Skip to content

python.mdx

Latest commit

 

History

History
121 lines (85 loc) · 3.57 KB

python.mdx

File metadata and controls

121 lines (85 loc) · 3.57 KB
title Python for Machine Learning
sidebar_label Python
description Mastering the Python essentials required for ML: from data structures to vectorization and the scientific ecosystem.
tags
python
programming
numpy
pandas
mathematics-for-ml

Python is the "lingua franca" of Machine Learning. Its simplicity allows researchers to focus on algorithms rather than syntax, while its robust ecosystem of libraries provides the heavy lifting for mathematical computations.

1. Why Python for ML?

The power of Python in ML doesn't come from its speed (it is actually quite slow compared to C++), but from its ecosystem.

mindmap
  root((Python ML Ecosystem))
    Data Processing
      Pandas
      NumPy
    Visualization
      Matplotlib
      Seaborn
      Plotly
    Modeling
      Scikit-Learn
      PyTorch
      TensorFlow
    Deployment
      FastAPI
      Flask
Loading

2. Core Data Structures for ML

In ML, we don't just store values; we store features and labels. Understanding how Python holds this data is vital.

Structure Syntax Best Use Case in ML
List [1, 2, 3] Storing a sequence of layer sizes or hyperparameter values.
Dictionary {"lr": 0.01} Passing hyperparameters to a model.
Tuple (640, 480) Storing immutable shapes of images or tensors.
Set {1, 2} Finding unique classes/labels in a dataset.

3. The Power of Vectorization (NumPy)

Standard Python for loops are slow. In ML, we use Vectorization via NumPy to perform operations on entire arrays at once. This pushes the computation down to optimized C and Fortran code.

import numpy as np

# Standard Python (Slow)
result = [x + 5 for x in range(1000000)]

# NumPy Vectorization (Fast)
arr = np.arange(1000000)
result = arr + 5

Multi-dimensional Data

Most ML data is represented as Tensors (ND-Arrays):

  • 1D Array: A single feature vector.
  • 2D Array: A dataset (rows = samples, columns = features).
  • 3D Array: A batch of grayscale images.
  • 4D Array: A batch of color images (Batch, Height, Width, Channels).

4. Functional Programming Tools

ML code often involves transforming data. These three tools are used constantly for feature engineering:

  1. List Comprehensions: Creating new lists from old ones in one line.
  • normalized_data = [x / 255 for x in pixels]
  1. Lambda Functions: Small, anonymous functions for quick transformations.
  • clean_text = lambda x: x.lower().strip()
  1. Map/Filter: Applying functions across datasets efficiently.

5. Object-Oriented Programming (OOP) in ML

Most ML frameworks (like Scikit-Learn and PyTorch) use Classes to define models. Understanding self, __init__, and inheritance is necessary for building custom model pipelines.

classDiagram
    class Model {
        +weights: array
        +bias: float
        +train(data)
        +predict(input)
    }
    Model <-- LinearRegression
    Model <-- LogisticRegression
Loading

6. Common ML Patterns in Python

The Fit-Transform Pattern

Almost all Python ML libraries follow this logical flow:

flowchart LR
    A[Raw Data] --> B["fit() : Learn parameters from data"]
    B --> C["transform() : Apply changes to data"]
    C --> D["predict() : Generate output"]
    style B fill:#e1f5fe,stroke:#01579b,color:#333
    style D fill:#f9f,stroke:#333,color:#333
Loading

Python provides the syntax, but for heavy mathematical operations, we need a specialized engine. Let's dive into the core library that makes numerical computing in Python possible.