First results for lambda-WASP
Wednesday 2010-01-27 15:55
Seems like I finally got lambda wasp implemented. Althrough it probably isn't stable yet I managed to get som initial results. So for a trainingset of 100 items and 50 test items I got a precision of 0.6666667 a recall ocf 0.16 and a willingess of 0.24. As it is above my avarage for WASP I take it as a good sign.
Improved parallelization
Wednesday 2010-01-06 19:17
I have managed to get the parallelisation code to return it's errors to the controlling host. A majore improvement in debuging conviniance. Althrough the code is by no means easy to work with only because of that, I just learned the hard way that it's a big no no to ever print lisp forms in the distributed code. I was unkowingly printing a list and this caused invalid values to be returned to the master host, namly the values I printed. If only clisp hade some way of distingushing stderr and stdout better. At least I know wath to be carefull with in the feature.
Writing
Thursday 2010-03-04 17:43
The writing has started to come along. I have so far written about 12 pages of text and am not out of material to write about yet. The current goal set at a meeting with mike today is to reach about 20 pages by tuesday morning. A target that sould be reachable and possibly surpassed if I'm ont to lazy.
Fixed a bug in lambda-WASP that caused incorrect rules
Friday 2010-02-05 13:45
I have found and fixed (by ignoring) a bug in lambda-WASP where rules where extracted for nodes in the parse tree that aligned with a substring of the nl that was a substring of another node that was not its ancestor. This caused rules on incorect syntactic form which caused the parser to crash.
Failing to Understand C-Phrase Dialog
Friday 2009-11-13 10:51
Frustraitingly enough C-Phrase has a function that works well the first time I call it (including returning values) but blocks indefinitly the second time. I think this has to do with something in called dialogs or sessions but I have failed to figure out how they work. Hopefully mike will turn up soon.
Parsing problems
Tuesday 2009-12-15 13:13
When expanding the posible rule formats I have either missed some limitation of the parser or struck a bug. The parser seems to infinitly recurse and causes a stack owerflow which aborts the program. So that makes me a bit stuck.
Initial test results produced
Tuesday 2010-01-12 16:06
I have finally managed to produce some statistics showing the preformance of my system. Running the WASP-algorithm with a control group of 50 items gives the following statistiks.
| Items | Precision | Recall | Willingness |
|---|---|---|---|
| 100 | 0.56 | 0.20 | 0.46 |
| 150 | 0.87 | 0.40 | 0.46 |
| 200 | 0.82 | 0.36 | 0.44 |
Playing Around with C-Phrase
Thursday 2009-11-12 18:07
I have been playing around with c-phrase a bit. Trying to understand how it processes queries, so far with mized success but thats a problem for tomorrow.
Solution Somewhat Parallelized
Friday 2010-01-01 17:24
Well now it seems like I'm back after a Christmas break. I have finished the parallelization I started before Christmas and while it made a significant improvement I'm still a bit disappointed. The improvement was about 3 times faster rather than the ten to fifteen times I had hoped for, the cause is a rather large overhead that prevents me from improving some parts of the code (the got slower parallelized due to overhead). But on a bright side 10 minutes is much nicer test times than 30 minutes.
Bug in C-Phrase?
Monday 2009-11-16 08:26
Seems like my C-Phrase problem is some sort of bug in C-Phrase were certain input strings causes the system to get stuck in long loops before crashing. I will have to discuss with mike how to proceed, hopefully he can confirm the bug.