# Table of Contents 1. [Context](#org6dbe148) 2. [Structure of the Repository](#org65603ff) 3. [Problems](#orga34978d) 4. [Learning](#orgcfaa906) 5. [Projects](#orge067afe) # Context This repo is my 10,000 Hours of Machine Learning. **IMPORTANT: this repository eventually grew in scope and thus [abaj.ai](https://abaj.ai) was born. All of the files in this repository are a submodule of that repo, and all of the code is tangled with org files in the site** > "Machine Learning is just lego for adults" - Dr. Kieran Samuel Owens > "S/he who has a why, can bear almost any how." - Friedrich Nietzche *Why do anything else, when the thing that you could do, would do, everything else?* --- To become an expert at anything, there is a common denominator:

10,000 hours of deliberate practise on the subject.

# Structure of the Repository There are 3 main features: 1. Problems - Briefly, these are solutions to **classical** problems, MNIST, Boston Housing, XOR, etc. 2. Learning - This folder contains textbook exercises, open courseware material, youtube lecture series and any UNSW coursework that it is appropriate for me to publicly version control. The folders contain highly non-trivial code: Shakespeare speaker, GPT-2 implementation, etc. 3. Projects - These are my more complex and non-trivial projects. They are more fun, but their scopes are undefined / defined loosely by myself. There is a Kanye West chatbot based on the Attention architecture. There is also a Peg Solitare Reinforcement Learner and plans for Ultimate Frisbee Rules App that is fine-tuned. # Problems In no particular order, here are a list of the methods you will find in the notebooks. The emphasis is on understanding their limitations, benefits and constructions. - Least Squares Regression - Random Forests - Boosting, Bagging - Ensemble Methods - Multilayer Perceptrons - Naive Bayes - K-means regression - K-nearest Neighbours Clustering - Logistic Regression - Decision Trees - SVM - Kernel Methods - GAN's - Stable Diffusion - Recurrent Neural Networks - Convolutional Neural Networks - Transformers - word2vec, GLoVE and NLP - LLM To gain proficiency in all of the above methods, I have solved classical problems that lend themselves well to that particular method:

Dataset	Accuracy	Model
MNIST	92%	Logistic Regression
FMNIST	B%	Random Forest
KMNIST	C%	2-layer CNN
CIFAR	D%	CNN
IRIS	E%	SVM
ImageNet	F%	ResNet50
Sentiment140	G%	LSTM
Boston Housing	H%	Linear Regression
Wine Quality	I%	Gradient Boosting
Pima Indians Diabetes	J%	Decision Tree
IMDB Reviews	K%	BERT
KDD Cup 1999	L%	K-Means Clustering
Digits	M%	Gaussian Mixture Model
CartPole	N%	Deep Q-Network

# Learning For mastery, a formal education is also required; either by way of open-courseware, or by paying an institution. I have done both, and overall benefitted as a result. - [X] Zero to Hero - Andrej Karpathy - [X] Neural Networks and Deep Learning - Michael Nielsen - [X] UNSW AI - [X] UNSW Machine Learning and Data Mining - [X] UNSW Deep Learning and Neural Networks - [ ] UNSW Computer Vision - [ ] Stanford CS229 (Machine Learning) - [ ] Stanford CS230 (Deep Learning) - [.] Mathematics for Machine Learning, Ong et al. - [o] HOML (Hands on Machine Learning) - [ ] All of Statistics, Larry Wasserman - [X] Coursera Machine Learning Specialisation - [X] Coursera Deep Learning Specialisation # Projects To become proficient, I have applied my ML skills to solve problems of personal and social interest. - [X] Kanye West Producer - [X] KiTS19 Grand Challenge: Kidney and Kidney Tumour Segmentation - [X] Ultimate Frisbee Rule App - [X] OCR - [X] Peg Solitaire RL > "Read 2 papers a week" - Andrew Ng