def run_fasttext_infer():
"""run infer with FastText"""
dataset = load_infer_dataset(batch_size=config.batch_size, datafile=args.data_path)
fasttext_model = FastText(config.vocab_size, config.embedding_dims, config.num_class)
parameter_dict = load_checkpoint(args.model_ckpt)
load_param_into_net(fasttext_model, parameter_dict=parameter_dict)
ft_infer = FastTextInferCell(fasttext_model)
model = Model(ft_infer)
predictions = []
target_sens = []
for batch in dataset.create_dict_iterator(output_numpy=True, num_epochs=1):
target_sens.append(batch['label_idx'])
src_tokens = Tensor(batch['src_tokens'], mstype.int32)
src_tokens_length = Tensor(batch['src_tokens_length'], mstype.int32)
predicted_idx = model.predict(src_tokens, src_tokens_length)
predictions.append(predicted_idx.asnumpy())
from sklearn.metrics import accuracy_score, classification_report
target_sens = np.array(target_sens).flatten()
predictions = np.array(predictions).flatten()
acc = accuracy_score(target_sens, predictions)
result_report = classification_report(target_sens, predictions, target_names=target_label1)
print("********Accuracy: ", acc)
print(result_report)
def __init__(self, vocab_size, embedding_dims, num_class):
super(FastTextNetWithLoss, self).__init__()
self.fasttext = FastText(vocab_size, embedding_dims, num_class)
self.loss_func = nn.SoftmaxCrossEntropyWithLogits(sparse=True,
reduction='mean')
self.squeeze = P.Squeeze(axis=1)
self.print = P.Print()
def run_fasttext_export():
"""export function"""
fasttext_model = FastText(config.vocab_size, config.embedding_dims, config.num_class)
parameter_dict = load_checkpoint(args.ckpt_file)
load_param_into_net(fasttext_model, parameter_dict)
ft_infer = FastTextInferExportCell(fasttext_model)
batch_size = config.batch_size
if args.device_target == 'GPU':
batch_size = config.distribute_batch_size
if args.data_name == "ag":
src_tokens_shape = [batch_size, 467]
src_tokens_length_shape = [batch_size, 1]
elif args.data_name == 'dbpedia':
src_tokens_shape = [batch_size, 1120]
src_tokens_length_shape = [batch_size, 1]
elif args.data_name == 'yelp_p':
src_tokens_shape = [batch_size, 2955]
src_tokens_length_shape = [batch_size, 1]
file_name = args.file_name + '_' + args.data_name
src_tokens = Tensor(np.ones((src_tokens_shape)).astype(np.int32))
src_tokens_length = Tensor(np.ones((src_tokens_length_shape)).astype(np.int32))
export(ft_infer, src_tokens, src_tokens_length, file_name=file_name, file_format=args.file_format)
def create_network(name, *args, **kwargs):
if name == "fasttext":
return FastText(*args, **kwargs)
raise NotImplementedError(f"{name} is not implemented in the repo")
def fasttext(*args, **kwargs):
return FastText(*args, **kwargs)