Remove metrics placeholder from training loop (blue-oil#959)

oatawa1 · Mar 31, 2020 · e5d138a · e5d138a
1 parent e21d0ae
commit e5d138a
Show file tree

Hide file tree

Showing 3 changed files with 14 additions and 43 deletions.
diff --git a/blueoil/cmd/evaluate.py b/blueoil/cmd/evaluate.py
@@ -103,8 +103,7 @@ def evaluate(config, restore_path, output_dir):
         model.summary(output, labels_placeholder)
 
         summary_op = tf.compat.v1.summary.merge_all()
-
-        metrics_summary_op, metrics_placeholders = executor.prepare_metrics(metrics_ops_dict)
+        metrics_summary_op = executor.metrics_summary_op(metrics_ops_dict)
 
         init_op = tf.compat.v1.global_variables_initializer()
         reset_metrics_op = tf.compat.v1.local_variables_initializer()
@@ -140,14 +139,7 @@ def evaluate(config, restore_path, output_dir):
         else:
             sess.run([metrics_update_op], feed_dict=feed_dict)
 
-    metrics_values = sess.run(list(metrics_ops_dict.values()))
-    metrics_feed_dict = {
-        # TODO: Fix to avoid the implementation depended on the order of dict implicitly
-        placeholder: value for placeholder, value in zip(metrics_placeholders, metrics_values)
-    }
-    metrics_summary, = sess.run(
-        [metrics_summary_op], feed_dict=metrics_feed_dict,
-    )
+    metrics_summary = sess.run(metrics_summary_op)
     validation_writer.add_summary(metrics_summary, last_step)
 
     is_tfds = "TFDS_KWARGS" in config.DATASET
@@ -160,8 +152,7 @@ def evaluate(config, restore_path, output_dir):
         'dataset_name': dataset_name,
         'dataset_path': dataset_path,
         'last_step': int(last_step),
-        # TODO: Fix to avoid the implementation depended on the order of dict implicitly
-        'metrics': {k: float(v) for k, v in zip(list(metrics_ops_dict.keys()), metrics_values)},
+        'metrics': {k: float(sess.run(op)) for k, op in metrics_ops_dict.items()},
     }
     save_json(output_dir, json.dumps(metrics_dict, indent=4,), metrics_dict["last_step"])
     validation_dataset.close()

diff --git a/blueoil/cmd/train.py b/blueoil/cmd/train.py
@@ -115,8 +115,7 @@ def start_training(config):
         model.summary(output, labels_placeholder)
 
         summary_op = tf.compat.v1.summary.merge_all()
-
-        metrics_summary_op, metrics_placeholders = executor.prepare_metrics(metrics_ops_dict)
+        metrics_summary_op = executor.metrics_summary_op(metrics_ops_dict)
 
         init_op = tf.compat.v1.global_variables_initializer()
         reset_metrics_op = tf.compat.v1.local_variables_initializer()
@@ -223,12 +222,7 @@ def start_training(config):
             # train_writer.add_run_metadata(run_metadata, "step: {}".format(step + 1))
             train_writer.add_summary(summary, step + 1)
 
-            metrics_values = sess.run(list(metrics_ops_dict.values()))
-            metrics_feed_dict = {placeholder: value for placeholder, value in zip(metrics_placeholders, metrics_values)}
-
-            metrics_summary, = sess.run(
-                [metrics_summary_op], feed_dict=metrics_feed_dict,
-            )
+            metrics_summary = sess.run(metrics_summary_op)
             train_writer.add_summary(metrics_summary, step + 1)
             train_writer.flush()
         else:
@@ -261,13 +255,7 @@ def start_training(config):
                 else:
                     sess.run([metrics_update_op], feed_dict=feed_dict)
 
-            metrics_values = sess.run(list(metrics_ops_dict.values()))
-            metrics_feed_dict = {
-                placeholder: value for placeholder, value in zip(metrics_placeholders, metrics_values)
-            }
-            metrics_summary, = sess.run(
-                [metrics_summary_op], feed_dict=metrics_feed_dict,
-            )
+            metrics_summary = sess.run(metrics_summary_op)
             if rank == 0:
                 val_writer.add_summary(metrics_summary, step + 1)
                 val_writer.flush()

diff --git a/blueoil/utils/executor.py b/blueoil/utils/executor.py
@@ -103,28 +103,20 @@ def save_pb_file(sess, output_dir, output_node_names=["output"], pb_name="minima
     return pb_name
 
 
-def prepare_metrics(metrics_ops_dict):
-    """Create summary_op and placeholders for training metrics.
+def metrics_summary_op(metrics_ops_dict):
+    """Create summary_op for training metrics.
 
     Args:
         metrics_ops_dict (dict): dict of name and metrics_op.
 
     Returns:
         metrics_summary_op: summary op of metrics.
-        metrics_placeholders: list of metrics placeholder.
 
     """
     with tf.compat.v1.name_scope("metrics"):
-        metrics_placeholders = []
-        metrics_summaries = []
-        for (metrics_key, metrics_op) in metrics_ops_dict.items():
-            metrics_placeholder = tf.compat.v1.placeholder(
-                tf.float32, name="{}_placeholder".format(metrics_key)
-            )
-            summary = tf.compat.v1.summary.scalar(metrics_key, metrics_placeholder)
-            metrics_placeholders.append(metrics_placeholder)
-            metrics_summaries.append(summary)
-
-        metrics_summary_op = tf.compat.v1.summary.merge(metrics_summaries)
-
-    return metrics_summary_op, metrics_placeholders
+        metrics_summaries = [
+            tf.compat.v1.summary.scalar(metrics_key, metrics_op)
+            for (metrics_key, metrics_op) in metrics_ops_dict.items()
+        ]
+
+    return tf.compat.v1.summary.merge(metrics_summaries)