项目作者: ElliotDahl1986

项目描述 :
ANN, RL, Gaussian process
高级语言: Jupyter Notebook
项目地址: git://github.com/ElliotDahl1986/ML-Projects-From-Scratch.git
创建时间: 2018-01-05T19:39:19Z
项目社区:https://github.com/ElliotDahl1986/ML-Projects-From-Scratch

开源协议:

下载


ML-Projects-From-Scratch

About

In ML-Projects-From-Scratch I explore the mathematics behind various machine learning algorithms by building them myself and doing some simple predictions.

Dependencies

ML-Projects-From-Scratch is tested to work under Python 3

Contains

ANN:

A artificial neural network implemented from scratch. I learn the weights and biases through batch gradient descent. I test the ANN by identifying handwritten numbers (a classic data set found here Number data set).

The backward propagation,



RL:

Reinforcement learning implented from scratch. I investigate the use of combined modules having their own Q-table. I apply Gibbs policy improvement. I apply this to an environment containing various rewards.

Gibbs policy improvement,


screen shot 2018-01-11 at 3 55 36 pm

Q-table update,


screen shot 2018-01-11 at 3 55 25 pm

Gaussian Process:

Gaussian process used to predict time-series data for motion movement. The hyper parameters are found by maximizing the log likelyhood function


screen shot 2018-02-22 at 4 18 33 pm

where Q is the kernel given by,


screen shot 2018-02-22 at 4 19 04 pm

I use steepest ascent to find the hyper parameters,


screen shot 2018-02-22 at 4 19 11 pm

Resources:

Bishop, C. (2006). Pattern Recognition and Machine Learning. Springer-Verlag New York

Excelent list of free, open source books on machine learning, statistics, data-mining, etc.