ValueError: Exception encountered when calling layer "sequential" (type Sequential)

[Citizen-Dan]

When fitting the mode for first time in the 01_neural_network_regression_in_tensorflow.ipynb notebook, I get the following error shown below. I have tried to reshape the training data but to no avail.

"ValueError: Exception encountered when calling layer "sequential_6" (type Sequential).

Input 0 of layer "dense_8" is incompatible with the layer: expected min_ndim=2, found ndim=1. Full shape received: (None,)"

[mrdbourke]

Hey @Citizen-Dan,

This looks like it's an update from TensorFlow 2.7.0 that will break some code.

See the TensorFlow 2.7.0 update here: https://github.com/tensorflow/tensorflow/releases/tag/v2.7.0

But the main thing is incorrect input shapes for models, in essence fit() no longer turns data from (batch_size, ) to (batch_size, 1).

So we have to do it manually by adding an extra dimension.

Error

The error you might see if passing a scalar to any TensorFlow training function:

ValueError: Exception encountered when calling layer "sequential" (type Sequential).
    
    Input 0 of layer "dense" is incompatible with the layer: expected min_ndim=2, found ndim=1. Full shape received: (None,)
    
    Call arguments received:
      • inputs=tf.Tensor(shape=(None,), dtype=float32)
      • training=True
      • mask=None

TensorFlow version

You will see this breakage if you're coming from TensorFlow <2.7.0 (all versions prior to 2.7.0).

You can check your TensorFlow version here:

import tensorflow as tf
print(tf.__version__)

>>> 2.7.0

Google Colab uses TensorFlow 2.7.0+ as of November 2021.

Code Before

import tensorflow as tf

# Create features (using tensors)
X = tf.constant([-7.0, -4.0, -1.0, 2.0, 5.0, 8.0, 11.0, 14.0])

# Create labels (using tensors)
y = tf.constant([3.0, 6.0, 9.0, 12.0, 15.0, 18.0, 21.0, 24.0])

# Set random seed
tf.random.set_seed(42)

# Create a model using the Sequential API
model = tf.keras.Sequential([
  tf.keras.layers.Dense(1)
])

# Compile the model
model.compile(loss=tf.keras.losses.mae, # mae is short for mean absolute error
              optimizer=tf.keras.optimizers.SGD(), # SGD is short for stochastic gradient descent
              metrics=["mae"])

# Fit the model
model.fit(X, y, epochs=5) # this will break with TensorFlow 2.7.0+

Code After (how to fix it)

Notice the change to the input of model.fit(). The shape of X gets expanded in the last dimension using tf.expand_dims(X, axis=-1).

import tensorflow as tf

# Create features (using tensors)
X = tf.constant([-7.0, -4.0, -1.0, 2.0, 5.0, 8.0, 11.0, 14.0])

# Create labels (using tensors)
y = tf.constant([3.0, 6.0, 9.0, 12.0, 15.0, 18.0, 21.0, 24.0])

# Set random seed
tf.random.set_seed(42)

# Create a model using the Sequential API
model = tf.keras.Sequential([
  tf.keras.layers.Dense(1)
])

# Compile the model
model.compile(loss=tf.keras.losses.mae, # mae is short for mean absolute error
              optimizer=tf.keras.optimizers.SGD(), # SGD is short for stochastic gradient descent
              metrics=["mae"])

# Fit the model
model.fit(tf.expand_dims(X, axis=-1), y, epochs=5) # <- updated line

That's the only major change that you should have to do when fitting a model on data that's a scalar (e.g. no second dimension, like X).

[mrdbourke]

@Yash-Wasalwar-07 the labels vector can stay how it is due to it already being a vector. If it had a shape issue after fitting the model, we could reshape it as well.

[happymondaynkanta]

I ran into a similar problem too. Reshaping the array into a 2-dimensional array solved the problem.
Use the code.reshape(-1,1)
Example:
X = tf.constant([-7.0, -4.0, -1.0, 2.0, 5.0, 8.0, 11.0, 14.0]).reshape(-1,1)
y = tf.constant([3.0, 6.0, 9.0, 12.0, 15.0, 18.0, 21.0, 24.0]).reshape(-1,1)

[Skorkmaz88]

Reshape causes error in version : 2.8.

X = tf.constant([12.2, 12.1]).reshape(-1,1)

# Create labels (using tensors)
y = tf.constant([12.5, 12.3]).reshape(-1,1)

'EagerTensor' object has no attribute 'reshape'.
If you are looking for numpy-related methods, please run the following:
from tensorflow.python.ops.numpy_ops import np_config
np_config.enable_numpy_behavior()

[DebjitHore]

As an alternative to this, I think when initialising the model if we explicitly define the input shape as having one single dimension that has the same effect?
I am pretty new to this and I am not sure if what I am suggesting is right, so I kindly take it with a pinch of salt and suggest if I am on track or wrong.

model= tf.keras.Sequential([tf.keras.Input(shape=(1,)), tf.keras.layers.Dense(1) ])

[C3ald]

ValueError: Attempt to convert a value (<TakeDataset element_spec=(TensorSpec(shape=(1,), dtype=tf.int64, name=None), TensorSpec(shape=(), dtype=tf.string, name=None))>) with an unsupported type (<class 'tensorflow.python.data.ops.dataset_ops.TakeDataset'>) to a Tensor.

[dsaha21]

import tensorflow as tf

Set random seed

tf.random.set_seed(42)

Create` features (using tensors)

X = tf.constant([-7.0, -4.0, -1.0, 2.0, 5.0, 8.0, 11.0, 14.0])

Create labels (using tensors)

y = tf.constant([3.0, 6.0, 9.0, 12.0, 15.0, 18.0, 21.0, 24.0])

Create a model using the Sequential API

m = tf.keras.Sequential([
tf.keras.layers.Dense(1),
tf.keras.Input(shape=(1,) # updated line
])

Compile the model

m.compile(loss=tf.keras.losses.mae,
optimizer=tf.keras.optimizers.SGD(),
metrics=["mae"])

Fit the model

m.fit(tf.expand_dims(X, axis = -1), y, epochs = 5) # updated line

[SKsatyam18]

why to use that tf.expand

[Dikshith4128]

it was done for us before the update now we have to do it manually