!3368 Restore the code to collect the graph network

Merge pull request !3368 from LiHongzhang/oh_graph

mindspore/train/callback/_summary_collector.py

@@ -182,6 +182,7 @@ class SummaryCollector(Callback):
         self._custom_lineage_data = custom_lineage_data
 
         self._temp_optimizer = None
+        self._has_saved_train_network = False
         self._has_saved_custom_data = False
         self._is_parse_loss_success = True
         self._first_step = True
@@ -215,7 +216,7 @@ class SummaryCollector(Callback):
     @staticmethod
     def _check_positive(name, value, allow_none=False):
         """Check if the value to be int type and positive."""
-        if allow_none:
+        if allow_none and value is None:
             return
         check_value_type(name, value, int)
         if value <= 0:
@@ -294,8 +295,9 @@ class SummaryCollector(Callback):
 
             self._collect_dataset_graph(cb_params)
             if self._collect_tensor_freq is None:
+                default_tensor_summary_limit = 50
                 total_step = cb_params.epoch_num * cb_params.batch_num
-                self._collect_tensor_freq = max(self._collect_freq, total_step // 50)
+                self._collect_tensor_freq = max(self._collect_freq, total_step // default_tensor_summary_limit)
 
         if self._custom_lineage_data and not self._has_saved_custom_data:
             packaged_custom_data = self._package_custom_lineage_data(self._custom_lineage_data)
@@ -309,6 +311,8 @@ class SummaryCollector(Callback):
         cb_params = run_context.original_args()
         if cb_params.mode != ModeEnum.TRAIN.value:
             return
+        if not self._has_saved_train_network:
+            self._collect_graphs(cb_params)
         if self._first_step:
             # Notice: This way of determining whether dataset sink mode is True does not work in the eval scenario
             self._dataset_sink_mode = cb_params.cur_step_num == cb_params.batch_num
@@ -424,6 +428,7 @@ class SummaryCollector(Callback):
         if graph_proto is None:
             return
 
+        self._has_saved_train_network = True
         self._record.add_value(PluginEnum.GRAPH.value, 'train_network/auto', graph_proto)
 
     def _collect_metric(self, cb_params):