCN_data_loader.make_label_loader (LabelSamples,args.batch_size_bert,fp16=False) ## if we fix encoder, then we don't have to worry about batch size, should be able to handle 32 or even 64 ## **** load protein data if args.ontology is None: add_name = "" else: add_name = '-' + args.ontology train_loader = protSeqLoader.ProtLoader (args.data_dir, 'train'+add_name+'.tsv', all_name_array, MAX_SEQ_LEN, 'random', args, args.do_kmer, label_to_test) dev_loader = protSeqLoader.ProtLoader (args.data_dir, 'dev'+add_name+'.tsv', all_name_array, MAX_SEQ_LEN, 'sequential', args, args.do_kmer, label_to_test)
processor = data_loader.LabelProcessor()
label_desc_examples = processor.get_examples(
args.main_dir, all_name_array) ## must get all labels
label_desc_features = data_loader.convert_labels_to_features(
label_desc_examples,
max_seq_length=MAX_SEQ_LEN,
tokenizer=Vocab,
all_name_array=all_name_array,
tokenize_style="space")
if args.batch_size_label_desc == 0:
args.batch_size_label_desc = args.num_label
label_desc_dataloader, label_desc_name = data_loader.make_label_loader(
label_desc_features,
batch_size=args.batch_size_label_desc,
fp16=args.fp16) # len(label_desc_examples)
print('num of label desc to be transformed by gcn {}'.format(
len(label_desc_examples)))
else:
label_desc_dataloader = None
## **** for the rest, we don't use the "GO:"
# all_name_array = [ re.sub(r"GO:","",g) for g in all_name_array ]
## read go terms entailment pairs to train
processor = data_loader.QnliProcessor()
label_list = processor.get_labels() ## no/yes entailment style