Question

我按照one of the TF beginner tutorial中的步骤创建了一个简单的分类模型。它们是：

from __future__ import absolute_import, division, print_function, unicode_literals
import numpy as np
import pandas as pd
import tensorflow as tf
from tensorflow import feature_column
from tensorflow.keras import layers
from sklearn.model_selection import train_test_split

URL = 'https://storage.googleapis.com/applied-dl/heart.csv'
dataframe = pd.read_csv(URL)
dataframe.head()

train, test = train_test_split(dataframe, test_size=0.2)
train, val = train_test_split(train, test_size=0.2)

def df_to_dataset(dataframe, shuffle=True, batch_size=32):
  dataframe = dataframe.copy()
  labels = dataframe.pop('target')
  ds = tf.data.Dataset.from_tensor_slices((dict(dataframe), labels))
  if shuffle:
    ds = ds.shuffle(buffer_size=len(dataframe))
  ds = ds.batch(batch_size)
  return ds

batch_size = 5 # A small batch sized is used for demonstration purposes
train_ds = df_to_dataset(train, batch_size=batch_size)
val_ds = df_to_dataset(val, shuffle=False, batch_size=batch_size)
test_ds = df_to_dataset(test, shuffle=False, batch_size=batch_size)

feature_columns = []
for header in ['age', 'trestbps', 'chol', 'thalach', 'oldpeak', 'slope', 'ca']:
  feature_columns.append(feature_column.numeric_column(header))
thal_embedding = feature_column.embedding_column(thal, dimension=8)
feature_columns.append(thal_embedding)

feature_layer = tf.keras.layers.DenseFeatures(feature_columns)

batch_size = 32
train_ds = df_to_dataset(train, batch_size=batch_size)
val_ds = df_to_dataset(val, shuffle=False, batch_size=batch_size)
test_ds = df_to_dataset(test, shuffle=False, batch_size=batch_size)


model = tf.keras.Sequential([
  feature_layer,
  layers.Dense(128, activation='relu'),
  layers.Dense(128, activation='relu'),
  layers.Dense(1, activation='sigmoid')
])

model.compile(optimizer='adam',
              loss='binary_crossentropy',
              metrics=['accuracy'],
              run_eagerly=True)

model.fit(train_ds,
          validation_data=val_ds,
          epochs=5)

然后我将模型保存为：

model.save("model/", save_format='tf')

然后，我尝试使用此TF tutorial为该模型提供服务。我执行以下操作：

docker pull tensorflow/serving
docker run -p 8501:8501 --mount type=bind,source=/path/to/model/,target=/models/model -e MODEL_NAME=mo

我尝试以这种方式调用模型：

curl -d '{"inputs": {"age": [0], "trestbps": [0], "chol": [0], "thalach": [0], "oldpeak": [0], "slope": [1], "ca": [0], "exang": [0], "restecg": [0], "fbs": [0], "cp": [0], "sex": [0], "thal": ["normal"], "target": [0] }}' -X POST http://localhost:8501/v1/models/model:predict

我收到以下错误：

{“错误”：“索引= 1不在[0，1）\ n \ t [[{{node StatefulPartitionedCall_51 / StatefulPartitionedCall / sequential / dense_features / thal_embedding / thal_embedding_weights / GatherV2}}]]}}}

它似乎与“ thal”功能的嵌入层有关。但是我不知道“索引= 1不在[0，1）中”是什么意思，为什么会这样。

发生错误时，这是TF docker服务器记录的内容：

2019-09-23 12：50：43.921721：W external / org_tensorflow / tensorflow / core / framework / op_kernel.cc：1502] OP_REQUIRES在lookup_table_op.cc:952失败：失败的前提条件：表已初始化。

知道错误来自哪里以及如何解决？

Python版本：3.6

tensorflow版本：2.0.0-rc0

最新TensorFlow /服务（截至2019年9月20日）

模型签名：

signature_def['__saved_model_init_op']:
  The given SavedModel SignatureDef contains the following input(s):
  The given SavedModel SignatureDef contains the following output(s):
    outputs['__saved_model_init_op'] tensor_info:
        dtype: DT_INVALID
        shape: unknown_rank
        name: NoOp
  Method name is: 

signature_def['serving_default']:
  The given SavedModel SignatureDef contains the following input(s):
    inputs['age'] tensor_info:
        dtype: DT_INT32
        shape: (-1, 1)
        name: serving_default_age:0
    inputs['ca'] tensor_info:
        dtype: DT_INT32
        shape: (-1, 1)
        name: serving_default_ca:0
    inputs['chol'] tensor_info:
        dtype: DT_INT32
        shape: (-1, 1)
        name: serving_default_chol:0
    inputs['cp'] tensor_info:
        dtype: DT_INT32
        shape: (-1, 1)
        name: serving_default_cp:0
    inputs['exang'] tensor_info:
        dtype: DT_INT32
        shape: (-1, 1)
        name: serving_default_exang:0
    inputs['fbs'] tensor_info:
        dtype: DT_INT32
        shape: (-1, 1)
        name: serving_default_fbs:0
    inputs['oldpeak'] tensor_info:
        dtype: DT_FLOAT
        shape: (-1, 1)
        name: serving_default_oldpeak:0
    inputs['restecg'] tensor_info:
        dtype: DT_INT32
        shape: (-1, 1)
        name: serving_default_restecg:0
    inputs['sex'] tensor_info:
        dtype: DT_INT32
        shape: (-1, 1)
        name: serving_default_sex:0
    inputs['slope'] tensor_info:
        dtype: DT_INT32
        shape: (-1, 1)
        name: serving_default_slope:0
    inputs['thal'] tensor_info:
        dtype: DT_STRING
        shape: (-1, 1)
        name: serving_default_thal:0
    inputs['thalach'] tensor_info:
        dtype: DT_INT32
        shape: (-1, 1)
        name: serving_default_thalach:0
    inputs['trestbps'] tensor_info:
        dtype: DT_INT32
        shape: (-1, 1)
        name: serving_default_trestbps:0
  The given SavedModel SignatureDef contains the following output(s):
    outputs['output_1'] tensor_info:
        dtype: DT_FLOAT
        shape: (-1, 1)
        name: StatefulPartitionedCall:0
  Method name is: tensorflow/serving/predict

Answer 1

我也试图提供一个由嵌入层，lstm层等组成的模型，但是我收到了其他一些错误。我什至在TF上提出了issue。

无论如何，我在代码中看到的问题是用于Docker的已保存模型的类型。如果您阅读here，它会显示以下几点-

A SavedModel to serve

不是keras model.save，而是另一个TF API，here是描述从keras训练模型中创建 SavedModel 的方法。试试看，让我们知道结果。

Answer 2

问题似乎与您发送的格式有关。您可以张贴模型的签名吗？由于声誉低下，因此无法将其发布为评论。

Answer 3

我遇到了同样的问题。更改为以下格式。

curl -d '{"inputs": {"age": [[0]], "trestbps": [[0]], "chol": [[0]], "thalach": [[0]], "oldpeak": [[0]], "slope": [[1]], "ca": [[0]], "exang": [[0]], "restecg": [[0]], "fbs": [[0]], "cp": [[0]], "sex": [[0]], "thal": [["normal"]], "target": [[0]] }}' -X POST http://localhost:8501/v1/models/model:predict

注意：全部改为[["normal"]]或[[0]]

在TF 2.0中使用Tensorflow / Keras模型时，嵌入层存在问题

3 个答案: