AutoGraph could not transform <bound method TopLevelFeature.decode_example of FeaturesDict

[olk]

System information

• OS Platform and Distribution: Arch Linux, 5.4.2-arch1-1-ARCH
• TensorFlow installed from: binary
• TensorFlow version: 2.1.0rc0-1
• Keras version: 2.2.4-tf
• Python version: 3.8
• GPU model and memory: 2x GTX 1080 Ti 11GB"`

Describe the current behavior
executing Tensorflow's MNIST handwriting example produces warning:

WARNING:tensorflow:AutoGraph could not transform <bound method TopLevelFeature.decode_example of FeaturesDict({
'image': Image(shape=(28, 28, 1), dtype=tf.uint8),
'label': ClassLabel(shape=(), dtype=tf.int64, num_classes=10),
})> and will run it as-is.
Please report this to the TensorFlow team. When filing the bug, set the verbosity to 10 (on Linux, export AUTOGRAPH_VERBOSITY=10) and attach the full output.
Cause: 'arguments' object has no attribute 'defaults'

Code to reproduce the issue
import tensorflow as tf
import tensorflow_datasets as tfds

from tensorflow.keras.optimizers import Adam

def build_model():
filters = 48
units = 24
kernel_size = 7
learning_rate = 1e-4
model = tf.keras.Sequential([
tf.keras.layers.Conv2D(filters=filters, kernel_size=(kernel_size, kernel_size), activation='relu', input_shape=(28, 28, 1)),
tf.keras.layers.MaxPooling2D(),
tf.keras.layers.Flatten(),
tf.keras.layers.Dense(units, activation='relu'),
tf.keras.layers.Dense(10, activation='softmax')
])
model.compile(loss='sparse_categorical_crossentropy', optimizer=Adam(learning_rate), metrics=['accuracy'])
return model

datasets, info = tfds.load(name='mnist', with_info=True, as_supervised=True)
mnist_train, mnist_test = datasets['train'], datasets['test']

num_train_examples = info.splits['train'].num_examples
num_test_examples = info.splits['test'].num_examples

BUFFER_SIZE = 10000
BATCH_SIZE = 32

def scale(image, label):
image = tf.cast(image, tf.float32)
image /= 255
return image, label

train_dataset = mnist_train.map(scale).shuffle(BUFFER_SIZE).repeat().batch(BATCH_SIZE).prefetch(buffer_size=tf.data.experimental.AUTOTUNE)
eval_dataset = mnist_test.map(scale).repeat().batch(BATCH_SIZE).prefetch(buffer_size=tf.data.experimental.AUTOTUNE)

model = build_model()

epochs=2
model.fit(
train_dataset,
validation_data=eval_dataset,
steps_per_epoch=num_train_examples/epochs,
validation_steps=num_test_examples/epochs,
epochs=epochs)

[oanush]

@olk ,
Hi,i tried running the code in TF-2.1rc-1 and I didn't face any error. Kindly find the gist of colab for your reference.Thanks!

[olk]

print('TensorFlow version: %s' % tf.__version__)
print('Keras version: %s' % tf.keras.__version__)

prints:

TensorFlow version: 2.1.0-rc0
Keras version: 2.2.4-tf

Does RC1 fix the issue?

[olk]

Tensorflow at Arch Linux upgraded to new version

TensorFlow version: 2.1.0-rc1
Keras version: 2.2.4-tf

but same warnings are displayed:

WARNING:tensorflow:AutoGraph could not transform <bound method TopLevelFeature.decode_example of FeaturesDict({
'image': Image(shape=(28, 28, 1), dtype=tf.uint8),
'label': ClassLabel(shape=(), dtype=tf.int64, num_classes=10),
})> and will run it as-is.
Please report this to the TensorFlow team. When filing the bug, set the verbosity to 10 (on Linux, export AUTOGRAPH_VERBOSITY=10) and attach the full output.
Cause: 'arguments' object has no attribute 'defaults'
WARNING:tensorflow:AutoGraph could not transform <bound method TopLevelFeature.decode_example of FeaturesDict({
'image': Image(shape=(28, 28, 1), dtype=tf.uint8),
'label': ClassLabel(shape=(), dtype=tf.int64, num_classes=10),
})> and will run it as-is.
Please report this to the TensorFlow team. When filing the bug, set the verbosity to 10 (on Linux, export AUTOGRAPH_VERBOSITY=10) and attach the full output.
Cause: 'arguments' object has no attribute 'defaults'
WARNING:tensorflow:AutoGraph could not transform <bound method TopLevelFeature.decode_example of FeaturesDict({
'image': Image(shape=(28, 28, 1), dtype=tf.uint8),
'label': ClassLabel(shape=(), dtype=tf.int64, num_classes=10),
})> and will run it as-is.
Please report this to the TensorFlow team. When filing the bug, set the verbosity to 10 (on Linux, export AUTOGRAPH_VERBOSITY=10) and attach the full output.
Cause: 'arguments' object has no attribute 'defaults'
WARNING:tensorflow:AutoGraph could not transform <bound method TopLevelFeature.decode_example of FeaturesDict({
'image': Image(shape=(28, 28, 1), dtype=tf.uint8),
'label': ClassLabel(shape=(), dtype=tf.int64, num_classes=10),
})> and will run it as-is.
Please report this to the TensorFlow team. When filing the bug, set the verbosity to 10 (on Linux, export AUTOGRAPH_VERBOSITY=10) and attach the full output.
Cause: 'arguments' object has no attribute 'defaults'
WARNING:tensorflow:AutoGraph could not transform <function scale at 0x7f1877956940> and will run it as-is.
Please report this to the TensorFlow team. When filing the bug, set the verbosity to 10 (on Linux, export AUTOGRAPH_VERBOSITY=10) and attach the full output.
Cause: invalid value for "node": expected "ast.AST", got "<class 'NoneType'>"; to visit lists of nodes, use "visit_block" instead
WARNING:tensorflow:AutoGraph could not transform <function scale at 0x7f1877956940> and will run it as-is.
Please report this to the TensorFlow team. When filing the bug, set the verbosity to 10 (on Linux, export AUTOGRAPH_VERBOSITY=10) and attach the full output.
Cause: invalid value for "node": expected "ast.AST", got "<class 'NoneType'>"; to visit lists of nodes, use "visit_block" instead
Train for 30000.0 steps, validate for 5000.0 steps

[olk]

I guess this issue is related to using Tensorflow with Python-3.8.

[olk]

The error vanishes if I downgraded my system to:

• Python 3.7.4
• Tensorflow-2.1.0-rc1

After system upgrade (Python 3.8.0) I get the error messages again:

WARNING:root:Limited tf.compat.v2.summary API due to missing TensorBoard installation.
WARNING:root:Limited tf.compat.v2.summary API due to missing TensorBoard installation.
WARNING:root:Limited tf.compat.v2.summary API due to missing TensorBoard installation.
WARNING:tensorflow:AutoGraph could not transform <bound method TopLevelFeature.decode_example of FeaturesDict({
'image': Image(shape=(28, 28, 1), dtype=tf.uint8),
'label': ClassLabel(shape=(), dtype=tf.int64, num_classes=10),
})> and will run it as-is.
Please report this to the TensorFlow team. When filing the bug, set the verbosity to 10 (on Linux, export AUTOGRAPH_VERBOSITY=10) and attach the full output.
Cause: 'arguments' object has no attribute 'defaults'
WARNING:tensorflow:AutoGraph could not transform <bound method TopLevelFeature.decode_example of FeaturesDict({
'image': Image(shape=(28, 28, 1), dtype=tf.uint8),
'label': ClassLabel(shape=(), dtype=tf.int64, num_classes=10),
})> and will run it as-is.
Please report this to the TensorFlow team. When filing the bug, set the verbosity to 10 (on Linux, export AUTOGRAPH_VERBOSITY=10) and attach the full output.
Cause: 'arguments' object has no attribute 'defaults'
WARNING:tensorflow:AutoGraph could not transform <bound method TopLevelFeature.decode_example of FeaturesDict({
'image': Image(shape=(28, 28, 1), dtype=tf.uint8),
'label': ClassLabel(shape=(), dtype=tf.int64, num_classes=10),
})> and will run it as-is.
Please report this to the TensorFlow team. When filing the bug, set the verbosity to 10 (on Linux, export AUTOGRAPH_VERBOSITY=10) and attach the full output.
Cause: 'arguments' object has no attribute 'defaults'
WARNING:tensorflow:AutoGraph could not transform <bound method TopLevelFeature.decode_example of FeaturesDict({
'image': Image(shape=(28, 28, 1), dtype=tf.uint8),
'label': ClassLabel(shape=(), dtype=tf.int64, num_classes=10),
})> and will run it as-is.
Please report this to the TensorFlow team. When filing the bug, set the verbosity to 10 (on Linux, export AUTOGRAPH_VERBOSITY=10) and attach the full output.
Cause: 'arguments' object has no attribute 'defaults'
WARNING:tensorflow:AutoGraph could not transform <function scale at 0x7f35a7d63040> and will run it as-is.
Please report this to the TensorFlow team. When filing the bug, set the verbosity to 10 (on Linux, export AUTOGRAPH_VERBOSITY=10) and attach the full output.
Cause: invalid value for "node": expected "ast.AST", got "<class 'NoneType'>"; to visit lists of nodes, use "visit_block" instead
WARNING:tensorflow:AutoGraph could not transform <function scale at 0x7f35a7d63040> and will run it as-is.
Please report this to the TensorFlow team. When filing the bug, set the verbosity to 10 (on Linux, export AUTOGRAPH_VERBOSITY=10) and attach the full output.
Cause: invalid value for "node": expected "ast.AST", got "<class 'NoneType'>"; to visit lists of nodes, use "visit_block" instead
Train for 30000.0 steps, validate for 5000.0 steps
Epoch 1/2
2019-12-17 21:01:35.085243: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcublas.so.10
2019-12-17 21:01:36.341770: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcudnn.so.7
30000/30000 [==============================] - 114s 4ms/step - loss: 0.0856 - accuracy: 0.9758 - val_loss: 0.0379 - val_accuracy: 0.9889
Epoch 2/2
29996/30000 [============================>.] - ETA: 0s - loss: 0.0161 - accuracy: 0.99552019-12-17 21:05:09.198963: W tensorflow/core/kernels/data/generator_dataset_op.cc:103] Error occurred when finalizing GeneratorDataset iterator: Cancelled: Operation was cancelled
30000/30000 [==============================] - 112s 4ms/step - loss: 0.0161 - accuracy: 0.9955 - val_loss: 0.0350 - val_accuracy: 0.9898
2019-12-17 21:05:20.499356: W tensorflow/core/kernels/data/generator_dataset_op.cc:103] Error occurred when finalizing GeneratorDataset iterator: Cancelled: Operation was cancelled
2019-12-17 21:05:20.516223: W tensorflow/core/kernels/data/generator_dataset_op.cc:103] Error occurred when finalizing GeneratorDataset iterator: Cancelled: Operation was cancelled
elapsed: 226.111

Seams that this is an issue of Python-3.8.0 and Tensorflow-2.1.0-rc1.

[gowthamkpr]

@olk Try building tensorflow nightly version with python 3.8 and let me know if the issue still persists. Thanks!

[tensorflowbutler]

It has been 14 days with no activity and the awaiting response label was assigned. Is this still an issue?

[olk]

I'm too busy to build tensorflow from scratch ...
Anyway, I've moved to MXNet - feel free to close this issue.

[gowthamkpr]

@olk I am sorry for misleading you in the above comment.

Tensorflow doesn't support python 3.8 yet. May be from Tensorflow 2.2, we might support python 3.8 and this is the cause of the error. Python 3.8 support for tensorflow has already been tracked here.

I am gonna close this issue. Thanks!

[tensorflow-bot]

Are you satisfied with the resolution of your issue?
Yes
No