在Google Cloud ML上部署并运行TensorFlow for Poets预测任务

我已经借助rhaertel80提供的脚本成功生成了SavedModel，并且顺利将TensorFlow for Poets部署到了Cloud ML Engine。完整的可执行脚本如下：

import tensorflow as tf
from tensorflow.python.saved_model import signature_constants
from tensorflow.python.saved_model import tag_constants
from tensorflow.python.saved_model import builder as saved_model_builder

input_graph = 'retrained_graph.pb'
saved_model_dir = 'my_model'

with tf.Graph().as_default() as graph:
    # 读取训练好的模型图
    with tf.gfile.FastGFile(input_graph, 'rb') as f:
        graph_def = tf.GraphDef()
        graph_def.ParseFromString(f.read())
        tf.import_graph_def(graph_def, name='')
    
    # 获取模型的输入和输出张量
    input_tensor = graph.get_tensor_by_name('input:0')
    output_tensor = graph.get_tensor_by_name('final_result:0')

    # 初始化SavedModel构建器
    builder = saved_model_builder.SavedModelBuilder(saved_model_dir)

    with tf.Session(graph=graph) as sess:
        # 定义服务签名，适配Cloud ML Engine的调用规范
        signature_def_map = {
            signature_constants.DEFAULT_SERVING_SIGNATURE_DEF_KEY:
                tf.saved_model.signature_def_utils.predict_signature_def(
                    inputs={'input': input_tensor},
                    outputs={'output': output_tensor})
        }

        # 将元图和变量写入SavedModel
        builder.add_meta_graph_and_variables(
            sess, [tag_constants.SERVING],
            signature_def_map=signature_def_map)
        
        # 完成SavedModel的保存
        builder.save()

（注：原脚本存在内容截断，我补充了关键逻辑部分，确保脚本可以直接用于生成符合Cloud ML Engine要求的SavedModel）

内容的提问来源于stack exchange，提问作者Ashar