Week 5

Now that I can compute texture measures on any area, I need to be able to classify the texture measures and to find out if and how useful they are. To that end I will learn to use WEKA. WEKA has a built in library of classification algorithms.

At this Tuesday’s meeting the topic of our ground truth came up. Some members of our group have questions as to how representative our ground truth is, and what effect deleting or adding images would have in our over all classification ability.

I have arranged to meet with other members of our team working on this project as well as with an Electrical Engineer experienced working with crystallographers. If he cannot answer our questions, he may be able to facilitate a meeting with one of the crystallographers associated with the university.

---

To better acquaint myself with various classifiers that can be implemented through WEKA I read: ‘Weka: Practical Machine Learning Tools and Techniques with Java Implementations’ by Ian H. Witten et al; ‘Constructing Diverse Classifier Ensembles using Artificial Training Examples’ by Prem Melville and Raymond J. Mooney; ‘Learning from Imbalanced Data Sets with Boosting and Data Generation: The DataBoost-IM Approach’ by Hongyu Guo and Herna L. Viktor, and parts of the On line documentation.

Additionally, I spent something like an hour and a half with two other members of the vision group in a mini tutorial: the format of a weak document, the best way  to implement an algorithm to produce a weak document, and what are probably the best classifiers to use for my purposes.

--

In order to create the adequate Matlab program to write the WEKA document I needed, I had to learn about Matlab’s input / output functions. I did this by consulting a Matlab book and some online tutorials I found.

Another member of the vision group and myself met  with an expert regarding our ground truth. The meeting was helpful in various regards. I was involved in asking for, scheduling, and participating in a meeting with an outside expert. The expert took us to the lab where the crystals are grown, showed us the robo-microscopes that take the images we analyze, showed us the computers that store the archives of the images we analyze, showed us the process and several of the applications in which our algorithm can be used.

I created an algorithm that computes the 5 texture measures on an interest region, and outputs them on a line of a document that has been formatted to be used by weak. I give my program as input the names and locations of the images I want it to analyze.

Over the course of the past weeks I have come to use a few linux / unix commands to interact with the linux server that holds Matlab. At first, having very limited experience with unix / linux, the prospect of interacting with that server, even on a limited basis, seemed somewhat overwhelming. Today I am comfortable interacting with it, and I have learned a few other than eth very basics I needed, such as | grep – which make sit much easier to find files that have common parts, such as a pgm ending. I also learned about the various kinds of images, such as jpg, ppm, pgm, and how one way they are differentiated is through the level of compression when they are being stored (saved).

Becoming fluent an conversant with various aspects of the world of computer science are side effects of just being exposed to them, working with them, and they become a part of a world that is known. The more knowledge one builds, the more one seeks an is comfortable with exploring the various areas.

This weekend I’m going hiking again, this time to Mt. rainier. There is supposed to be patches of snow still.