Prediction API error for dense_vector_sequence input #1703
Description
Got the following error message when calling converter.
Traceback (most recent call last): File "src/PaddlePredictor.py", line 102, in process_predict result.success = self._handler.predict(args.tenantId, args.predictorId, args.input) File "src/paddle_predictor_server.py", line 173, in predict result = adapter_method(model_obj, parsed_input) File "src/adapter.py", line 191, in entity_tagging result = model.predict(modified_input) File "src/paddle_predictor_server.py", line 106, in predict result = self.model.predict(data) File "src/model.py", line 39, in predict input = self.converter(self.data_provider_module.convert_data(data)) File "/usr/local/lib/python2.7/site-packages/py_paddle/dataprovider_converter.py", line 233, in __call__ return self.convert(dat, argument) File "/usr/local/lib/python2.7/site-packages/py_paddle/dataprovider_converter.py", line 225, in convert scanner.scan(each_step) File "/usr/local/lib/python2.7/site-packages/py_paddle/dataprovider_converter.py", line 184, in scan self.__inner_scanner__.scan(each) File "/usr/local/lib/python2.7/site-packages/py_paddle/dataprovider_converter.py", line 109, in scan self.__mat__[self.__height__] = dat TypeError: 'NoneType' object does not support item assignment I checked the source code, and found that DenseScanner has a finish_pre_scan method, which assigns value to self.__mat__. But if it's a dense_vector_sequence, finish_pre_scan is not called by the SequenceScanner, which caused the None error.