Python Can'；t将带有特征列的keras.models.Model转换为估计模型。TypeError:缺少参数：'；功能列'&引用；

使用TF1.14.0 我正在尝试将带有特征列的keras模型转换为估计模型。 但它不起作用，我如何解决它？ 我的代码是：

import tensorflow as tf
from tensorflow import keras
from tensorflow.keras.layers import Input, Dense, Lambda
from tensorflow.keras.models import Model
from tensorflow.keras.layers import dot
from tensorflow import feature_column

numerical_columns = ['age','fare']
feature_columns = []

def make_feature_column(numerical_columns):
    feature_columns = []
    feature_layer_inputs = {}
    # numeric cols
    for header in numerical_columns:
        header_str = str(header)
        feature_columns.append(feature_column.numeric_column(header_str))
        feature_layer_inputs[header_str] = tf.keras.Input(shape=(1,), name=header_str)
    return feature_columns,feature_layer_inputs

feature_columns,feature_layer_inputs = make_feature_column(numerical_columns)

feature_layer = tf.keras.layers.DenseFeatures(feature_columns,trainable=False)(feature_layer_inputs)
x = Dense(units=2048, activation='relu')(feature_layer)
x = Dense(units=1024, activation='relu')(x)
x = Dense(units=512, activation='relu')(x)
x = Dense(units=200, activation='relu',name='user_vec')(x)
item_input = Input((200,), name='item_vec')
x = Lambda(lambda t: t / tf.linalg.norm(t, ord=1))(x)

x = dot([x, item_input], axes=-1)
main_output = Dense(1, activation='sigmoid', name='main_output')(x)

feature_layers = feature_layer_inputs.values()
inputs = [v for v in feature_layers]
inputs.append(item_input)

model = Model(inputs=inputs, outputs=main_output)
model.compile(optimizer='rmsprop', loss={'main_output': 'binary_crossentropy'}, loss_weights={'main_output': 1.})

estimator = tf.keras.estimator.model_to_estimator(model)

但是它返回了一个类型错误，我不能理解这个错误。 错误是：

Traceback (most recent call last):
  File "/Users/lideyang/PycharmProjects/NewUserSort/test.py", line 42, in <module>
    estimator = tf.keras.estimator.model_to_estimator(model)
  File "/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow/python/keras/estimator/__init__.py", line 73, in model_to_estimator
    config=config)
  File "/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow_estimator/python/estimator/keras.py", line 450, in model_to_estimator
    config)
  File "/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow_estimator/python/estimator/keras.py", line 318, in _save_first_checkpoint
    custom_objects)
  File "/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow_estimator/python/estimator/keras.py", line 201, in _clone_and_build_model
    optimizer_iterations=global_step)
  File "/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow/python/keras/models.py", line 538, in clone_and_build_model
    clone = clone_model(model, input_tensors=input_tensors)
  File "/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow/python/keras/models.py", line 326, in clone_model
    model, input_tensors=input_tensors, layer_fn=clone_function)
  File "/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow/python/keras/models.py", line 154, in _clone_functional_model
    new_layer = layer_fn(layer)
  File "/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow/python/keras/models.py", line 54, in _clone_layer
    return layer.__class__.from_config(layer.get_config())
  File "/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow/python/keras/engine/base_layer.py", line 446, in from_config
    return cls(**config)
TypeError: __init__() missing 1 required positional argument: 'feature_columns'

回溯（最近一次呼叫最后一次）：
文件“/Users/lideyang/PycharmProjects/NewUserSort/test.py”，第42行，在
估计器=tf.keras.estimator.model_to_估计器（model）
文件“/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site packages/tensorflow/Python/keras/estimator/__init____;.py”，第73行，在model_to_uestimator中
config=config）
文件“/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site packages/tensorflow_-estimator/Python/estimator/keras.py”，第450行，模型_-to_-estimator
配置）
文件“/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site packages/tensorflow\u estimator/Python/estimator/keras.py”，第318行，保存第一个检查点
自定义对象）
文件“/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site packages/tensorflow\u estimator/Python/estimator/keras.py”，第201行，在克隆和构建模型中
优化器_迭代次数=全局_步）
文件“/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site packages/tensorflow/Python/keras/models.py”，第538行，在clone_和_build_模型中
克隆=克隆\模型（模型，输入\张量=输入\张量）
文件“/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site packages/tensorflow/Python/keras/models.py”，第326行，在clone_模型中
模型，输入\张量=输入\张量，层\ fn=克隆\函数）
文件“/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site packages/tensorflow/Python/keras/models.py”，第154行，在克隆功能模型中
新层=层层（层）
文件“/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site packages/tensorflow/Python/keras/models.py”，第54行，在克隆层
返回层。来自层配置的层类（layer.get\u config（））
文件“/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site packages/tensorflow/Python/keras/engine/base_layer.py”，第446行，from_config
返回cls（**配置）
TypeError:\uuuu init\uuuuu（）缺少1个必需的位置参数：“feature\u columns”

---

如果我不使用功能列，而是使用Inputlayer，比如“user\u input=input（（300），，name='user\u input'），它就可以工作了。如何解决此问题。

一旦将tensorflow升级到

1.15

或

2.x

，您的问题就可以得到解决

%tensorflow_version 2.x
import tensorflow as tf
print(tf.__version__)
from tensorflow import keras
from tensorflow.keras.layers import Input, Dense, Lambda
from tensorflow.keras.models import Model
from tensorflow.keras.layers import dot
from tensorflow import feature_column

numerical_columns = ['age','fare']
feature_columns = []

def make_feature_column(numerical_columns):
    feature_columns = []
    feature_layer_inputs = {}
    # numeric cols
    for header in numerical_columns:
        header_str = str(header)
        feature_columns.append(feature_column.numeric_column(header_str))
        feature_layer_inputs[header_str] = tf.keras.Input(shape=(1,), name=header_str)
    return feature_columns,feature_layer_inputs

feature_columns,feature_layer_inputs = make_feature_column(numerical_columns)

feature_layer = tf.keras.layers.DenseFeatures(feature_columns,trainable=False)(feature_layer_inputs)
x = Dense(units=2048, activation='relu')(feature_layer)
x = Dense(units=1024, activation='relu')(x)
x = Dense(units=512, activation='relu')(x)
x = Dense(units=200, activation='relu',name='user_vec')(x)
item_input = Input((200,), name='item_vec')
x = Lambda(lambda t: t / tf.linalg.norm(t, ord=1))(x)

x = dot([x, item_input], axes=-1)
main_output = Dense(1, activation='sigmoid', name='main_output')(x)

feature_layers = feature_layer_inputs.values()
inputs = [v for v in feature_layers]
inputs.append(item_input)

model = Model(inputs=inputs, outputs=main_output)
model.compile(optimizer='rmsprop', loss={'main_output': 'binary_crossentropy'}, loss_weights={'main_output': 1.})

estimator = tf.keras.estimator.model_to_estimator(model)

输出：

2.4.0
INFO:tensorflow:Using default config.
WARNING:tensorflow:Using temporary folder as model directory: /tmp/tmpwa8ialaq
INFO:tensorflow:Using the Keras model provided.
/usr/local/lib/python3.6/dist-packages/tensorflow/python/keras/backend.py:434: UserWarning: `tf.keras.backend.set_learning_phase` is deprecated and will be removed after 2020-10-11. To update it, simply pass a True/False value to the `training` argument of the `__call__` method of your layer or model.
  warnings.warn('`tf.keras.backend.set_learning_phase` is deprecated and '
INFO:tensorflow:Using config: {'_model_dir': '/tmp/tmpwa8ialaq', '_tf_random_seed': None, '_save_summary_steps': 100, '_save_checkpoints_steps': None, '_save_checkpoints_secs': 600, '_session_config': allow_soft_placement: true
graph_options {
  rewrite_options {
    meta_optimizer_iterations: ONE
  }
}
, '_keep_checkpoint_max': 5, '_keep_checkpoint_every_n_hours': 10000, '_log_step_count_steps': 100, '_train_distribute': None, '_device_fn': None, '_protocol': None, '_eval_distribute': None, '_experimental_distribute': None, '_experimental_max_worker_delay_secs': None, '_session_creation_timeout_secs': 7200, '_checkpoint_save_graph_def': True, '_service': None, '_cluster_spec': ClusterSpec({}), '_task_type': 'worker', '_task_id': 0, '_global_id_in_cluster': 0, '_master': '', '_evaluation_master': '', '_is_chief': True, '_num_ps_replicas': 0, '_num_worker_replicas': 1}