Few optimizations with numpy (#4982)

examples/trials/benchmarking/automlbenchmark/nni/extensions/NNI/architectures/run_mlp.py

@@ -117,7 +117,7 @@ def run_mlp(dataset, config, tuner, log):
             
                 # Here score is the output of score() from the estimator
                 cur_score = cross_val_score(cur_model, X_train, y_train)
-                cur_score = sum(cur_score) / float(len(cur_score))
+                cur_score = np.mean(cur_score)
                 if np.isnan(cur_score):
                     cur_score = 0
             

examples/trials/benchmarking/automlbenchmark/nni/extensions/NNI/architectures/run_random_forest.py

@@ -138,7 +138,7 @@ def run_random_forest(dataset, config, tuner, log):
             
                 # Here score is the output of score() from the estimator
                 cur_score = cross_val_score(cur_model, X_train, y_train)
-                cur_score = sum(cur_score) / float(len(cur_score))
+                cur_score = np.mean(cur_score)
                 if np.isnan(cur_score):
                     cur_score = 0
             

examples/trials/kaggle-tgs-salt/metrics.py

@@ -66,7 +66,7 @@ def compute_eval_metric(gt, predictions):
     thresholds = [0.5, 0.55, 0.6, 0.65, 0.7, 0.75, 0.8, 0.85, 0.9, 0.95]
     ious = compute_ious(gt, predictions)
     precisions = [compute_precision_at(ious, th) for th in thresholds]
-    return sum(precisions) / len(precisions)
+    return np.mean(precisions)
 
 
 def intersection_over_union(y_true, y_pred):

nni/algorithms/compression/v2/pytorch/pruning/tools/task_generator.py

@@ -260,13 +260,10 @@ class SimulatedAnnealingTaskGenerator(TaskGenerator):
         num_weights = sorted([self.weights_numel[op_name] for op_name in op_names])
         sparsity = sorted(random_sparsity)
         
-        total_weights = 0
-        total_weights_pruned = 0
-
         # calculate the scale
-        for idx, num_weight in enumerate(num_weights):
-            total_weights += num_weight
-            total_weights_pruned += int(num_weight * sparsity[idx])
+        total_weights = np.sum(num_weights)
+        total_weights_pruned = np.sum([int(num_weight * sparsity[idx]) for idx, num_weight in enumerate(num_weights)])
+        
         if total_weights_pruned == 0:
             return None
 

nni/algorithms/hpo/ppo_tuner/ppo_tuner.py

@@ -633,7 +633,7 @@ class PPOTuner(Tuner):
             # use mean of finished trials as the result of this failed trial
             values = [val for val in self.trials_result if val is not None]
             logger.warning('In trial_end, values: %s', values)
-            self.trials_result[trial_info_idx] = (sum(values) / len(values)) if values else 0
+            self.trials_result[trial_info_idx] = (np.mean(values)) if values else 0
             self.finished_trials += 1
             if self.finished_trials == self.inf_batch_size:
                 logger.debug('Start next round inference in trial_end')