2025-04-27 07:49:33 -04:00

73 lines
1.8 KiB
INI

# Configuration file for Traditional Chinese WinCE recognizer
# Configuration
Locale=CHT
Config=dev_cjk
# The list of catagories to select from the .loc file for this configuration.
CatagoryList=basic,ExtPunc,FilledShapes,CJK
# The file containing the list of FFF file to train from.
TrainSets="\
//asian/island/TextInk/cht/cht-boxed-20020222/big5--list.txt \
//asian/island/TextInk/cht/cht-boxed-20020222/capital--list.txt \
//asian/island/TextInk/cht/cht-boxed-20020222/natural--list.txt \
//asian/island/TextInk/cht/cht-boxed-20020222/cjk--list.txt \
//asian/island/TextInk/cht/cht-boxed-20020222/addpanel--list.txt \
"
# Free input training set
FreeTrainSet=""
# Tuning sets
NatTuneSet="//tpg/reco/test/cht/tune-natural000.tfl"
UniTuneSet="//tpg/reco/test/cht/tune-uniform000.tfl"
FreeTuneSet=$NatTuneSet
# Unicode text file to generate unigrams and bigrams from.
TextFile="//tpg/reco/corpora/cht/all02.utf"
# List of tests to run on recognizer.
TestList="natural address Big5 CJK free"
# Target size to shrink zilla to.
ZillaSize=7000000
# Select the recognizer for 1 and 2 stroke characters
UseFugu=0
UseOtter=1
UseSole=0
UseJaws=0
# Select the postprocessor
UseCrane=1
UseHawk=0
# Shrink parameters for otter.
OtterStrongCluster=30
OtterWeakCluster=10
OtterSize=1000000
# Otter limit on number of samples per (folded) code point
OtterSampleLimit=25000
# Number of classes to use for Centipede
CentipedeClasses=32
# Weight of natural frequency vs. uniform frequency in training set.
NatuFreqWeight=1.0
# Flag for WinCE configuration.
# Set to 1 if true, 0 if false.
IsWinCE=0
# Character detector settings
UseCharDet=0
# Break net and segmentation net settings
UseBrkNet=0
UseSegNet=0
# Whether to do IFELang3 tuning and testing
UseIFELang3=0