\ft\tiResult of Applying Noisy Harmonic Grammar to NeutralPattern.txt

\cn2-19-2017, 1:05 p.m.

\cnOTSoft 2.5, release date 1/17/2017

\hrNHG Results for NeutralPattern.txt	2-19-2017	\pn
\ks

\h1Weights Found
\ts2
Constraints	Weights Found
Constraint 1	0.001
Constraint 2	0.000
\te\ke

\ks

\h1Matchup to Input Frequencies
\ts5
/Input/	Input Frequencies	Generated Frequencies	Input Number (exact)	Generated Number
Candidate 1	0.500	0.501	50088	50137
Candidate 2	0.500	0.499	49912	49863
\te\ke
\ts3\ks
/Input/	Constraint 1 (.001)	Constraint 2 (0)
\wsF\we Candidate 1 (0.501)		*
Candidate 2 (0.499)	* 	*
\te\ke
\ks

\h1Hasse Diagram
The following Hasse diagram (in output folder at NeutralPatternHasse.gif) summarizes the rankings obtained.

\hf
\ke


\h1Testing the Grammar:  Details
The grammar was tested for 100000 cycles.
Average error per candidate:  0.000 percent
Learning time:  0.013 minutes


\h1Parameter Values Used

   Noise was set at 1.

   Negative weights were permitted.
   The number of times that a negative weight was recorded during learning and testing was 223796 out of 450180 chances.

   The number of ties between winner and rival was  1 out of 400000 chances.  Tie is within 0.00001.

   Noise was applied to constraints, hence the same for all candidates.

   Noise was applied before weights multiplied by violations.
\h2Initial weights

All constraints started out at the default value of 0.
\ke
\ks
\h2Schedule for GLA Parameters
\ts6
Stage	Trials	PlastMark	PlastFaith
1	25000	1.000	1.000
2	25000	0.100	0.100
3	25000	0.010	0.010
4	25000	0.001	0.001
\te
There were a total of 100000 learning trials.
\ke
