Variables and Data Types

In Python, variables are not "buckets" that hold values; they are labels (references) that point to objects in memory. In Machine Learning, understanding how these labels work is the difference between writing clean code and hunting for memory-leak bugs.

1. Dynamic Typing

Python is dynamically typed. You don't need to declare that a variable is an integer or a string; Python figures it out at runtime.

x = 0.01  # Initially a float (learning rate)
x = "Adam" # Now a string (optimizer name)

2. Fundamental Data Types in ML

Every feature in your dataset will eventually be mapped to one of these fundamental Python types.

A. Numerical Types

• int: Whole numbers (e.g., number of layers, epochs).
• float: Decimal numbers. Most weights and probabilities in ML are 64-bit or 32-bit floats.
• complex: Used in signal processing and Fourier transforms.

B. Boolean Type (bool)

• Represents True or False. Used for binary masks (e.g., selecting rows in a dataset where age > 30).

C. Sequence Types

• str: Strings are used for category labels or raw text in Natural Language Processing (NLP).
• list: A mutable, ordered sequence.
• tuple: An immutable (unchangeable) sequence. Often used for tensor shapes like (3, 224, 224).

3. Mutability: The "Gotcha" in ML

Understanding Mutability is crucial when passing data through pre-processing functions.

• Mutable (Can be changed): list, dict, set, numpy.ndarray.
• Immutable (Cannot be changed): int, float, str, tuple.

ML Common Bug

If you pass a list of hyperparameters to a function and the function modifies that list, the original list outside the function will also be changed. Always use .copy() if you want to preserve the original data!

4. Type Casting

In ML, we frequently need to convert data types (e.g., converting integer pixel values 0-255 to floats 0.0-1.0).

# Converting types
pixel_val = 255
normalized = float(pixel_val) / 255.0

# Checking types
print(type(normalized)) # Output: <class 'float'>

5. Summary Reference Table

Type	Example	Mutable?	Typical ML Usage
int	32	No	Batch size, epoch count.
float	0.001	No	Learning rate, weights.
str	"cat"	No	Target labels, text data.
list	[1, 2, 3]	Yes	Collecting loss values over time.
tuple	(28, 28)	No	Input dimensions of an image.
dict	{"id": 1}	Yes	Storing model configuration.

---

Now that we know how to store single values and lists, we need to know how to organize them logically for complex tasks. Let's look at more advanced data structures.