added optimization config to examples/seq2seq_attn

asyml · Oct 9, 2018 · d223165 · d223165
1 parent a73d727
commit d223165
Show file tree

Hide file tree

Showing 3 changed files with 28 additions and 1 deletion.
diff --git a/examples/seq2seq_attn/config_model.py b/examples/seq2seq_attn/config_model.py
@@ -27,3 +27,11 @@
         'attention_layer_size': num_units
     }
 }
+opt = {
+    'optimizer': {
+        'type':  'AdamOptimizer',
+        'kwargs': {
+            'learning_rate': 0.001,
+        },
+    },
+}
diff --git a/examples/seq2seq_attn/config_model_full.py b/examples/seq2seq_attn/config_model_full.py
@@ -125,3 +125,21 @@
     'max_decoding_length_infer': None,
     'name': 'attention_rnn_decoder'
 }
+# --------------------- Optimization --------------------- #
+opt = {
+    'optimizer': {
+        'type':  'AdamOptimizer',
+        'kwargs': {
+            'learning_rate': 0.001,
+            # Other keyword arguments for the optimizer class
+        },
+    },
+    'learning_rate_decay': {
+        # Hyperparameters of learning rate decay
+    },
+    'gradient_clip': {
+        # Hyperparameters of gradient clipping
+    },
+    'gradient_noise_scale': None,
+    'name': None
+}
diff --git a/examples/seq2seq_attn/seq2seq_attn.py b/examples/seq2seq_attn/seq2seq_attn.py
@@ -63,7 +63,8 @@ def build_model(batch, train_data):
         tx.losses.sequence_sparse_softmax_cross_entropy(
             labels=batch['target_text_ids'][:, 1:],
             logits=training_outputs.logits,
-            sequence_length=batch['target_length'] - 1))
+            sequence_length=batch['target_length'] - 1),
+        hparams=config_model.opt)
 
     start_tokens = tf.ones_like(batch['target_length']) * \
             train_data.target_vocab.bos_token_id