Fix TypeError: Expected instance of keras.layers.Layer, got None

I was building a custom model using tf.keras.Sequential(). The error appeared when I tried to add a layer, which turned out to be None. Debugging the issue led me to common mistakes, which I will explain here.

While working on a deep learning model with TensorFlow and Keras, I encountered an issue that blocked my progress. The error message was:

TypeError: Expected instance of keras.layers.Layer, got None

Why This Error Happens

This error occurs when a function or variable that should return a valid Keras layer returns None. A few common causes include:

• Misspelled or incorrect function calls – If a function does not return a valid Keras layer, it may return None instead.

• Improper layer imports – If keras.layers is not imported properly, references to its layers may be undefined.

• Variable name conflicts – If a variable name overwrites a function name, calling that function may return None.

• Misconfigured model architecture – If layers are not defined correctly in a model, Keras cannot recognize them.

• Issues in subclassed models – If you create a custom model class, an improper call() method may lead to this error.

How to Fix TypeError: Expected instance of keras.layers.Layer, got None

Check for Incorrect Function Calls

If a function should return a Keras layer but returns None, ensure you are calling it correctly. For example,

def get_custom_layer():
    return tf.keras.layers.Dense(64, activation='relu')

layer = get_custom_layer()
model = tf.keras.Sequential([layer])

If get_custom_layer() does not return a valid layer, check its logic.

Verify Proper Imports

Ensure you are importing layers correctly,

from tensorflow.keras import layers

If you mistakenly write,

import keras.layers

It may cause issues, depending on the TensorFlow version.

Avoid Name Conflicts

If a variable has the same name as a function, calling it will not work as expected. For example,

Dense = None  # Overwriting the Dense function
layer = Dense(64)  # This will raise an error

Ensure your variable names do not overwrite important functions.

Check Model Layer Definitions

When defining a model, verify that each layer is assigned correctly:

model = tf.keras.Sequential([
    tf.keras.layers.Dense(64, activation='relu'),
    None,  # This will cause the error
    tf.keras.layers.Dense(10, activation='softmax')
])

Avoid inserting None values into layer lists.

Fix Issues in Custom Models

If you define a custom model, ensure that call() returns the expected output:

class MyModel(tf.keras.Model):
    def __init__(self):
        super(MyModel, self).__init__()
        self.dense = tf.keras.layers.Dense(64, activation='relu')

    def call(self, inputs):
        return self.dense(inputs)

If call() does not return an output, the model will not function properly.

Summary

The error TypeError: Expected instance of keras.layers.Layer, got None usually occurs due to incorrect function calls, import errors, variable name conflicts, improper model definitions, or issues in custom models. By checking each of these areas step by step, I resolved the issue and ensured that my model worked correctly.

If you encounter this error, review your code for missing layer definitions, incorrect imports, and function return values. By following these solutions, you can fix the issue and continue building your model without interruptions.

Explore more articles on deep learning troubleshooting and TensorFlow solutions.