commit 1870d00b659eeeb373e7f6bc83f815f8622f9cc1
author Jörn Kottmann <joern@apache.org> Fri Feb 01 10:45:58 2019 +0100
committer Joern Kottmann <kottmann@users.noreply.github.com> Mon Feb 04 16:12:12 2019 +0100
tree 471ffa8cd60f1db07ff0860fbc5e61116c495277
parent 7b6e85d1cc25d805a045fb6c18ff4ee92619f04a

Compute max target length instead of fixed value in normalizer

tf-ner-poc/src/main/python/normalizer/normalizer.py

diff --git a/tf-ner-poc/src/main/python/normalizer/normalizer.py b/tf-ner-poc/src/main/python/normalizer/normalizer.py
index bc2c6ac..a0eabe8 100644
--- a/tf-ner-poc/src/main/python/normalizer/normalizer.py
+++ b/tf-ner-poc/src/main/python/normalizer/normalizer.py
@@ -219,8 +219,12 @@
 
     batch_size = 20
 
-    # TODO: Don't hard code this ...
-    target_max_len = 9
+    target_max_len = -1
+    for token in (target_train + target_dev + target_test):
+        target_max_len = max(target_max_len, len(token))
+
+    # Increase size by one for termination char
+    target_max_len += 1
 
     train_graph = tf.Graph()
     eval_graph = tf.Graph()