InvalidArgumentError：无法将 dtype 资源的张量转换为 NumPy 数组

问题描述

我想使用 tf-hub 构建文本分类模型并导出为 tflite 模型但是
在转换包括 tf hub 的 tensorflow 模型时，我遇到了错误。请帮我解决。

import tensorflow as tf
import tensorflow_hub as hub 
model = tf.keras.Sequential()
model.add(tf.keras.layers.InputLayer(dtype=tf.string,input_shape=()))

model.add(hub.KerasLayer("https://tfhub.dev/google/tf2-preview/nnlm-en-dim50/1"))
converter=tf.lite.TFLiteConverter.from_keras_model(model)
tflite_model = converter.convert()

我尝试了 tf-lite python 和命令行 api。但是我遇到了 InvalidArgumentError。

---

InvalidArgumentError                      Traceback (most recent call last)
<ipython-input-15-5a8dbd778645> in <module>()
      5 model.add(hub.KerasLayer("https://tfhub.dev/google/tf2-preview/nnlm-en-dim50/1"))
      6 converter = tf.lite.TFLiteConverter.from_keras_model(model)
----> 7 tflite_model = converter.convert()

6 frames
/usr/local/lib/python3.6/dist-packages/tensorflow/lite/python/lite.py in convert(self)
    850     frozen_func,graph_def = (
    851         _convert_to_constants.convert_variables_to_constants_v2_as_graph(
--> 852             self._funcs[0],lower_control_flow=False))
    853 
    854     input_tensors = [

/usr/local/lib/python3.6/dist-packages/tensorflow/python/framework/convert_to_constants.py in convert_variables_to_constants_v2_as_graph(func,lower_control_flow,aggressive_inlining)
   1103       func=func,1104       lower_control_flow=lower_control_flow,-> 1105       aggressive_inlining=aggressive_inlining)
   1106 
   1107   output_graph_def,converted_input_indices = _replace_variables_by_constants(

/usr/local/lib/python3.6/dist-packages/tensorflow/python/framework/convert_to_constants.py in __init__(self,func,aggressive_inlining,variable_names_allowlist,variable_names_denylist)
    804         variable_names_allowlist=variable_names_allowlist,805         variable_names_denylist=variable_names_denylist)
--> 806     self._build_tensor_data()
    807 
    808   def _build_tensor_data(self):

/usr/local/lib/python3.6/dist-packages/tensorflow/python/framework/convert_to_constants.py in _build_tensor_data(self)
    823         data = map_index_to_variable[idx].numpy()
    824       else:
--> 825         data = val_tensor.numpy()
    826       self._tensor_data[tensor_name] = _TensorData(
    827           numpy=data,/usr/local/lib/python3.6/dist-packages/tensorflow/python/framework/ops.py in numpy(self)
   1069     """
   1070     # Todo(slebedev): Consider avoiding a copy for non-cpu or remote tensors.
-> 1071     maybe_arr = self._numpy()  # pylint: disable=protected-access
   1072     return maybe_arr.copy() if isinstance(maybe_arr,np.ndarray) else maybe_arr
   1073 

/usr/local/lib/python3.6/dist-packages/tensorflow/python/framework/ops.py in _numpy(self)
   1037       return self._numpy_internal()
   1038     except core._NotOkStatusException as e:  # pylint: disable=protected-access
-> 1039       six.raise_from(core._status_to_exception(e.code,e.message),None)  # pylint: disable=protected-access
   1040 
   1041   @property

/usr/local/lib/python3.6/dist-packages/six.py in raise_from(value,from_value)

InvalidArgumentError: Cannot convert a Tensor of dtype resource to a NumPy array.

解决方法

上次我检查时，TFLite 不支持查找表，这是 TF Hub 模型中资源张量的主要来源（除了变量，但这些肯定有效）。