wirick / mlcode Goto Github PK

This repository contains various code that is applicable to machine learning.

I'm going to be doing some ML work now.
So I'm going to be doing a lot
of math, and a lot of ml fun.

The Probability Model can create uniform, single
and multivariable normal, categorical, and mixture model samples from
distributions with a specified mean and (co)variance. the plotsample.py is
a script that I wrote with plotting functions so that I could test my models
by eye. I use the Box Muller transform and the Cholesky decomposition for normal samples.

regression.py is the skeleton for the first assignment for UC Berkeley's 2011 machine
learning course. It does k-local linear regression on a location, phase, station pair
to predict the time residual of a seismic event. other code is a precomputation procedure
to determine the k-local linear regression. It is too big for my computer to manage without
being painfully slow, so now I think I'll have to use some other trick up my sleeve to try and
work it out.

util.py contains helper functions. K-nearest neighbors, k-fold cross validation, and 
some functions created to make a toy test for my cross-validation function, which predicted
a plane nonlinear equation using nearest neighbors

util.discrete_histogram(data, attr, *computationFn) takes a data file(csv) and discrete attributes(list), or one computed from those attributes and returns (value, absolute_frequency, relative_frequency) tuples. the computation function is the identity for one attribute if none other is supplied
util.confusion_matrix(attr1, attr2) returns a matrix M indexed by the values of each attribute such that M_kl is the fraction of records that have Z = z_l out of all records having X = x_k. It also returns the column values and the row values in order of their matrix indicies.

plotregression.py is a visual test for my cross validation functions.