| |||||||||||||||||||||
| |||||||||||||||||||||
| |||||||||||||||||||||
| |||||||||||||||||||||
Initial Experiment Results
All experiments here were done with the first 500 instances
from the training data set
and tested with the first 500 instances from the test data set.
This was done to get a feel for the optimal k value to do
further testing with. Since time is limited, no fancy indexing
scheme was used. This makes the search through all the training
instances very lengthy. Once the optimal k is found, tests will be done
with the full training and testing data sets.
For the details of each experiment, please see the
test.txt file for that test.
The best results from this set of testing was from the k = 6 test. Actually, k = 7 got the same result.
Summary of these results:
| Test | Resulting accuracy over test data (%) |
|---|---|
| k = 1 | 77.4451 |
| k = 2 | 78.8423 |
| k = 3 | 79.2415 |
| k = 4 | 81.8363 |
| k = 5 | 82.4351 |
| k = 6 | 82.6347 |
| k = 7 | 82.6347 |
| k = 8 | 82.0359 |
| k = 9 | 82.4351 |
| k = 10 | 81.0379 |
So, further testing will be done with k = 6. This saves a miniscule but measurable time over k = 7.
by: Keith A. Pray
Last Modified: July 4, 2004 9:00 AM
|
© 2004 - 1975 Keith A. Pray.
All rights reserved. |