nyancrimew/LatinIME
1
0
Fork 0
Code Issues Pull requests Projects Releases Wiki Activity

Optimize the parameters for the typing algorithm

Browse source 
The new set of parameters are optimized for both
keystroke savings and regression test scores.

This change is corresponding to the patch 6 of
Change-Id: Ib8439a85de26cbbba58837344717f0acbfcf0ab0
in the regression test repository.

Here is the link to the doc analyzing the
“nnow → know” case:
https://docs.google.com/a/google.com/document/d/1XIHfAHLlP1f3F8R-u7yxPGWcP58tokygzN_lQKXFFgM/edit

Here is the link to the spreadsheet showing
the diff of the words with frequencies >= 10
in en_user_log_phones_2011_08.csv.
https://docs.google.com/a/google.com/spreadsheet/ccc?key=0Am_RQlV4zC6SdFFVTlhJT0RLUENCWWZOZ3JQTUpMQ2c&usp=drive_web#gid=0

Updated TRT Results for this change:
[Category diff]
+1     783
-1     717
+2      20
-2      34
+3      34
-3      20
+4     575
-4     777
+5     765
-5     755
+6     785
-6     666
+7     479
-7     472

[Weighted category diff]
+1     999
-1     965
+2      21
-2      35
+3      35
-3      21
+4     804
-4     980
+5     970
-5     971
+6     954
-6     831
+7     645
-7     625

Bug: 10613502

Change-Id: I5cd447673a196ad5af641a2118602bec879bdefb
This commit is contained in:
Xiaojun Bi 2013-10-03 18:15:29 -07:00
parent 9d76391058
commit 2ed09c214a
1 changed files with 24 additions and 24 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

48
native/jni/src/suggest/policyimpl/typing/scoring_params.cpp
View file

@ -27,30 +27,30 @@ const int ScoringParams::MAX_CACHE_DIC_NODE_SIZE = 170;
const int ScoringParams::MAX_CACHE_DIC_NODE_SIZE_FOR_SINGLE_POINT = 310;
const int ScoringParams::THRESHOLD_SHORT_WORD_LENGTH = 4;
const float ScoringParams::DISTANCE_WEIGHT_LENGTH = 0.132f;
const float ScoringParams::PROXIMITY_COST = 0.095f;
const float ScoringParams::FIRST_CHAR_PROXIMITY_COST = 0.102f;
const float ScoringParams::FIRST_PROXIMITY_COST = 0.019f;
const float ScoringParams::OMISSION_COST = 0.458f;
const float ScoringParams::OMISSION_COST_SAME_CHAR = 0.491f;
const float ScoringParams::OMISSION_COST_FIRST_CHAR = 0.582f;
const float ScoringParams::INSERTION_COST = 0.730f;
const float ScoringParams::TERMINAL_INSERTION_COST = 0.93f;
const float ScoringParams::INSERTION_COST_SAME_CHAR = 0.586f;
const float ScoringParams::INSERTION_COST_PROXIMITY_CHAR = 0.70f;
const float ScoringParams::INSERTION_COST_FIRST_CHAR = 0.623f;
const float ScoringParams::TRANSPOSITION_COST = 0.526f;
const float ScoringParams::SPACE_SUBSTITUTION_COST = 0.319f;
const float ScoringParams::ADDITIONAL_PROXIMITY_COST = 0.380f;
const float ScoringParams::SUBSTITUTION_COST = 0.383f;
const float ScoringParams::COST_NEW_WORD = 0.042f;
const float ScoringParams::COST_SECOND_OR_LATER_WORD_FIRST_CHAR_UPPERCASE = 0.25f;
const float ScoringParams::DISTANCE_WEIGHT_LANGUAGE = 1.123f;
const float ScoringParams::COST_FIRST_LOOKAHEAD = 0.545f;
const float ScoringParams::COST_LOOKAHEAD = 0.073f;
const float ScoringParams::HAS_PROXIMITY_TERMINAL_COST = 0.093f;
const float ScoringParams::HAS_EDIT_CORRECTION_TERMINAL_COST = 0.041f;
const float ScoringParams::HAS_MULTI_WORD_TERMINAL_COST = 0.447f;
const float ScoringParams::DISTANCE_WEIGHT_LENGTH = 0.1524f;
const float ScoringParams::PROXIMITY_COST = 0.0694f;
const float ScoringParams::FIRST_CHAR_PROXIMITY_COST = 0.072f;
const float ScoringParams::FIRST_PROXIMITY_COST = 0.07788f;
const float ScoringParams::OMISSION_COST = 0.4676f;
const float ScoringParams::OMISSION_COST_SAME_CHAR = 0.399f;
const float ScoringParams::OMISSION_COST_FIRST_CHAR = 0.5256f;
const float ScoringParams::INSERTION_COST = 0.7248f;
const float ScoringParams::TERMINAL_INSERTION_COST = 0.9828f;
const float ScoringParams::INSERTION_COST_SAME_CHAR = 0.5508f;
const float ScoringParams::INSERTION_COST_PROXIMITY_CHAR = 0.674f;
const float ScoringParams::INSERTION_COST_FIRST_CHAR = 0.639f;
const float ScoringParams::TRANSPOSITION_COST = 0.5608f;
const float ScoringParams::SPACE_SUBSTITUTION_COST = 0.339f;
const float ScoringParams::ADDITIONAL_PROXIMITY_COST = 0.4576f;
const float ScoringParams::SUBSTITUTION_COST = 0.3806f;
const float ScoringParams::COST_NEW_WORD = 0.0292f;
const float ScoringParams::COST_SECOND_OR_LATER_WORD_FIRST_CHAR_UPPERCASE = 0.3224f;
const float ScoringParams::DISTANCE_WEIGHT_LANGUAGE = 1.1214f;
const float ScoringParams::COST_FIRST_LOOKAHEAD = 0.4786f;
const float ScoringParams::COST_LOOKAHEAD = 0.00624f;
const float ScoringParams::HAS_PROXIMITY_TERMINAL_COST = 0.06836f;
const float ScoringParams::HAS_EDIT_CORRECTION_TERMINAL_COST = 0.0362f;
const float ScoringParams::HAS_MULTI_WORD_TERMINAL_COST = 0.4182f;
const float ScoringParams::TYPING_BASE_OUTPUT_SCORE = 1.0f;
const float ScoringParams::TYPING_MAX_OUTPUT_SCORE_PER_INPUT = 0.1f;
const float ScoringParams::NORMALIZED_SPATIAL_DISTANCE_THRESHOLD_FOR_EDIT = 0.045f;