add seq2seq test

Author: nbcsm

tests/test_seq2seq.py

@@ -0,0 +1,70 @@
+# Copyright (c) Microsoft Corporation. All rights reserved.
+# Licensed under the MIT license.
+
+""" Unit Tests for tf.contrib.seq2seq """
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import numpy as np
+import tensorflow as tf
+from tensorflow.contrib import rnn
+from tensorflow.python.ops import init_ops
+from backend_test_base import Tf2OnnxBackendTestBase
+from common import check_opset_min_version, unittest_main
+
+
+# pylint: disable=missing-docstring
+
+class Seq2SeqTests(Tf2OnnxBackendTestBase):
+    def test_dynamic_decode_maximum_iterations(self):
+        batch_size = 2
+        num_units = 4
+        vocab_size = 5
+        embedding_size = 3
+        GO_SYMBOL = 0
+        END_SYMBOL = 1
+
+        embedding = tf.constant(np.ones([vocab_size, embedding_size], dtype=np.float32))
+        state_val = np.reshape([np.ones([num_units], dtype=np.float32) * i for i in range(batch_size)],
+                               [batch_size, num_units])
+        encoder_state = tf.nn.rnn_cell.LSTMStateTuple(state_val, state_val)
+        initializer = init_ops.constant_initializer(0.5)
+        cell = rnn.LSTMCell(
+            num_units=num_units,
+            initializer=initializer,
+            state_is_tuple=True)
+
+        helper = tf.contrib.seq2seq.GreedyEmbeddingHelper(
+            embedding=embedding,
+            start_tokens=tf.tile([GO_SYMBOL], [batch_size]),
+            end_token=END_SYMBOL)
+
+        output_layer = tf.layers.Dense(vocab_size, kernel_initializer=initializer)
+        decoder = tf.contrib.seq2seq.BasicDecoder(
+            cell=cell,
+            helper=helper,
+            initial_state=encoder_state,
+            output_layer=output_layer)
+
+        outputs, state, sequence_lengths = tf.contrib.seq2seq.dynamic_decode(
+            decoder=decoder,
+            maximum_iterations=6)
+
+        _ = tf.identity(outputs.rnn_output, name="rnn_output")
+        _ = tf.identity(outputs.sample_id, name="sample_id")
+        _ = tf.identity(state, name="state")
+        _ = tf.identity(sequence_lengths, name="sequence_lengths")
+
+        output_names_with_port = [
+            "rnn_output:0",
+            # "sample_id:0",  # incomplete type support for Transpose on onnxruntime 0.2.1
+            "state:0",
+        ]
+
+        self.run_test_case({}, [], output_names_with_port, atol=1e-06, rtol=1e-6)
+
+
+if __name__ == '__main__':
+    unittest_main()