python - 值错误: Duplicate node name in graph: 'write_summary' while using custom callback with keras and tensorboard

Question

我在创建自定义回调时收到图中的重复节点名称。这是完整的代码。

import os
import datetime
import numpy as np
import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import layers


class MyCallback(tf.keras.callbacks.Callback):

    def __init__(self):
        log_dir = datetime.datetime.now().strftime("%Y%m%d-%H%M%S")
        self._logdir = os.path.join('logs', log_dir)

    def on_train_begin(self, logs=None):
        self.summary_writer = tf.summary.create_file_writer(self._logdir)

    def on_epoch_end(self, epoch, logs=None):
        t = tf.reduce_sum(self.model._targets[0])
        with self.summary_writer.as_default():
            tf.summary.scalar("test", t, step=epoch)

    def on_train_end(self, logs=None):
        self.summary_writer.close()


def get_model():
    model = tf.keras.Sequential([
    # Adds a densely-connected layer with 64 units to the model:
    layers.Dense(64, activation='relu', input_shape=(32,)),
    # Add another:
    layers.Dense(64, activation='relu'),
    # Add a softmax layer with 10 output units:
    layers.Dense(10, activation='softmax')])

    model.compile(optimizer=tf.keras.optimizers.Adam(0.01),
        loss='categorical_crossentropy',
        metrics=['accuracy'])

    return model

if __name__ == '__main__':

    model = get_model()
    data = np.random.random((1000, 32))
    labels = np.random.random((1000, 10))

    model.fit(data, labels, epochs=10, batch_size=32, callbacks=[MyCallback()])

我收到的错误是。

$ python3 test.py 
2020-01-20 20:09:00.694065: I tensorflow/core/platform/cpu_feature_guard.cc:142] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 FMA
2020-01-20 20:09:00.718085: I tensorflow/core/platform/profile_utils/cpu_utils.cc:94] CPU Frequency: 2592000000 Hz
2020-01-20 20:09:00.718952: I tensorflow/compiler/xla/service/service.cc:168] XLA service 0x4bead60 executing computations on platform Host. Devices:
2020-01-20 20:09:00.718974: I tensorflow/compiler/xla/service/service.cc:175]   StreamExecutor device (0): Host, Default Version
Train on 1000 samples
Epoch 1/10
1000/1000 [==============================] - 1s 569us/sample - loss: 45.9223 - accuracy: 0.0980
Epoch 2/10
  32/1000 [..............................] - ETA: 0s - loss: 85.0311 - accuracy: 0.0938Traceback (most recent call last):
  File "/usr/local/lib/python3.6/dist-packages/tensorflow_core/python/framework/ops.py", line 1610, in _create_c_op
    c_op = c_api.TF_FinishOperation(op_desc)
tensorflow.python.framework.errors_impl.InvalidArgumentError: Duplicate node name in graph: 'write_summary'

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "test.py", line 48, in <module>
    model.fit(data, labels, epochs=10, batch_size=32, callbacks=[MyCallback()])
  File "/usr/local/lib/python3.6/dist-packages/tensorflow_core/python/keras/engine/training.py", line 728, in fit
    use_multiprocessing=use_multiprocessing)
  File "/usr/local/lib/python3.6/dist-packages/tensorflow_core/python/keras/engine/training_v2.py", line 372, in fit
    prefix='val_')
  File "/usr/lib/python3.6/contextlib.py", line 88, in __exit__
    next(self.gen)
  File "/usr/local/lib/python3.6/dist-packages/tensorflow_core/python/keras/engine/training_v2.py", line 685, in on_epoch
    self.callbacks.on_epoch_end(epoch, epoch_logs)
  File "/usr/local/lib/python3.6/dist-packages/tensorflow_core/python/keras/callbacks.py", line 298, in on_epoch_end
    callback.on_epoch_end(epoch, logs)
  File "test.py", line 21, in on_epoch_end
    tf.summary.scalar("test", t, step=epoch)
  File "/usr/local/lib/python3.6/dist-packages/tensorboard/plugins/scalar/summary_v2.py", line 65, in scalar
    metadata=summary_metadata)
  File "/usr/local/lib/python3.6/dist-packages/tensorflow_core/python/ops/summary_ops_v2.py", line 646, in write
    _should_record_summaries_v2(), record, _nothing, name="summary_cond")
  File "/usr/local/lib/python3.6/dist-packages/tensorflow_core/python/framework/smart_cond.py", line 54, in smart_cond
    return true_fn()
  File "/usr/local/lib/python3.6/dist-packages/tensorflow_core/python/ops/summary_ops_v2.py", line 640, in record
    name=scope)
  File "/usr/local/lib/python3.6/dist-packages/tensorflow_core/python/ops/gen_summary_ops.py", line 868, in write_summary
    summary_metadata=summary_metadata, name=name)
  File "/usr/local/lib/python3.6/dist-packages/tensorflow_core/python/framework/op_def_library.py", line 793, in _apply_op_helper
    op_def=op_def)
  File "/usr/local/lib/python3.6/dist-packages/tensorflow_core/python/framework/func_graph.py", line 548, in create_op
    compute_device)
  File "/usr/local/lib/python3.6/dist-packages/tensorflow_core/python/framework/ops.py", line 3429, in _create_op_internal
    op_def=op_def)
  File "/usr/local/lib/python3.6/dist-packages/tensorflow_core/python/framework/ops.py", line 1773, in __init__
    control_input_ops)
  File "/usr/local/lib/python3.6/dist-packages/tensorflow_core/python/framework/ops.py", line 1613, in _create_c_op
    raise ValueError(str(e))
ValueError: Duplicate node name in graph: 'write_summary'

它是python3.6和tensorflow-2.0.0

在 TensorflowSupport 的评论后编辑以下实现工作。

class MyCallback(tf.keras.callbacks.TensorBoard):

    def __init__(self):
        log_dir = datetime.datetime.now().strftime("%Y%m%d-%H%M%S")
        logdir = os.path.join('logs', log_dir)
        super(MyCallback, self).__init__(logdir)


    def on_epoch_end(self, epoch, logs=None):
        super(MyCallback, self).on_epoch_end(epoch, logs)
        with self._get_writer("train").as_default():
            t = tf.reduce_sum(self.model._targets[0])
            tf.summary.scalar("test", 90, step=epoch)

以下也有效

class MyCallback(tf.keras.callbacks.TensorBoard):

    def __init__(self):
        log_dir = datetime.datetime.now().strftime("%Y%m%d-%H%M%S")
        logdir = os.path.join('logs', log_dir)
        super(MyCallback, self).__init__(logdir)


    def on_epoch_end(self, epoch, logs=None):
        super(MyCallback, self).on_epoch_end(epoch, logs)
        with self._get_writer("train").as_default():
            tf.summary.scalar("test", 90, step=epoch)

但这不起作用并引发相同的错误。

class MyCallback(tf.keras.callbacks.TensorBoard):

    def __init__(self):
        log_dir = datetime.datetime.now().strftime("%Y%m%d-%H%M%S")
        logdir = os.path.join('logs', log_dir)
        super(MyCallback, self).__init__(logdir)


    def on_epoch_end(self, epoch, logs=None):
        super(MyCallback, self).on_epoch_end(epoch, logs)
        with self._get_writer("train").as_default():
            t = tf.reduce_sum(self.model._targets[0])
            tf.summary.scalar("test", t, step=epoch)

Answer 1

我的理解是，如果您将 Tensor 传递给 tf.summary.scalar，则会产生错误，但如果您将 Scalar 传递给它，它就会执行而不会出现任何错误。

因此，我们可以将 Tensor 转换为 Scalar，然后将该 Scalar 传递给 tf.summary.scalar，从而成功执行。

下面提到的代码应该可以工作:

class MyCallback(tf.keras.callbacks.TensorBoard):

    def __init__(self):
        log_dir = datetime.datetime.now().strftime("%Y%m%d-%H%M%S")
        logdir = os.path.join('logs', log_dir)
        super(MyCallback, self).__init__(logdir)


    def on_epoch_end(self, epoch, logs=None):
        super(MyCallback, self).on_epoch_end(epoch, logs)
        with self._get_writer("train").as_default():
            t = tf.reduce_sum(self.model._targets[0])
            t_scalar = t.numpy() #This line will convert Tensor to a Scalar
            tf.summary.scalar("test", t_scalar, step=epoch)