Re: [裝死] 很不幸的要麻煩大家了
※ 引述《hamigwa (哈密瓜)》之銘言：
※ 引述《JLTChiu (隨風)》之銘言：
: 最新資料登場
: /share/data/cp/ntu_lectures/ss/master
/home/hamigwa/Projects/ss/sets/alltrans/master_revised.txt
: /share/data/cp/ntu_lectures/ss/lex/lexicon
: /share/data/cp/ntu_lectures/ss/wav_files/whole/*.wav
: /diska1/home/RA/SSS_Data/Slide
: 有問題請再與我聯絡 謝謝＝）
我稍微根據/share/data/cp/ntu_lectures/ss/sets/中的scp檔
弄出了
/home/hamigwa/Projects/ss/sets/train.trans
/home/hamigwa/Projects/ss/sets/dev.trans
/home/hamigwa/Projects/ss/sets/test.trans
/home/hamigwa/Projects/ss/sets/train.seg
/home/hamigwa/Projects/ss/sets/dev.seg
/home/hamigwa/Projects/ss/sets/test.seg
//然後目前/share/data/cp/ntu_lectures/ss/sets/*.(seg/trans) 當還是舊的
LM:
看了train.seg，然後有KN和GT smoothing
/home/hamigwa/Projects/ss/KN-Ngram/LMs/kn-1gram.lm.ascii ppl=513.599
/home/hamigwa/Projects/ss/KN-Ngram/LMs/kn-2gram.lm.ascii ppl=222.716
/home/hamigwa/Projects/ss/KN-Ngram/LMs/kn-3gram.lm.ascii ppl=211.146
/home/hamigwa/Projects/ss/KN-Ngram/LMs/kn-4gram.lm.ascii ppl=209.654
看了train.seg + dev.seg，然後有KN和GT smoothing
/home/hamigwa/Projects/ss/KN-Ngram-Trn+Dev/LMs/kn-1gram-TD.lm.ascii p=503.306
/home/hamigwa/Projects/ss/KN-Ngram-Trn+Dev/LMs/kn-2gram-TD.lm.ascii p=206.31
/home/hamigwa/Projects/ss/KN-Ngram-Trn+Dev/LMs/kn-3gram-TD.lm.ascii p=193.816
/home/hamigwa/Projects/ss/KN-Ngram-Trn+Dev/LMs/kn-4gram-TD.lm.ascii p=192.465
用train.seg訓練DTs 再用dev.seg調DTs的random forest(10000 DTs)
/home/hamigwa/Projects/ss/rflm/trees/rf-ss.lm.ascii ppl=178.302
還沒有用out-of-domain corpora去做LM adaptaiton
請大家先將就著用一下唄:P
--
※ 發信站: 批踢踢實業坊(ptt.cc)
◆ From: 140.112.21.11
討論串 (同標題文章)