Switching to a fractional split between test and training data (#30)

mtlynch · May 1, 2018 · d3d8c20 · d3d8c20
1 parent f7d196a
commit d3d8c20
Show file tree

Hide file tree

Showing 4 changed files with 1,375 additions and 1,379 deletions.
diff --git a/test_e2e b/test_e2e
@@ -11,8 +11,10 @@ set -e
 set -x
 
 LABELLED_DATA_FILE=nyt-ingredients-snapshot-2015.csv
-COUNT_TRAIN=20000
-COUNT_TEST=2000
+LABELLED_EXAMPLE_COUNT=22000
+TRAINING_DATA_PERCENT=0.9
+COUNT_TRAIN=$(python -c "print int($TRAINING_DATA_PERCENT * $LABELLED_EXAMPLE_COUNT)")
+COUNT_TEST=$(python -c "print int((1.0 - $TRAINING_DATA_PERCENT) * $LABELLED_EXAMPLE_COUNT)")
 OUTPUT_DIR=$(mktemp -d)
 CRF_TRAINING_FILE="${OUTPUT_DIR}/training_data.crf"
 CRF_TESTING_FILE="${OUTPUT_DIR}/testing_data.crf"

diff --git a/tests/golden/eval_output b/tests/golden/eval_output
@@ -1,10 +1,10 @@
 
 Sentence-Level Stats:
-	correct:  1854
-	total:  2000
-	% correct:  92.7
+	correct:  2044
+	total:  2199
+	% correct:  92.9513415189
 
 Word-Level Stats:
-	correct: 11252
-	total: 11459
-	% correct: 98.1935596474
+	correct: 12328
+	total: 12547
+	% correct: 98.2545628437