Python for Machine Learning

Python, Numpy and Matplotlib basics with examples, for Machine Learning beginners.

Content summary

The lectures were presented during the course Python Programming for Machine Learning, at TU Berlin: Technische Universität Berlin, in the 2017's Summer Semester, held by Professor Grégoire Montavon.

Each of the 4 lectures are Jupyter Notebooks, exported as pdf files, with examples of basic concepts from subjects related to Python, Numpy, Matplotlib and Machine Learning algorithms.

In the assignments folders one can find the 4 given homeworks, corresponding to each lecture. Each folder contains the original requirements, the solved version with feedback from the reviewers and any additional material generated or used during the task.

Lecture 1

Introduction into Python programming with:

• the traditional "hello world";
• some data structures: lists, dictionaries;
• functions & classes;
• Functional Programming concepts: list comprehension, map, filter & !reduce!;
• ! Reading Data from Files !

Have some fun by following a simple Python implementation of a Decision Tree.

Try to distinguish between fruits based on their color and size.

Lecture 2

Introduction into Numpy and Matplotlib libraries.

Why and when should one use Numpy? You will see time performance arguments and useful tricks for matrix manipulations (such as indexing, reshaping, broadcasting, ...).

How to plot graphs?

Take a grasp of the Boston dataset from sklearn library. Analyze and find insides about the data.

Lecture 3

Simulate random processes, such as Monte Carlo Markov Chain, with Numpy.

Brief talk about integrating highly optimized functions, written in C/C++, Fortran or Cuda, into your Python code.

Apply these concepts in image filtering.

Lecture 4

Rounding, Overflow and Linear Algebra.

Analyze what problems can appear when working with large numbers in Numpy.

Short implementation of the Linear Regression and the Principal Component Analysis algorithms.

Final Exam

Test the main topics discussed in the lectures.

Universal Approximation

Presentation made for the Neural Networks Seminary. Proof of the Neural Networks' Universal Approximation Theorem.