summaryrefslogtreecommitdiff
path: root/graehl/NOTES
diff options
context:
space:
mode:
authorgraehl@gmail.com <graehl@gmail.com@ec762483-ff6d-05da-a07a-a48fb63a330f>2010-08-02 16:43:59 +0000
committergraehl@gmail.com <graehl@gmail.com@ec762483-ff6d-05da-a07a-a48fb63a330f>2010-08-02 16:43:59 +0000
commit2fd80bbceadd625b74f8cbd989c945ce24a60fcc (patch)
tree9fa78b3cf6595bd6d95d9576a58c6a0f2e440a6a /graehl/NOTES
parent506cdc7562956b8bd2460f7dd55a307775eb68cb (diff)
notes
git-svn-id: https://ws10smt.googlecode.com/svn/trunk@469 ec762483-ff6d-05da-a07a-a48fb63a330f
Diffstat (limited to 'graehl/NOTES')
-rwxr-xr-xgraehl/NOTES18
1 files changed, 18 insertions, 0 deletions
diff --git a/graehl/NOTES b/graehl/NOTES
new file mode 100755
index 00000000..77e99fee
--- /dev/null
+++ b/graehl/NOTES
@@ -0,0 +1,18 @@
+BUG: tune is bad - urdu conf=baseline tuning (16 dev bleu score???)
+
+conf=baseline force=1 ./tune.sh
+
+ decode is good.
+
+ UPDATE: maybe tuning is fine; chris never gave me a dev-corpus-filtered grammar and so a bleu of 16 may be what we always got; i just never checked. this means i need to redo tuned-first-pass experiments
+
+valgrind is ok
+
+ dist-vest?
+
+ (changes made to scoring? plusequals? shared_ptr? small_vector?)
+
+ scorer_test is good
+
+
+ line_optimer fast_score scorer