Validation split in TensorFlow - Practice Problems & Coding Challenges

Challenge - 5 Problems

❓ Predict Output

intermediate

What is the size of the validation set?

Given the code below, what is the number of samples in the validation set?

TensorFlow

import tensorflow as tf
from tensorflow.keras import layers

(x_train, y_train), _ = tf.keras.datasets.mnist.load_data()

model = tf.keras.Sequential([
    layers.Flatten(input_shape=(28, 28)),
    layers.Dense(10, activation='softmax')
])

history = model.fit(x_train, y_train, epochs=1, batch_size=32, validation_split=0.2)

A 12000

B 10000

C 20000

D 6000

💡 Hint

The MNIST training set has 60,000 samples. Validation split 0.2 means 20% of training data is used for validation.

The training data has 60,000 samples. 20% of 60,000 is 12,000, so the validation set size is 12,000 samples.

❓ Model Choice

intermediate

Which validation_split value splits 25% of data for validation?

You want to reserve exactly 25% of your training data for validation during model.fit. Which validation_split value should you use?

A 0.25

B 0.75

C 0.5

D 0.2

💡 Hint

validation_split is the fraction of training data used for validation.

validation_split=0.25 means 25% of the training data is used for validation.

❓ Hyperparameter

advanced

What happens if validation_split is set to 0.0?

In TensorFlow's model.fit, what is the effect of setting validation_split=0.0?

A The entire dataset is split equally between training and validation

B All data is used as validation data

C An error is raised because 0.0 is invalid

D No validation data is used during training

💡 Hint

validation_split controls the fraction of data reserved for validation.

Setting validation_split=0.0 means no data is reserved for validation, so training uses all data.

🔧 Debug

advanced

Why does validation_split not work with a tf.data.Dataset?

Consider this code: import tensorflow as tf train_ds = tf.data.Dataset.from_tensor_slices((x_train, y_train)) model.fit(train_ds, epochs=3, validation_split=0.2) Why does this raise an error?

A validation_split must be an integer, not a float

B tf.data.Dataset objects cannot be used for training

C validation_split only works with NumPy arrays or tensors, not tf.data.Dataset

D The batch size is missing, causing validation_split to fail

💡 Hint

Check the data type compatibility of validation_split parameter.

validation_split works only when input data is a NumPy array or tensor, not a tf.data.Dataset object.

🧠 Conceptual

expert

Why is validation_split applied before shuffling in model.fit?

In TensorFlow's model.fit, the validation_split is applied before shuffling the training data. What is the main reason for this behavior?

A To speed up training by reducing shuffling overhead

B To ensure the validation set is a fixed subset of the original data, not affected by shuffling

C To shuffle the validation data separately from training data

D To randomly assign samples to validation or training sets each epoch

💡 Hint

Think about consistency of validation data across epochs.

Applying validation_split before shuffling ensures the validation set is always the same fixed subset, providing consistent evaluation.