Python TF2/Keras:加载自定义模型

我正在不断地训练一个定制的keras模型。我已经像这样覆盖了这个方法

class MyModel(keras.models.Model):

    def train_step(self, data):
        # doing my magic and returning metrics

训练就像一种魅力，能给你带来高分（是啊！）。然后我通过

model.save

进入tf文件夹。我想在收集更多数据后继续培训

然后我通过

model = keras.models.load_model(saved_model)

然而，现在的模型是

tensorflow.python.keras.engine.functional.Functional

而不是

MyModel

并且缺少自定义的

训练步骤

方法。当然，当我在加载的模型上调用

fit

时，一切都会中断。如何在TF2/keras中正确加载自定义模型

//编辑：示例代码

#!/usr/bin/env python3

from tensorflow import keras
from tensorflow.keras import layers
import tensorflow as tf


class MyModel(keras.Model):
    def train_step(self, data):

        x, y_true = data

        with tf.GradientTape() as tape:
            y_pred = self(x, training=True)

            loss = tf.losses.mean_squared_error(y_true['a'], y_pred['a'])
            loss += tf.losses.mean_squared_error(y_true['b'], y_pred['b'])

            trainable_vars = self.trainable_variables
            gradients = tape.gradient(loss, trainable_vars)

            self.optimizer.apply_gradients(zip(gradients, trainable_vars))

        return {'loss': loss}


def get_model():
    input = layers.Input(shape=(20, 1))
    x = layers.Conv1D(10, 3, activation='relu')(input)
    x = layers.Conv1D(20, 5, activation='relu')(x)
    x = layers.Conv1D(30, 7, activation='relu')(x)
    x = layers.GlobalMaxPool1D()(x)
    x_a = layers.Dense(2)(x)
    x_b = layers.Dense(3)(x)

    return MyModel(input, {'a': x_a, 'b': x_b}, name='MyModel')


def data_gen(n=640):
    for _ in range(n):
        features = tf.random.uniform(shape=(20, 1))
        labels = {
            'a': tf.random.uniform(shape=(2,), minval=0, maxval=2, dtype=tf.float32),
            'b': tf.random.uniform(shape=(3,), minval=-1, maxval=0, dtype=tf.float32),
        }
        yield features, labels


ds = tf.data.Dataset.from_generator(data_gen, output_types=(tf.float32, {'a': tf.float32, 'b': tf.float32}), output_shapes=((20, 1), {'a': (2,), 'b': (3,)}))
ds = ds.batch(64)

model = get_model()
model.compile('adam')
print(model)
model.summary()
model.fit(ds, epochs=5)
model.save('/tmp/mymodel')

# load
model = keras.models.load_model('/tmp/mymodel')
print(model)
model.summary()
model.fit(ds, epochs=5)

执行此操作时，我可以使用自定义train_步骤训练模型。但加载保存的模型后，列车步骤丢失，无法正确处理输出。因此，它抱怨：

Epoch 1/5
Traceback (most recent call last):
File "./example.py", line 63, in <module>
    model.fit(ds, epochs=5)
File "/home/user/venv/default/lib/python3.6/site-packages/tensorflow/python/keras/engine/training.py", line 108, in _method_wrapper
    return method(self, *args, **kwargs)
File "/home/user/venv/default/lib/python3.6/site-packages/tensorflow/python/keras/engine/training.py", line 1098, in fit
    tmp_logs = train_function(iterator)
File "/home/user/venv/default/lib/python3.6/site-packages/tensorflow/python/eager/def_function.py", line 780, in __call__
    result = self._call(*args, **kwds)
File "/home/user/venv/default/lib/python3.6/site-packages/tensorflow/python/eager/def_function.py", line 823, in _call
    self._initialize(args, kwds, add_initializers_to=initializers)
File "/home/user/venv/default/lib/python3.6/site-packages/tensorflow/python/eager/def_function.py", line 697, in _initialize
    *args, **kwds))
File "/home/user/venv/default/lib/python3.6/site-packages/tensorflow/python/eager/function.py", line 2855, in _get_concrete_function_internal_garbage_collected
    graph_function, _, _ = self._maybe_define_function(args, kwargs)
File "/home/user/venv/default/lib/python3.6/site-packages/tensorflow/python/eager/function.py", line 3213, in _maybe_define_function
    graph_function = self._create_graph_function(args, kwargs)
File "/home/user/venv/default/lib/python3.6/site-packages/tensorflow/python/eager/function.py", line 3075, in _create_graph_function
    capture_by_value=self._capture_by_value),
File "/home/user/venv/default/lib/python3.6/site-packages/tensorflow/python/framework/func_graph.py", line 986, in func_graph_from_py_func
    func_outputs = python_func(*func_args, **func_kwargs)
File "/home/user/venv/default/lib/python3.6/site-packages/tensorflow/python/eager/def_function.py", line 600, in wrapped_fn
    return weak_wrapped_fn().__wrapped__(*args, **kwds)
File "/home/user/venv/default/lib/python3.6/site-packages/tensorflow/python/framework/func_graph.py", line 973, in wrapper
    raise e.ag_error_metadata.to_exception(e)
ValueError: in user code:

    /home/user/venv/default/lib/python3.6/site-packages/tensorflow/python/keras/engine/training.py:806 train_function  *
        return step_function(self, iterator)
    /home/user/venv/default/lib/python3.6/site-packages/tensorflow/python/keras/engine/training.py:796 step_function  **
        outputs = model.distribute_strategy.run(run_step, args=(data,))
    /home/user/venv/default/lib/python3.6/site-packages/tensorflow/python/distribute/distribute_lib.py:1211 run
        return self._extended.call_for_each_replica(fn, args=args, kwargs=kwargs)
    /home/user/venv/default/lib/python3.6/site-packages/tensorflow/python/distribute/distribute_lib.py:2585 call_for_each_replica
        return self._call_for_each_replica(fn, args, kwargs)
    /home/user/venv/default/lib/python3.6/site-packages/tensorflow/python/distribute/distribute_lib.py:2945 _call_for_each_replica
        return fn(*args, **kwargs)
    /home/user/venv/default/lib/python3.6/site-packages/tensorflow/python/keras/engine/training.py:789 run_step  **
        outputs = model.train_step(data)
    /home/user/venv/default/lib/python3.6/site-packages/tensorflow/python/keras/engine/training.py:757 train_step
        self.trainable_variables)
    /home/user/venv/default/lib/python3.6/site-packages/tensorflow/python/keras/engine/training.py:2737 _minimize
        trainable_variables))
    /home/user/venv/default/lib/python3.6/site-packages/tensorflow/python/keras/optimizer_v2/optimizer_v2.py:562 _aggregate_gradients
        filtered_grads_and_vars = _filter_grads(grads_and_vars)
    /home/user/venv/default/lib/python3.6/site-packages/tensorflow/python/keras/optimizer_v2/optimizer_v2.py:1271 _filter_grads
        ([v.name for _, v in grads_and_vars],))

    ValueError: No gradients provided for any variable: ['conv1d/kernel:0', 'conv1d/bias:0', 'conv1d_1/kernel:0', 'conv1d_1/bias:0', 'conv1d_2/kernel:0', 'conv1d_2/bias:0', 'dense/kernel:0', 'dense/bias:0', 'dense_1/kernel:0', 'dense_1/bias:0'].

1/5纪元
回溯（最近一次呼叫最后一次）：
文件“/example.py”，第63行，在
模型拟合（ds，历元=5）
文件“/home/user/venv/default/lib/python3.6/site packages/tensorflow/python/keras/engine/training.py”，第108行，在方法包装中
返回方法（self、*args、**kwargs）
文件“/home/user/venv/default/lib/python3.6/site packages/tensorflow/python/keras/engine/training.py”，第1098行
tmp_logs=训练函数（迭代器）
文件“/home/user/venv/default/lib/python3.6/site packages/tensorflow/python/eager/def_function.py”，第780行，在调用中__
结果=自身调用（*args，**kwds）
文件“/home/user/venv/default/lib/python3.6/site packages/tensorflow/python/eager/def_function.py”，第823行，在调用中
self.\u初始化（参数、KWD、添加初始值设定项到=初始值设定项）
文件“/home/user/venv/default/lib/python3.6/site packages/tensorflow/python/eager/def_function.py”，第697行，在
*args，**科威特第纳尔）
文件“/home/user/venv/default/lib/python3.6/site-packages/tensorflow/python/eager/function.py”，第2855行，位于“获取”\u具体”\u函数\u内部\u垃圾收集
图函数，自我，可能定义函数（args，kwargs）
文件“/home/user/venv/default/lib/python3.6/site packages/tensorflow/python/eager/function.py”，第3213行，在函数定义中
graph\u function=self.\u create\u graph\u function（args，kwargs）
文件“/home/user/venv/default/lib/python3.6/site packages/tensorflow/python/eager/function.py”，第3075行，在“创建图形”函数中
按值捕获=自身。_按值捕获），
文件“/home/user/venv/default/lib/python3.6/site packages/tensorflow/python/framework/func_graph.py”，第986行，在func_graph_中，从_py_func
func_outputs=python_func（*func_args，**func_kwargs）
文件“/home/user/venv/default/lib/python3.6/site packages/tensorflow/python/eager/def_function.py”，第600行，包装为
返回弱_-wrapped_-fn（）
包装器中的文件“/home/user/venv/default/lib/python3.6/site packages/tensorflow/python/framework/func_graph.py”，第973行
将e.ag\u错误\u元数据引发到\u异常（e）
ValueError:在用户代码中：
/home/user/venv/default/lib/python3.6/site packages/tensorflow/python/keras/engine/training.py:806 train_函数*
返回步骤_函数（self、迭代器）
/home/user/venv/default/lib/python3.6/site packages/tensorflow/python/keras/engine/training.py:796 step_函数**
输出=模型。分配策略。运行（运行步骤，参数=（数据，）
/home/user/venv/default/lib/python3.6/site-packages/tensorflow/python/distribute/distribute_-lib.py:1211运行
返回self.\u扩展。为每个\u副本调用\u（fn，args=args，kwargs=kwargs）
/home/user/venv/default/lib/python3.6/site-packages/tensorflow/python/distribute/distribute_-lib.py:2585为每个_副本调用_
返回自我。为每个副本（fn、ARG、kwargs）调用
/home/user/venv/default/lib/python3.6/site packages/tensorflow/python/distribute/distribute\u lib.py:2945\u调用每个副本
返回fn（*args，**kwargs）
/home/user/venv/default/lib/python3.6/site-packages/tensorflow/python/keras/engine/training.py:789运行步骤**
输出=型号列车步进（数据）
/home/user/venv/default/lib/python3.6/site-packages/tensorflow/python/keras/engine/training.py:757训练步骤
可自我训练的变量）
/home/user/venv/default/lib/python3.6/site-packages/tensorflow/python/keras/engine/training.py:2737\u
可训练变量）
/home/user/venv/default/lib/python3.6/site packages/tensorflow/python/keras/optimizer\u v2/optimizer\u v2.py:562\u aggregate\u gradients
过滤的梯度和变量=\u过滤梯度（梯度和变量）
/home/user/venv/default/lib/python3.6/site packages/tensorflow/python/keras/optimizer\u v2/optimizer\u v2.py:1271\u filter\u grads
（[v.name代表u，v在grads和vars中]，）
ValueError:没有为任何变量提供渐变：['conv1d/kernel:0'，'conv1d/bias:0'，'conv1d\u 1/kernel:0'，'conv1d\u 2/kernel:0'，'conv1d\u 2/bias:0'，'densite/kernel:0'，'densite\u 1/kernel 0'，'densite\u 1/bias:0']。

请显示

打印的输出（“加载的型号：，加载的型号）

然后您说

所有东西都断了。

请显示断了的，并附加堆栈跟踪。谢谢。请分享错误日志以了解问题。并检查自定义对象保存和加载模型。ThanksI用一个例子更新了这个问题，说明了问题所在。这个例子是否更清楚地显示了错误？@papaya这个例子是否显示了明显的错误？你有什么建议吗？请显示

打印的输出（“加载的模型：，加载的模型）

，然后你说

所有东西都坏了。

请显示坏了的东西，并附加堆栈跟踪。谢谢。请分享错误日志以了解问题。并检查自定义对象保存和加载模型。ThanksI用一个例子更新了这个问题，说明了问题所在。这个例子是否更清楚地显示了错误？@papaya这个例子是否显示了明显的错误？你有什么建议吗？