Numpy interoperability in TensorFlow - Practice Problems & Coding Challenges

Challenge - 5 Problems

❓ Predict Output

intermediate

TensorFlow to NumPy conversion output

What is the output of this code snippet converting a TensorFlow tensor to a NumPy array?

TensorFlow

import tensorflow as tf
x = tf.constant([[1, 2], [3, 4]])
np_array = x.numpy()
print(np_array)

A

tf.Tensor([[1 2]
 [3 4]], shape=(2, 2), dtype=int32)

B [[1, 2], [3, 4]]

C

[[1 2]
 [3 4]]

D array([[1, 2], [3, 4]], dtype=int32)

💡 Hint

TensorFlow tensors converted to NumPy arrays print without commas.

When converting a TensorFlow tensor to a NumPy array using .numpy(), the output is a NumPy array printed without commas between elements.

❓ Model Choice

intermediate

Choosing correct TensorFlow operation for NumPy interoperability

Which TensorFlow operation correctly creates a tensor from a NumPy array without copying data?

A tf.convert_to_tensor(np_array)

B tf.constant(np_array)

C tf.Variable(np_array)

D tf.make_tensor_proto(np_array)

💡 Hint

Look for the operation that avoids copying data and shares memory.

tf.convert_to_tensor creates a tensor that shares memory with the NumPy array if possible, avoiding data copy.

❓ Metrics

advanced

Effect of NumPy array modification on TensorFlow tensor

Given a TensorFlow tensor created from a NumPy array, what happens to the tensor's values if the original NumPy array is modified?

TensorFlow

import tensorflow as tf
import numpy as np
np_array = np.array([1, 2, 3])
tf_tensor = tf.convert_to_tensor(np_array)
np_array[0] = 100
print(tf_tensor.numpy())

A [0 2 3]

B [100 2 3]

C Raises RuntimeError

D [1 2 3]

💡 Hint

Consider if the tensor shares memory or copies data.

TensorFlow tensors created with tf.convert_to_tensor copy data from the NumPy array, so modifying the original NumPy array does not affect the tensor.

🔧 Debug

advanced

Debugging NumPy to TensorFlow conversion error

What error does this code raise and why? import tensorflow as tf import numpy as np np_array = np.array(['a', 'b', 'c']) tf_tensor = tf.convert_to_tensor(np_array)

TensorFlow

import tensorflow as tf
import numpy as np
np_array = np.array(['a', 'b', 'c'])
tf_tensor = tf.convert_to_tensor(np_array)

A TypeError: Cannot convert string to tensor with default dtype

B No error, tensor created successfully

C ValueError: Incompatible shape for tensor conversion

D RuntimeError: TensorFlow does not support string arrays

💡 Hint

TensorFlow supports string tensors natively.

tf.convert_to_tensor automatically converts NumPy arrays of strings (dtype='

🧠 Conceptual

expert

Understanding zero-copy behavior in TensorFlow and NumPy interoperability

Which statement best describes when TensorFlow tensors share memory with NumPy arrays (zero-copy) and when they do not?

A TensorFlow tensors share memory with NumPy arrays only if the NumPy array has a compatible dtype and is contiguous in memory.

B TensorFlow tensors always copy data from NumPy arrays to ensure immutability.

C TensorFlow tensors never share memory with NumPy arrays due to different memory layouts.

D TensorFlow tensors share memory with NumPy arrays only when created with tf.constant.

💡 Hint

Consider dtype compatibility and memory layout requirements.

Zero-copy sharing happens only if the NumPy array's dtype matches TensorFlow's expected dtype and the array is contiguous in memory.