DistributedDatasetInterface is not an attribute of input_lib - engine\data_adapter.py

[samuelsennev]

Issue type

Bug

Have you reproduced the bug with TensorFlow Nightly?

No

Source

source

TensorFlow version

2.13

Custom code

No

OS platform and distribution

No response

Mobile device

No response

Python version

3.10.9

Bazel version

No response

GCC/compiler version

No response

CUDA/cuDNN version

No response

GPU model and memory

No response

Current behavior?

I'm new to machine learning and I was setting up this network with just one layer, with only one neuron inside it, just to see how it works. Setting up went fine. The error occured when I asked it to predict.

I'm not sure what exactly happend, but apparently, a non existent attribute (input_lib.DistributedDatasetInterface) was passed to isinstance() in the file: ...\site-packages\tensorflow\python\keras\engine\data_adapter.py (line 1699).

I've replaced by the wrong one by the one suggested and worked as expected.

Standalone code to reproduce the issue

import numpy as np
import tensorflow as tf
from tensorflow.python.keras.layers import Dense, Input
from tensorflow.python.keras import Sequential
from tensorflow.python.keras.activations import sigmoid

import logging
logging.getLogger("tensorflow").setLevel(logging.ERROR)
tf.autograph.set_verbosity(0)

X_train = np.array([0., 1, 2, 3, 4, 5], dtype=np.float32).reshape(-1,1)
Y_train = np.array([0,  0, 0, 1, 1, 1], dtype=np.float32).reshape(-1,1)

model = Sequential([
    Input(shape=(1,)),
    Dense(1, activation=sigmoid, name = "L1")
])
model.summary()

# Setting the weights and bias of the neuron
logistic_layer = model.get_layer('L1')
set_w = np.array([[2]])
set_b = np.array([-4.5])
logistic_layer.set_weights([set_w, set_b])

# Performance test
a1 = model.predict(X_train[0].reshape(1,1)) # this line caused the error
print(a1)
alog = sigmoid(np.dot(set_w,X_train[0].reshape(1,1)) + set_b)
print(alog)

Relevant log output

...\site-packages\tensorflow\python\keras\engine\data_adapter.py", line 1699, in _is_distributed_dataset
    return isinstance(ds, input_lib.DistributedDatasetInterface)
AttributeError: module 'tensorflow.python.distribute.input_lib' has no attribute 'DistributedDatasetInterface'. Did you mean: 'DistributedDatasetSpec'?

[hrishikeshkh]

this is the same error I had: AttributeError: module 'tensorflow.python.distribute.input_lib' has no attribute 'DistributedDatasetInterface'. This error comes up even when I try model.fit and any other operation with the model beyond its compile line.

[samuelsennev]

this is the same error I had: AttributeError: module 'tensorflow.python.distribute.input_lib' has no attribute 'DistributedDatasetInterface'. This error comes up even when I try model.fit and any other operation with the model beyond its compile line.

I've solved it by replacing DistributedDatasetInterface by DistributedDatasetSpec right on:

...\site-packages\tensorflow\python\keras\engine\data_adapter.py", line 1699, in _is_distributed_dataset

This is what I did:

def _is_distributed_dataset(ds):
#   return isinstance(ds, input_lib.DistributedDatasetInterface)
    return isinstance(ds, input_lib.DistributedDatasetSpec)

Definetely not a good practice, I know, but...

[vatsarishabh22]

return isinstance(ds, input_lib.DistributedDatasetSpec)

do you have any update if input_lib.DistributedDatasetSpec carries same functionality as input_lib.DistributedDatasetInterface ?

[tilakrayal]

@samuelsennev,
tensorflow/python/keras code is a legacy copy of Keras since the TensorFlow v2.7 release, and will be deleted in the v2.12 release.

Please remove any import of tensorflow.python.keras and use the public API with from tensorflow import keras or import tensorflow as tf; tf.keras.

I tried to execute the mentioned code on tensorflow v2.13 by using below imports and it was executed without any issues.

from tensorflow.keras.layers import Dense, Input
from tensorflow.keras import Sequential
from tensorflow.keras.activations import sigmoid

Kindly find the gist of it here.
https://github.com/tensorflow/tensorflow/releases

Thank you!

[samuelsennev]

@tilakrayal
Got it!

Worked with the public API and from tensorflow import keras

Thanks ;)

[google-ml-butler]

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

[motto0808]

@samuelsennev, tensorflow/python/keras code is a legacy copy of Keras since the TensorFlow v2.7 release, and will be deleted in the v2.12 release.

Please remove any import of tensorflow.python.keras and use the public API with from tensorflow import keras or import tensorflow as tf; tf.keras.

I tried to execute the mentioned code on tensorflow v2.13 by using below imports and it was executed without any issues.

from tensorflow.keras.layers import Dense, Input
from tensorflow.keras import Sequential
from tensorflow.keras.activations import sigmoid

Kindly find the gist of it here. https://github.com/tensorflow/tensorflow/releases

Thank you!

it does works!
but after i change the imports code, the IDE(vs code) cannot recognize any type under keras， is there any way to resolve it?

[RubTalha]

https://stackoverflow.com/a/77733620/13086128