1

我需要优化旧版 tensorflow 模型，我正在关注这篇博客文章：https : //medium.com/google-cloud/optimizing-tensorflow-models-for-serving-959080e9ddbf（我首先尝试了 tensorflow lite转换器没有成功）。

博客文章中的步骤有效，除了最后一个，将优化的 Graph 转换回 SavedModel 格式。经过一些实验，我最终得到了这个功能：

def convert_graph_def_to_saved_model(export_dir, graph_filepath):
    if tf.gfile.Exists(export_dir):
        tf.gfile.DeleteRecursively(export_dir)
    graph_def = get_graph_def_from_file(graph_filepath)
    with tf.Session(graph=tf.Graph()) as session:
        tf.import_graph_def(graph_def, name='')
        inputs = {name_map['{}:0'.format(node.name)]: 
            session.graph.get_tensor_by_name(
                '{}:0'.format(node.name))
            for node in graph_def.node if node.op=='Placeholder'}
        outputs = {'class_ids': session.graph.get_tensor_by_name(
            'head/predictions/class_ids:0')}
        signature = tf.saved_model.signature_def_utils.predict_signature_def(
    inputs=inputs, outputs=outputs)
        session.run([tf.global_variables_initializer(), tf.tables_initializer()])
        b = tf.saved_model.builder.SavedModelBuilder(export_dir)
        b.add_meta_graph_and_variables(session, [tag_constants.SERVING], signature_def_map={'predict':signature}, 
                                       legacy_init_op = tf.group(tf.tables_initializer(), name='legacy_init_op'))
        b.save()
    print('Optimized graph converted to SavedModel!')

这将正确创建 SavedModel 文件。

然后进行预测，遗留代码使用

self.predictor = tensorflow.contrib.predictor.from_saved_model(model_path. signature_def_key="predict" 
predictions = self.predictor(feed_dict)

其中 feed_dict 包含数据。

在这一点上，但是我收到以下错误：

in predict(self, df, feed_dict, single_feed_dict)
    218             raise Exception('You must provide a df or a feed_dict to predict')
    219 
--> 220         predictions = self.predictor(feed_dict)
    221 
    222         return predictions

venv_py3.7/lib/python3.7/site-packages/tensorflow_core/contrib/predictor/predictor.py in __call__(self, input_dict)
     75       if value is not None:
     76         feed_dict[self.feed_tensors[key]] = value
---> 77     return self._session.run(fetches=self.fetch_tensors, feed_dict=feed_dict)

venv_py3.7/lib/python3.7/site-packages/tensorflow_core/python/client/session.py in run(self, fetches, feed_dict, options, run_metadata)
    954     try:
    955       result = self._run(None, fetches, feed_dict, options_ptr,
--> 956                          run_metadata_ptr)
    957       if run_metadata:
    958         proto_data = tf_session.TF_GetBuffer(run_metadata_ptr)

venv_py3.7/lib/python3.7/site-packages/tensorflow_core/python/client/session.py in _run(self, handle, fetches, feed_dict, options, run_metadata)
   1178     if final_fetches or final_targets or (handle and feed_dict_tensor):
   1179       results = self._do_run(handle, final_targets, final_fetches,
-> 1180                              feed_dict_tensor, options, run_metadata)
   1181     else:
   1182       results = []

venv_py3.7/lib/python3.7/site-packages/tensorflow_core/python/client/session.py in _do_run(self, handle, target_list, fetch_list, feed_dict, options, run_metadata)
   1357     if handle is None:
   1358       return self._do_call(_run_fn, feeds, fetches, targets, options,
-> 1359                            run_metadata)
   1360     else:
   1361       return self._do_call(_prun_fn, handle, feeds, fetches)

venv_py3.7/lib/python3.7/site-packages/tensorflow_core/python/client/session.py in _do_call(self, fn, *args)
   1382                     '\nsession_config.graph_options.rewrite_options.'
   1383                     'disable_meta_optimizer = True')
-> 1384       raise type(e)(node_def, op, message)
   1385 
   1386   def _extend_graph(self):

FailedPreconditionError: Table not initialized.

我使用 python 3.7 和 tensorflow 1.15。

有人可以帮忙吗？

4

0 回答 0