Dataset from files in TensorFlow - Practice Problems & Coding Challenges

Challenge - 5 Problems

❓ Predict Output

intermediate

Output of loading text files with tf.data

What is the output of this code snippet that loads text lines from two files using TensorFlow's tf.data API?

TensorFlow

import tensorflow as tf

# Create two small text files
with open('file1.txt', 'w') as f:
    f.write('apple\nbanana')
with open('file2.txt', 'w') as f:
    f.write('cherry\ndate')

# Load dataset from files
files = ['file1.txt', 'file2.txt']
dataset = tf.data.TextLineDataset(files)

# Collect all lines into a list
lines = list(dataset.as_numpy_iterator())
print(lines)

A [b'file1.txt', b'file2.txt']

B SyntaxError

C [b'apple banana', b'cherry date']

D [b'apple', b'banana', b'cherry', b'date']

💡 Hint

tf.data.TextLineDataset reads each line from all files in order.

The TextLineDataset reads each line from the files in the order they are listed. So it outputs each line as a byte string in sequence.

❓ data_output

intermediate

Number of elements in a dataset from multiple CSV files

Given three CSV files each with 2 rows, what is the number of elements in the dataset created by tf.data.experimental.CsvDataset loading all files?

TensorFlow

import tensorflow as tf

# Assume files: data1.csv, data2.csv, data3.csv each with 2 rows
files = ['data1.csv', 'data2.csv', 'data3.csv']
dataset = tf.data.experimental.CsvDataset(files, [tf.float32, tf.int32])

count = 0
for _ in dataset:
    count += 1
print(count)

A 6

B 3

C 2

D 9

💡 Hint

Each file has 2 rows, and dataset reads all rows from all files.

CsvDataset reads all rows from all files. 3 files * 2 rows each = 6 elements.

🔧 Debug

advanced

Error raised when loading non-existent files

What error will this code raise when trying to create a TextLineDataset from a file that does not exist?

TensorFlow

import tensorflow as tf

files = ['missing_file.txt']
dataset = tf.data.TextLineDataset(files)

for line in dataset:
    print(line.numpy())

A No error, prints nothing

B tf.errors.NotFoundError

C ValueError

D FileNotFoundError

💡 Hint

TensorFlow raises its own error type for missing files.

TensorFlow raises tf.errors.NotFoundError when a file is missing during dataset creation or iteration.

🚀 Application

advanced

Creating a dataset from image files with labels

You have a folder with images and a CSV file mapping image filenames to labels. Which code snippet correctly creates a tf.data.Dataset yielding (image_tensor, label) pairs?

A Use tf.data.TextLineDataset on CSV, parse lines, then map to load images with tf.io.read_file and tf.image.decode_jpeg

B Use tf.data.Dataset.list_files on images, then map to load images and assign labels manually

C Use tf.data.experimental.CsvDataset on CSV, then map to load images and parse labels

D Use tf.data.Dataset.from_tensor_slices with image paths and labels loaded into memory

💡 Hint

CsvDataset is designed to read CSV files with typed columns.

tf.data.experimental.CsvDataset reads CSV files with typed columns, allowing you to get filenames and labels. Then mapping loads images and pairs them with labels.

🧠 Conceptual

expert

Effect of interleave on dataset from multiple files

What is the main difference between tf.data.TextLineDataset(files) and tf.data.Dataset.from_tensor_slices(files).interleave(tf.data.TextLineDataset, cycle_length=2) when reading multiple text files?

A TextLineDataset reads files sequentially; interleave reads lines from files in parallel, mixing lines

B TextLineDataset reads files in parallel; interleave reads files sequentially

C Both produce the same output order

D TextLineDataset reads only first file; interleave reads all files

💡 Hint

Interleave cycles through datasets to mix their elements.

TextLineDataset reads all lines from the first file, then second, etc. Interleave cycles through files, reading lines from each in turn, mixing lines from files.