Prerequisite Review & refresh

🐍 Python Foundations & Advanced Features

PyTorch is a Python library, and idiomatic Python makes deep-learning code short and readable. Beyond the basics, a handful of features appear constantly in PyTorch and data code; know them and the framework stops feeling mysterious.

Core Python

Data types (numbers, strings, lists, dicts), control flow (if, for, while), functions, and modules.
Classes and objects: bundling related data and behavior together (the basis of most libraries).
Slicing and negative indexing to read parts of a sequence.

Idiomatic and advanced features

List, dict, and set comprehensions: build a collection in one readable expression.
Generators and iterators (the yield keyword): produce values lazily, one at a time.
lambda (small anonymous functions) and *args / **kwargs (flexible argument lists).
enumerate (index with item), zip (pair up iterables), f-strings, and context managers (the with statement).
Decorators at a reading level: a function that wraps another to change its behavior.

Useful dunder (magic) methods

__init__: the constructor that sets up a new object.
__len__ and __getitem__: make an object respond to len(obj) and obj[i].
__call__: make an instance callable like a function, so obj(x) runs the object's logic.
__repr__: a readable text representation of an object, useful for debugging.

NumPy versus Python lists

An ndarray stores one numeric type in contiguous memory; a Python list holds arbitrary objects and is slower.
Vectorized operations apply to a whole array at once, replacing explicit element-by-element loops.
Broadcasting (combining different but compatible shapes) and dtypes (the element type).
Converting between arrays and tensors with simple helper functions.

What recurs in PyTorch and DL code

Fancy indexing and boolean masks to select elements by position or by a condition.
Broadcasting to combine arrays of different shapes without writing loops.
In-place operations (methods ending in an underscore) that modify data directly.
Moving data between the CPU and the GPU.

Readiness check

Rewrite a for-loop that builds a list as a comprehension.
Write a generator that yields batches from a list.
Implement a tiny class with __len__ and __getitem__.
Explain why a NumPy vectorized op beats a Python loop.

❓Self-check questions

Multiple-choice questions on the topic itself. Pick an answer, then reveal it. If several are unclear, work through the review above first.

1. The list comprehension [x*x for x in range(4)] produces:

[0, 1, 2, 3]
[1, 4, 9, 16]
[0, 1, 4, 9]
[0, 2, 4, 6]

Show answer

Correct: C. range(4) is 0,1,2,3; their squares are 0,1,4,9.

2. A generator function is one that uses:

return
yield
lambda
async

Show answer

Correct: B. yield turns a function into a generator that produces values lazily, one at a time.

3. The main advantage of a generator over a list is:

faster indexing
lazy, memory-efficient iteration
automatic sorting
built-in JSON support

Show answer

Correct: B. A generator produces items on demand without holding the whole sequence in memory.

4. Which dunder methods let an object support len(obj) and obj[i]?

__init__ and __str__
__len__ and __getitem__
__call__ and __repr__
__add__ and __eq__

Show answer

Correct: B. __len__ backs len(); __getitem__ backs indexing and iteration (exactly what a Dataset implements).

5. Defining __call__ on a class allows:

printing the object
calling the instance like a function: obj(x)
adding two objects
iterating it

Show answer

Correct: B. __call__ makes instances callable; e.g. model(x) runs the object's logic.

6. zip([1, 2, 3], ['a', 'b', 'c']) yields:

(1, 2, 3, 'a', 'b', 'c')
(1, 'a'), (2, 'b'), (3, 'c')
a dictionary
[1, 'a', 2, 'b']

Show answer

Correct: B. zip pairs elements positionally from the iterables.

7. enumerate(['a', 'b']) yields:

'a', 'b'
(0, 'a'), (1, 'b')
(1, 'a'), (2, 'b')
['a', 'b']

Show answer

Correct: B. enumerate pairs each item with its index, starting at 0 by default.

8. In def f(*args, **kwargs), kwargs is:

a list of positional args
a dict of keyword args
a single value
the return value

Show answer

Correct: B. *args collects extra positional arguments as a tuple; **kwargs collects keyword arguments as a dict.

9. The slice a[1:4] returns the elements at indices:

1, 2, 3, 4
1, 2, 3
0, 1, 2, 3
2, 3, 4

Show answer

Correct: B. Slicing is start-inclusive and stop-exclusive: indices 1, 2, 3.

10. a[-1] refers to:

an error
the first element
the last element
index minus one, which is skipped

Show answer

Correct: C. Negative indices count from the end; -1 is the last element.

11. The main reason a NumPy array is faster than a Python list for numeric work:

it uses more memory
typed, contiguous storage with vectorized operations
it is written in Java
it caches results

Show answer

Correct: B. Homogeneous typed memory and vectorized C-level operations avoid slow per-element Python loops.

12. NumPy broadcasting aligns array shapes starting from:

the leading (first) dimension
the trailing (last) dimension
the largest dimension
alphabetical order

Show answer

Correct: B. Shapes are compared from the trailing axis; sizes must be equal or one of them 1.

13. A Python list is ____ and a tuple is ____:

immutable; mutable
mutable; immutable
both mutable
both immutable

Show answer

Correct: B. Lists can be modified in place; tuples cannot.

14. The expression {k: v for k, v in pairs} is a:

set comprehension
generator
dict comprehension
lambda

Show answer

Correct: C. The {key: value for ...} form builds a dictionary.

15. A lambda in Python is:

a class
a small anonymous function
a loop
a module

Show answer

Correct: B. lambda defines a short, unnamed function, e.g. lambda x: x + 1.

📚Refresher resources

Refresh

The Python Tutorial (official) docs.python.org

Refresh

Real Python: List Comprehensions realpython.com

Refresh

Real Python: Generators realpython.com

Refresh

NumPy: the absolute basics for beginners numpy.org

← All prerequisites Course home