In paulemms/datamining: Data mining

Introduction

This is Lab 2 on Data Mining.

First we load the datamining package and the data for the lab.

library(datamining)

The data set

Interspersed throughout this lab are questions that you will have to answer at check-off. 1. Download the files for this lab from the course web page to the desktop: http: //www.stat.cmu.edu/~*minka/courses/36-350/lab/ 2. Open a Word or Notepad document to record your work. Start R. 3. Start -> All Programs -> Class software -> R 1.7.0 4. Load the special functions for this lab: File -—-> Source R code... Browse to the desktop and pick lab2.r (it may have been renamed to lab2.r.txt when you downloaded it). Another window will immediately pop up for you to pick the mining.zip file you downloaded. The image dataset o. The dataset is in a table called imgs, containing 20 images of “Action Sailing” and 20 images of “Auto Racing”. They have already been converted into color counts over 64 prototypical colors in the HSV cube. A vector of labels named img.labels is also defined. How many pixels in sailing5 have color thistle? 6. Using the commands from lab 1, compute a matrix of distances between the images. For example: imgs = remove.singletons (imgs) imgs = idf.weight (imgs)

x = div_by_euc_length(imgs)

d = distances(x) Nearest-neighbor classification /. For this lab, let’s pretend that the mages sailingi-sailing10 and racing11—racing20 just arrived, and we don’t know what class any of them belongs to. These are the “test” images. The other images are “training” images. Using the commands described below, create a variable called i.test which holds the row numbers of the test images, and a variable called i.train which holds the row numbers of the training images. If you do this right, imgs[i.test,] should be the test images and imgs[i.train,] should be the training images. 8. Using the index vectors and the function closest, determine the training image which is the nearest neighbor of each test image. Out of the 20 test images, which would be misclassified by the nearest-neighbor method? Nearest-prototype classification 9. Using the training images and the labels in img.labels, compute prototypes for sailing and racing: xp = prototypes (imgs|i.train,|,img.labels[i.train] ,sum) 10. Concatenate the prototypes with the rows for test images, to get a new data matrix, and compute all distances. Note that IDF weights need to be assigned before taking prototypes, and division by Euclidean length needs to be done after taking prototypes. Out of the 20 test images, how many are misclassified, and which are they? Which is the better classification method here? 11. You can now get checked off. Constructing vectors ‘There are two basic ways to create a vector: 1:10 0(1,2,3,4,5,6,7,8,9,10) The first way creates a vector of all integers from a starting value to an ending value. ‘The second way uses the function c to concatenate a bunch of numbers together. You can also use it to concatenate vectors, such as c(1:10,15:20) The result can be assigned to a variable: x = 5:10 x13] # answer is 7 Selecting a block from a matrix You know that you can use numbers and names to select from a matrix, as in imgs["racing4",4]. But you can also use vectors of numbers or names, to select an entire block of the matrix. For example, imgs[,1:4] is the first 4 columns of the matrix, imgs[c(2,4),] is rows 2 and 4 of the matrix, and imgs[c (2,4) ,1:4] is the first 4 columns of rows 2 and 4. Computing nearest-neighbors Ifd is a distance matrix, then closest(d[i.train,i.test]) will tell you which training image is closest to each of the test images. In general, closest takes a sub-block of a distance matrix, and returns the closest row for each column. Concatenating matrices ‘To concatenate the rows of matrix x with the rows of matrix y, type: zZ = rbind(x,y) This defines a new matrix z containing all rows.

paulemms/datamining documentation built on March 1, 2023, 4:01 p.m.