Make batch size for normalizer inference dynamic
diff --git a/tf-ner-poc/src/main/java/org/apache/opennlp/normalizer/Normalizer.java b/tf-ner-poc/src/main/java/org/apache/opennlp/normalizer/Normalizer.java
index 2ad4809..52d44cf 100644
--- a/tf-ner-poc/src/main/java/org/apache/opennlp/normalizer/Normalizer.java
+++ b/tf-ner-poc/src/main/java/org/apache/opennlp/normalizer/Normalizer.java
@@ -73,8 +73,6 @@
public String[] normalize(String[] texts) {
- // TODO: Batch size is hard coded in the graph, make it dynamic or at padding here
-
int textLengths[] = Arrays.stream(texts).mapToInt(String::length).toArray();
int maxLength = Arrays.stream(textLengths).max().getAsInt();
@@ -89,11 +87,13 @@
}
try (Tensor<?> charTensor = Tensor.create(charIds);
- Tensor<?> textLength = Tensor.create(textLengths)) {
+ Tensor<?> textLength = Tensor.create(textLengths);
+ Tensor<?> batchSize = Tensor.create(texts.length)) {
List<Tensor<?>> result = session.runner()
.feed("encoder_char_ids", charTensor)
.feed("encoder_lengths", textLength)
+ .feed("batch_size", batchSize)
.fetch("decode", 0).run();
try (Tensor<?> translationTensor = result.get(0)) {
diff --git a/tf-ner-poc/src/main/python/normalizer/normalizer.py b/tf-ner-poc/src/main/python/normalizer/normalizer.py
index 86e735e..f721bb0 100644
--- a/tf-ner-poc/src/main/python/normalizer/normalizer.py
+++ b/tf-ner-poc/src/main/python/normalizer/normalizer.py
@@ -78,9 +78,11 @@
encoder_char_dim = 100
num_units = 256
+ batch_size_ph = tf.placeholder_with_default(batch_size, shape=(), name="batch_size")
+
# Encoder
- encoder_char_ids_ph = tf.placeholder(tf.int32, shape=[batch_size, None], name="encoder_char_ids")
- encoder_lengths_ph = tf.placeholder(tf.int32, shape=[batch_size], name="encoder_lengths")
+ encoder_char_ids_ph = tf.placeholder(tf.int32, shape=[None, None], name="encoder_char_ids")
+ encoder_lengths_ph = tf.placeholder(tf.int32, shape=[None], name="encoder_lengths")
encoder_embedding_weights = tf.get_variable(name="char_embeddings", dtype=tf.float32,
shape=[encoder_nchars, encoder_char_dim])
@@ -90,7 +92,7 @@
encoder_emb_inp = tf.transpose(encoder_emb_inp, perm=[1, 0, 2])
encoder_cell = tf.nn.rnn_cell.BasicLSTMCell(num_units)
- initial_state = encoder_cell.zero_state(batch_size, dtype=tf.float32)
+ initial_state = encoder_cell.zero_state(batch_size_ph, dtype=tf.float32)
encoder_outputs, encoder_state = tf.nn.dynamic_rnn(
encoder_cell, encoder_emb_inp, initial_state=initial_state,
@@ -98,8 +100,8 @@
time_major=True, swap_memory=True)
# Decoder
- decoder_char_ids_ph = tf.placeholder(tf.int32, shape=[batch_size, None], name="decoder_char_ids")
- decoder_lengths = tf.placeholder(tf.int32, shape=[batch_size], name="decoder_lengths")
+ decoder_char_ids_ph = tf.placeholder(tf.int32, shape=[None, None], name="decoder_char_ids")
+ decoder_lengths = tf.placeholder(tf.int32, shape=[None], name="decoder_lengths")
# decoder output (decoder_input shifted to the left by one)
@@ -121,7 +123,7 @@
attention_layer_size=num_units)
# decoder_initial_state = encoder_state
- decoder_initial_state = decoder_cell.zero_state(dtype=tf.float32, batch_size=batch_size)
+ decoder_initial_state = decoder_cell.zero_state(dtype=tf.float32, batch_size=batch_size_ph)
if "TRAIN" == mode:
@@ -166,16 +168,14 @@
if "EVAL" == mode:
helperE = tf.contrib.seq2seq.GreedyEmbeddingHelper(
decoder_embedding_weights,
- tf.fill([batch_size], decoder_nchars-2), decoder_nchars-1)
+ tf.fill([batch_size_ph], decoder_nchars-2), decoder_nchars-1)
decoderE = tf.contrib.seq2seq.BasicDecoder(
decoder_cell, helperE, decoder_initial_state,
output_layer=projection_layer)
- outputs, _, _ = tf.contrib.seq2seq.dynamic_decode(decoderE, maximum_iterations=15)
-
+ outputs, _, _ = tf.contrib.seq2seq.dynamic_decode(decoderE, maximum_iterations=20)
translations = tf.identity(outputs.sample_id, name="decode")
- # the outputs don't decode anything ...
return encoder_char_ids_ph, encoder_lengths_ph, translations
def encode_chars(names):
