fix evaluation mistake: "|"-separated reference IDs are composite, no…

…t alternatives (data)

config

@@ -1,6 +1,6 @@
 [DEFAULT]
+timestamp = 20180613-092832
 rootpath = /mnt/storage/karr/users/furrer/prlnk
-timestamp = 20180611-112025
 workers = 10
 
 [general]

log

@@ -1,26 +1,11 @@
-2018-06-11 11:20:28,792 - 'pattern' package not found; tag filters are not available for English
-2018-06-11 11:20:28,799 - loading pretrained embeddings...
-2018-06-11 11:20:28,799 - loading projection weights from /mnt/storage/karr/users/furrer/prlnk/data/embeddings/wvec_50_haodi-li-et-al.bin
-2018-06-11 11:20:34,286 - loaded (309058, 50) matrix from /mnt/storage/karr/users/furrer/prlnk/data/embeddings/wvec_50_haodi-li-et-al.bin
-2018-06-11 11:20:34,628 - loading terminology...
-2018-06-11 11:20:34,845 - loading vectorizer...
-2018-06-11 11:20:34,845 - loading candidate generator...
-2018-06-11 11:20:46,545 - preprocessing validation data...
-2018-06-11 11:20:46,545 - loading corpus...
-2018-06-11 11:20:46,557 - generating candidates with 0 workers...
-2018-06-11 11:21:00,534 - generated 5651 pair-wise samples (11585 with duplicates)
-2018-06-11 11:21:00,534 - compiling model architecture...
-2018-06-11 11:21:01,026 - preprocessing training data...
-2018-06-11 11:21:01,027 - loading corpus...
-2018-06-11 11:21:01,091 - generating candidates with 0 workers...
-2018-06-11 11:22:07,034 - generated 26228 pair-wise samples (71125 with duplicates)
-2018-06-11 11:22:07,036 - training CNN...
-2018-06-11 11:22:28,634 - Ranking accuracy: 0.715375
-2018-06-11 11:22:49,605 - Ranking accuracy: 0.743329
-2018-06-11 11:23:10,267 - Ranking accuracy: 0.752224
-2018-06-11 11:23:30,968 - Ranking accuracy: 0.753494
-2018-06-11 11:23:51,620 - Ranking accuracy: 0.767471
-2018-06-11 11:24:12,168 - Ranking accuracy: 0.766201
-2018-06-11 11:24:32,784 - Ranking accuracy: 0.770013
-2018-06-11 11:24:32,784 - Epoch 00007: early stopping
-2018-06-11 11:24:32,784 - done training.
+2018-06-13 09:28:35,734 - 'pattern' package not found; tag filters are not available for English
+2018-06-13 09:28:35,747 - loading pretrained embeddings...
+2018-06-13 09:28:35,747 - loading projection weights from /mnt/storage/karr/users/furrer/prlnk/data/embeddings/wvec_50_haodi-li-et-al.bin
+2018-06-13 09:28:41,258 - loaded (309058, 50) matrix from /mnt/storage/karr/users/furrer/prlnk/data/embeddings/wvec_50_haodi-li-et-al.bin
+2018-06-13 09:28:41,608 - loading terminology...
+2018-06-13 09:28:41,835 - loading vectorizer...
+2018-06-13 09:28:41,835 - loading candidate generator...
+2018-06-13 09:28:53,597 - preprocessing validation data...
+2018-06-13 09:28:53,597 - loading corpus...
+2018-06-13 09:28:53,617 - generating candidates with 0 workers...
+2018-06-13 09:28:58,012 - generated 11585 pair-wise samples (11585 with duplicates)

results

@@ -1,7 +1,6 @@
-accuracy     0.770012706480305
-correct        606
+accuracy     0.7560355781448539
+correct        595
 total          787
-unreachable    110
+unreachable    129
 nocandidates    10
 ambiguous        2
-compound        10