Coding Ethics

You might be a great programmer/ coder, but as far as my experience tells me, never get carried away by your instincts to sit and directly write a piece of code. Start with the algorithm first ,and then proceed towards writing the code, or you will spend double the time on the same piece of code.

Always!!

Required Fixation Locations

I-DT Algorithm

The low matching scores for the raw dataset of Eye Movement Tracking is being encountered mainly due to a large amount of noise and variation in the data. Therefore, the method employed for pre-processing the dataset is of cardinal Importance. Salvucci et al have showed many ways of finding the right fixations, so that it could be attached to the RFLs. I have tried out Velocity Based Algorithms and Dispersion Based Algorithms on the dataset. The scores should now drastically improve.

Next Course Up

After finishing with my Machine Learning course through online video lectures by Prof Pedro Domingos, the next course i am looking forward to is Artificial Neural Network. I am excited about the course, and wish to learn just more than the Algorithms, but muse over, as how it can be applied in a more efficient manner for Eye Movement Tracking Project i am working on currently.

Machine Learning

The roller-coaster ride of the Machine Learning is going to end today. The last series of lectures includes Unsupervised Learning. It has been a beatific journey of completing a 10 week course in just a mere 15 odd days. I have learned a lot from this course not only in the terms of the basic machine learning algorithms, but ways to approach a new set of problem with the given repertoire and extrapolating it to solve it. I am highly thankful to Coursera and most of all Prof Pedro Domingos for such pristine concepts. I really appreciate the style of teaching of the professor, as he concentrates more on reasoning than just merely giving the straight forward algorithms. I totally compare his way of teaching to that of Socrates, "Learning By Questioning".

Scores

Working on this piece of code, i feel like that i am doing more work on Data Mining than core Eye Movement.

MATLAB Command Window

Snippet from Score Matching Code

 I have been working on calculating the matching scores for my project, and it is getting quite complicated, as i am not getting the type of results one would like to achieve for an Authentication Access System. I am therefore trying all the permutations and combinations of score matching, in the book. Lets hope, i have a breakthrough.

Lazy Learner

Just saw a video on how technology has progressed over the years, and what drives it, better than those technologies that failed by Joicho Ito, MIT Media Labs. This was a stunning piece of presentation given by the Director, Media Labs. In the presentation, Joichi Ito talks about his unique approach to innovation in the age of internet and shares his 9 principles which include 'risk over safety', ' practice over theory', and 'disobedience over compliance'. One of the striking feature of the presentation that really hit me was the concept that i have now been hearing over a month now, at my Machine Learning online Class i.e The Lazy Learner. This concept is omnipresent and can change the face of new ways of solving a problem. Joicho mentions the same principle in relation to the Internet Databases.

Please check out the video at:

Link To Video

New courses

Finally, i am done with my practicals, and i have almost a month before the next semester commences, to assimilate some new concepts into my kitty. I plan to complete the Machine Learning course, in which i am already into the third last week of the video lectures. I am looking forward to learn Fuzzy Networks and Comprehensive insight into Artificial Neural Networks. This month is gonna be full of fun learning new and interesting concepts.

5 Hour Coding Spree

The second phase of the code has some glitches. I know i have made a small mistake somewhere in the code, which is the toughest task to catch.
Debugging begins, have to check the code line by line first, and if still the bug remains unscathed, i will have to write the complete code once again.

One Full Length Code vs Fragmented Code

This really varies person to person. The first phase of coding, i used Fragmented Code, just fired the .m files in order, because it was relatively easier to manage between the m files. Now for the second phase of coding, i am finding a it a little cumbersome, as the number of m files has increased drastically. Therefore, i am using a single m file for the second round, and this process is quite fast than what i employed in the first phase. The code is about 200 lines, and counting......

Scores

The feature level scores are not coming as i had supposed them to be, i have to drop the feature level score matching for the moment and concentrate on the other metrics.

Coding In Matlab

So, my end-term exams are finally over, and i am back onto my project with full throttle.
Before my exams, one piece of code was not completed, due to time constraints. I have sat down to complete the code now, and thank god for my Comments in the Code i am still able to make out which line fires what process. If i had failed to write the comments in the code, i would have ended up writing the entire code once again.

Golden Rule: However easy the code might be, comment as what you are doing, otherwise it would seriously hamper your progress, once you return back to the code to analyze it again.