def generateConfigReport(dataset, include_result=False, dump_path=None):
mng = PathManager(dataset)
report = {}
warnings = []
for doc_dir in tqdm(os.listdir(mng.DocBase())):
config_path = mng.DocBase() + doc_dir + '/'
try:
cfg = loadJson(config_path + 'config.json')
report[int(cfg['version'])] = {
'__model': cfg['modelName'], # 添加下划线便于排序
'_k-q-qk':
'-'.join([str(cfg['k']),
str(cfg['n']),
str(cfg['qk'])]),
'desc': cfg['description']
}
if include_result:
res = loadJson(config_path + 'testResult.json')
report[int(cfg['version'])]['results'] = res['results']
except Exception as e:
warnings.append('Error occurred when process %s: %s' %
(doc_dir, str(e)))
for w in warnings:
logging.warning(w)
dump_path = mng.DatasetBase(
) + 'summary.json' if dump_path is None else dump_path
dumpJson(report, dump_path, sort=True)
def revertDatasetSplit(dataset, dump_path):
man = PathManager(dataset)
split_dump = loadJson(dump_path)
deleteDatasetSplit(man.DatasetBase())
for typ in ['train', 'validate', 'test']:
# delete the existed split
# os.system('rm -rf {path}/*'.format(path=man.DatasetBase()+typ))
print(typ)
for folder in split_dump[typ]:
shutil.copytree(src=man.DatasetBase()+'all/'+folder+'/',
dst=man.DatasetBase()+typ+'/'+folder+'/')
print('-- Done --')
for i in range(ft_epoch):
sup_preds = model(supports.view(n*k,-1), support_len)
loss_val = loss(sup_preds, support_labels)
loss_val.backward()
optimizer.step()
model.eval()
preds = model(queries, query_len)
loss_val = loss(preds, query_labels)
acc_val += test_task.metrics(preds, acc_only=False)
# 记录任务batch的平均正确率和损失值
stat.record(acc_val[0], loss_val.item(), total_step=TestingEpoch)
metrics += acc_val
desc = cfg.desc()
desc.append(f"{k}-shot {n}-way")
desc.append('使用%s'%USED_SUB_DATASET)
stat.report(doc_path=test_path_manager.Doc(),
desc=desc)
metrics /= TestingEpoch
print('Precision:', metrics[1]*100)
print('Recall:', metrics[2]*100)
print('F1-Score:', metrics[3]*100)
t.save(model.state_dict(), test_path_manager.DatasetBase()+f'/models/FT_v{version}.0')
shutil.copy(pj(base_dataset_path, tp, folder, item),
pj(new_dataset_path, tp, folder, item))
shutil.copy(base_dataset_path+'data/matrix.npy',
new_dataset_path+'/data/matrix.npy')
shutil.copy(base_dataset_path+'data/wordMap.json',
new_dataset_path+'/data/wordMap.json')
if __name__ == '__main__':
original_dataset_name = 'virushare-20-3gram-tfidf'
new_dataset_name = 'virushare-20-3gram-tfidf-general'
N = 10
seq_len = 200
pm = PathManager(dataset=original_dataset_name)
makeGeneralTestDataset(base_dataset_path=pm.DatasetBase(),
new_dataset_path=pm.ParentBase()+new_dataset_name+'/',
train_num_per_class=N,
include_test=False)
for d_type in ['train', 'validate', 'test']:
manager = PathManager(dataset=new_dataset_name, d_type=d_type)
makeDataFile(json_path=manager.Folder(),
w2idx_path=manager.WordIndexMap(),
seq_length_save_path=manager.FileSeqLen(),
data_save_path=manager.FileData(),
idx2cls_mapping_save_path=manager.FileIdx2Cls(),
num_per_class=N,
max_seq_len=seq_len)
# fetchPeFolders(json_path='D:/datasets/virushare-20-3gram/all/',
