COMP 3132 - Python ML, part 2

Course outline

Draft

Book

Deep Learning with Python by François Chollet
- Online audio version (karaoke style, first 3 chapters are free)
- The audio version is also available on Audible
- Book GitHub page

Introductory videos

Neural networks playlist by Grant Sanderson of 3blue1brown.com
Hello World Canada: The Rise of AI - a motivating 50 minute documentary by Bloomberg
The Neuron - how real neurons in our brain work (2 min). Remember that artificial neural networks are only "inspired" by biological neurons, and don't fully mimic how brain works. Moreover, we don't yet fully understand how brain works.
Overview of DL tools by Siraj Raval

Installing Keras on your laptop

Note: This will take some time and download several hundred MB.

If you already have Anacanoda, open Anaconda prompt and first update conda by running

conda update conda

Then install keras

conda install keras

Test it by running import keras in python, you should see output like Using TensorFlow backend or Using CNTK backend - either is good. If you see any warnings about lack of GPU, that's also ok.

If you have a recent Nvidia GPU you can try following the CNTK instructions to istall the GPU version of CNTK but this is optional.

Week 1

Lab - exercises/intro_MNIST.md
Reading for next week:
- Chapter 1 of the book (available free on book website)
- First 2 videos of the Neural networks playlist from 3Blue1Brown
Home assignment 1 - exercises/home_assignment1.md (due Jan 23rd)

Week 2

Lecture slides
- Animation of 2D convolution
- How Blurs & Filters Work - Computerphile (8 min)
Lab - exercises/convolution.ipynb
Reading for next week:
- Chapter 2 of the book (skip the heavy math)
- The remaining 2 videos of the Neural networks playlist from 3Blue1Brown
Reminder: home assignment 1 is due Jan 23rd

Week 3

Lecture slides
Lab - exercises/imdb_reviews.md
Reading for next week:
- Review model evaluation techiques - videos #5 and #7 from Kevin's sklearn videos
- Chapter 3 intro and sections 3.1, 3.2 and 3.4

Week 4

Lecture slides
Lab - exercises/boston_housing.md
Reading for next week:
- Book sections 3.5, 3.6, 3.7
- Walkthrough of the MNIST example with Keras

Week 5

Lecture slides
Lab - exercises/kaggle_housing.md

Week 6

There was no lecture
Lab. Revising the MPG dataset using this notebook from the previous semester. Replace the decision tree with a neural network.

Week 7 - Midterm

Prep advice

Solve lab exercises
Go over the MNIST video from sentdex
Read chapters 3 of the book (skip section 3.3)
Solve more lab exercises
From chapter 4 read till section 4.3
Read chapters 1 and 2 of the book

Main topics and concepts to be familiar with

Data manipulation in NumPy and Pandas
Metrics used for evaluating and comparing ML models of different types
Overfitting
Train/validation/test data splitting and why we do this
Concepts about neural nets - weights, layers, activation functions, loss functions