HOME PROJECT JOURNAL |
When i met with Dr Moses on my first day (05/31/2011), she explained that she wanted me to work on optimizing the AFSA
to handle seperating multiple AFS runs into seperate files and then analyze the data from these files. I finshed this on the 06/01/2011.
The next step was to do two runs for using the old AFS;
1) Using larger colony sizes for the threes ant speices and the the same GA varibles from last year.
2) Using the distances for Maricopa, for both Desertorum and Rugosus.
Dr Moses wants to see if distance plays any significant role in the foraging habits of the ants.
I completed all the runs for 32 seed piles on Friday .
There wasn't much for me to do this week, i was still waiting on Kenneth to send me
the new AFS. I focused on finishing the runs from last week using 256 seed piles
as Dr Moses had asked.
Kenneth finally emailed me the source code of the new AFS model Tuesday night.
I started work on modifying the model to run for as much iteration as needed,
as opposed to running only once, the following day. It took me about 2 days to modify the program.
In the process of modifying the source code i ran into a few hiccups but was able to solve the problems
after looking at the old AFS model and exchanging emails with Kenneth.
Dr Moses, Kenneth and I met on Friday to discuss the foraging scenarios we should investigate using the new model.
And we decided to run simulations varying seed pile distances from the nest and the amount of
back ground food (i.e. The naturally occurring food). I started running the different scenarios, some scenarios ran fine but for others
it went into an infinite loop. I emailed Kenneth about the scenarios that did not work to see if he had any idea why they went into an infinite loop.
Kenneth and I sat down to fix the infinite loop problem for some of the scenarios I was running
simulations for. While looking through the source code for the AFS, we realized the problem was
with the block of code responsible for setting up the seed distributions on the grid for the ants to collect.
The ants were not efficiently collecting seeds in these scenarios and as a result there were seeds left
over from previous iterations which were confusing the ants. To solve the problem, Kenneth included a block of code
that would make sure that all seed distributions are set to zero at the end of each iteration.
The evolved GA values Kenneth sent me are broken into 4 categories;
1. Recruitment: Using these GA values the ants will forage based on pheromone trails.
2. Patch: Using these GA values the ants will forage based on site fidelity (i.e. based on their remembrance of where they last picked up food from).
3. Both: Using both their memory and pheromone trails to forage for food.
4. None: Using neither pheromone trails nor memory.
I completed the runs for the recruitment GA values on Thursday.
Aside from working with the new AFS, I also did some work on the GUI version of the AFSA.
I worked on the runs for the other 3 categories and carried
out the statistical analysis for each of them.
They were the only things I worked on this week,
because I was dealing with a lot of data and did not want to make any mistakes.
This week I focused all my attention on finishing the GUI version of the AFSA, which I finished on Thursday.
I used the GUI AFSA to analyze old data I had already analyzed to make sure I got the same results.
I noticed some inconsistences in the Pile Seed Ratio values.
In the process of trying to figure out what was wrong,
I noticed I had forgotten to make use of a certain function that empties data structure I use
to hold all the seed value data.
it can function well as a standalone application, outside of my Netbeans IDE. I use a windows 7 O.S.
After determining it will work fine, I deployed it on my linux desktop at work, after a some research and
poking, I got it to work. Unlike windows, I realized I had to start it from the command line and not
double clicking the icon.
I worked on using Minitab15 to carry out an ANOVA test comparing the results from the simulations to
to field data. The results were different from what we expected and didn't make any sense.
Kenneth decided to take a look the source code for the new AFS to see if something had gone
due to my modifications. It turns out i commented out the function responsible for placing
natural food . We corrected the mistake and i ahd to run the simulations again.
I spent the whole week and weekend working on them.
I worked on using Minitab15 to redo ANOVA test to compare the model data to the field data.
The data still came out different from what we expected but it made sense. The model runs i had done
so far had the background seeds randomly distributed but not in piles. Next step was to run simulations
using a power law distribution. With a power law distribution, the background seeds are
distributed randomly in piles and individually. The results of the ANOVA test weren't to different
from the tests carried out to compare field data to the first set of runs.
I started working on my final report, doing a few rewrites after corrections and suggestion from Dr Moses.
I also spent some time redesigning my DREU webpage.