When migrating legacy TensorFlow 1.x code to TensorFlow 2.x, several common attribute errors may occur due to API changes. Below are typical issues and their solutions:

1. module 'tensorflow' has no attribute 'placeholder'

tf.placeholder was removed in TensorFlow 2.x. To retain 1.x behavior:

import tensorflow.compat.v1 as tf
tf.disable_v2_behavior()

2. module 'tensorflow' has no attribute 'random_normal'

The function was renamed and moved under the tf.random namespace:

# Old (TF 1.x)
tf.random_normal(...)

# New (TF 2.x)
tf.random.normal(...)

3. name 'X' is not defined

This usually happens when a variable like X is defined inside a function but accessed globally. Declare it as global where first assigned:

global X
X = ...

4. module 'tensorflow._api.v2.train' has no attribute 'GradientDescentOptimizer'

Use the v1 compatibility module:

# Instead of
optimizer = tf.train.GradientDescentOptimizer(learning_rate)

# Use
optimizer = tf.compat.v1.train.GradientDescentOptimizer(learning_rate)

5. module 'tensorflow' has no attribute 'Session'

tf.Session is unavailable in eager execution mode (default in TF 2.x). Use the compatibility layer:

sess = tf.compat.v1.Session()

If you encounter "The Session graph is empty...", disable eager execution:

tf.compat.v1.disable_eager_execution()

6. module 'tensorboard.summary._tf.summary' has no attribute 'FileWriter'

Replace the deprecated FileWriter with the new API:

# Old
writer = tf.summary.FileWriter(logdir)

# New
writer = tf.summary.create_file_writer(logdir)

7. module 'tensorflow' has no attribute 'get_default_graph'

This often occurs when mixing standalone Keras with TensorFlow. Always import Keras from TensorFlow:

# Avoid
import keras
from keras import layers

# Prefer
from tensorflow import keras
from tensorflow.keras import layers