use candidates from cosine similarity only (data)

config

@@ -1,6 +1,6 @@
 [DEFAULT]
-timestamp = 20180611-110130
 rootpath = /mnt/storage/karr/users/furrer/prlnk
+timestamp = 20180611-112025
 workers = 10
 
 [general]
@@ -15,8 +15,7 @@ summary_fn = runs/summaries/${timestamp}.txt
 prediction_fn = runs/predictions/${timestamp}.tsv
 
 [candidates]
-generator = SGramCosine(.5, 10, [(2, 1), (3, 1)])
-	PhraseVecFixedSet(10, "mean")
+generator = SGramCosine(.5, 20, [(2, 1), (3, 1)])
 oracle = none
 workers = 0
 

log

@@ -1,23 +1,26 @@
-2018-06-11 11:01:33,835 - 'pattern' package not found; tag filters are not available for English
-2018-06-11 11:01:33,845 - loading pretrained embeddings...
-2018-06-11 11:01:33,845 - loading projection weights from /mnt/storage/karr/users/furrer/prlnk/data/embeddings/wvec_50_haodi-li-et-al.bin
-2018-06-11 11:01:39,387 - loaded (309058, 50) matrix from /mnt/storage/karr/users/furrer/prlnk/data/embeddings/wvec_50_haodi-li-et-al.bin
-2018-06-11 11:01:39,729 - loading terminology...
-2018-06-11 11:01:39,958 - loading vectorizer...
-2018-06-11 11:01:39,958 - loading candidate generator...
-2018-06-11 11:01:56,543 - preprocessing validation data...
-2018-06-11 11:01:56,544 - loading corpus...
-2018-06-11 11:01:56,554 - generating candidates with 0 workers...
-2018-06-11 11:02:11,068 - generated 6180 pair-wise samples (13250 with duplicates)
-2018-06-11 11:02:11,068 - compiling model architecture...
-2018-06-11 11:02:11,570 - preprocessing training data...
-2018-06-11 11:02:11,570 - loading corpus...
-2018-06-11 11:02:11,633 - generating candidates with 0 workers...
-2018-06-11 11:03:19,915 - generated 28647 pair-wise samples (84693 with duplicates)
-2018-06-11 11:03:19,917 - training CNN...
-2018-06-11 11:03:43,450 - Ranking accuracy: 0.707751
-2018-06-11 11:04:06,082 - Ranking accuracy: 0.74587
-2018-06-11 11:04:28,678 - Ranking accuracy: 0.747141
-2018-06-11 11:04:51,120 - Ranking accuracy: 0.743329
-2018-06-11 11:04:51,121 - Epoch 00004: early stopping
-2018-06-11 11:04:51,121 - done training.
+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.

results

@@ -1,7 +1,7 @@
-accuracy     0.7433290978398983
-correct        585
+accuracy     0.770012706480305
+correct        606
 total          787
-unreachable    108
-nocandidates     0
+unreachable    110
+nocandidates    10
 ambiguous        2
 compound        10