Constraint Grammar/Optimisation - Revision history

Unhammer: /* Current Performance */

2014-09-18T11:31:58Z

Current Performance

2014-09-18T11:31:04Z

Current Performance

2014-09-18T11:29:26Z

Current Performance: update :-)

2011-04-05T17:50:37Z

Performance as of 2011-04-05

2011-02-10T14:14:33Z

Rationale

2011-02-07T21:39:24Z

Current Performance

2011-02-07T18:59:00Z

Rationale

2011-02-07T18:25:06Z

2011-02-07T18:18:44Z

2011-02-07T18:12:19Z

@@ Line 6: / Line 6: @@
 === Current Performance ===
 For the [http://beta.visl.sdu.dk/cg3_performance.html reference test set]:
-* As of 2014 (r10044) [http://beta.visl.sdu.dk/cg3_performance.html CG-3 performance] is at 378 cohorts per second. Total run time 2.08 seconds.
+* As of 2014-07-14 (r10044) [http://beta.visl.sdu.dk/cg3_performance.html CG-3 performance] is at 378 cohorts per second. Total run time 2.08 seconds.
 * vislcg is at 32 cohorts per second. Total run time 24.63 seconds.
 * CG-2 cannot handle the grammar size.

@@ Line 11: / Line 11: @@
 For all 192042 lines of input:
-* CG-3 takes 231.97s
+* CG-3 (r6898) takes 231.97s
 * vislcg takes 2521.59, or 980% slower than CG-3
 * CG-2 cannot handle the grammar size.
 If the grammar is split into smaller morf and syn parts:
-* CG-3: morf 53.30s + syn 56.54s
+* CG-3 (r6898): morf 53.30s + syn 56.54s
 * vislcg: morf 85.15s + syn 98.69s, or 60-70% slower than CG-3
 * CG-2: morf 6.41s + syn 7.01s, or 80-90% faster than CG-3

@@ Line 6: / Line 6: @@
 === Current Performance ===
 For the [http://beta.visl.sdu.dk/cg3_performance.html reference test set]:
-* As of 2011-04-05, [http://beta.visl.sdu.dk/cg3_performance.html CG-3 performance] is at 282 cohorts per second. Total run time 2.79 seconds.
+* As of 2014 (r10044) [http://beta.visl.sdu.dk/cg3_performance.html CG-3 performance] is at 378 cohorts per second. Total run time 2.08 seconds.
 * vislcg is at 32 cohorts per second. Total run time 24.63 seconds.
 * CG-2 cannot handle the grammar size.

@@ Line 6: / Line 6: @@
 === Current Performance ===
 For the [http://beta.visl.sdu.dk/cg3_performance.html reference test set]:
-* As of 2011-02-07, [http://beta.visl.sdu.dk/cg3_performance.html CG-3 performance] is at 203 cohorts per second. Total run time 3.88 seconds.
+* As of 2011-04-05, [http://beta.visl.sdu.dk/cg3_performance.html CG-3 performance] is at 282 cohorts per second. Total run time 2.79 seconds.
 * vislcg is at 32 cohorts per second. Total run time 24.63 seconds.
 * CG-2 cannot handle the grammar size.
 For all 192042 lines of input:
-* CG-3 takes 325.93s
+* CG-3 takes 231.97s
-* vislcg takes 2521.59, or 670% slower than CG-3
+* vislcg takes 2521.59, or 980% slower than CG-3
 * CG-2 cannot handle the grammar size.
 If the grammar is split into smaller morf and syn parts:
-* CG-3: morf 76.91s + syn 79.60s
+* CG-3: morf 53.30s + syn 56.54s
-* vislcg: morf 85.15s + syn 98.69s, or 10-20% slower than CG-3
+* vislcg: morf 85.15s + syn 98.69s, or 60-70% slower than CG-3
-* CG-2: morf 6.41s + syn 7.01s, or 92% faster than CG-3
+* CG-2: morf 6.41s + syn 7.01s, or 80-90% faster than CG-3
 == Algorithm Pseudocode ==

@@ Line 2: / Line 2: @@
 == Rationale ==
-VISL CG-3 is too slow for some users. It is clearly faster than the old [http://visl.sdu.dk/visl/vislcg-doc.html vislcg], but it is also clearly slower than [http://google.com/search?q=Pasi+Tapanainen+CG-2 Pasi Tapanainen's CG-2] (up to 4 times slower for compatible features).
+VISL CG-3 is too slow for some users. It is clearly faster than the old [http://visl.sdu.dk/visl/vislcg-doc.html vislcg], but it is also clearly slower than [http://google.com/search?q=Pasi+Tapanainen+CG-2 Pasi Tapanainen's CG-2] (up to 10 times slower for compatible features).
 === Current Performance ===

← Older revision	Revision as of 18:25, 7 February 2011
(No difference)

@@ Line 35: / Line 35: @@
 }
 </pre>
+* Sections are re-run if any rule actually did anything, meaning worst case runtime of O(num_rules * factorial(num_sections)). On average, this happens twice so expected runtime is Θ(num_rules * num_sections * 2).
 * Rule order is fully deterministic.

← Older revision		Revision as of 18:12, 7 February 2011
Line 1:		Line 1:
	Ideas and discussions on how to optimize [http://beta.visl.sdu.dk/cg3.html VISL CG-3].		Ideas and discussions on how to optimize [http://beta.visl.sdu.dk/cg3.html VISL CG-3]. When it says "I", that means [[User:Tino Didriksen\|Tino Didriksen]].

	== Rationale ==		== Rationale ==