maxmustermann/DeepLearningExamples
1
0
Fork 0
Code Issues Pull requests Projects Releases Wiki Activity

[UNetmed/TF] Fix registered metric names

Browse source
This commit is contained in:
Michal Marcinkiewicz 2021-09-21 05:00:26 -07:00 committed by Krzysztof Kudrynski
parent 8d31298c56
commit 03a93a1cb2
2 changed files with 18 additions and 15 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

5
TensorFlow/Segmentation/UNet_Medical/utils/hooks/profiling_hook.py
View file

@ -50,5 +50,6 @@ class ProfilingHook(tf.estimator.SessionRunHook):
def end(self, session):
if hvd.rank() == 0:
stats = process_performance_stats(np.array(self._timestamps),
self._global_batch_size)
self.logger.log(step=(), data={metric: value for (metric, value) in stats})
self._global_batch_size,
self.mode)
self.logger.log(step=(), data=stats)

28
TensorFlow/Segmentation/UNet_Medical/utils/parse_results.py
View file

@ -1,4 +1,4 @@
# Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
# Copyright (c) 2021, NVIDIA CORPORATION. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
@ -17,18 +17,21 @@ import numpy as np
import argparse
def process_performance_stats(timestamps, batch_size):
timestamps_ms = 1000 * timestamps
latency_ms = timestamps_ms.mean()
std = timestamps_ms.std()
n = np.sqrt(len(timestamps_ms))
throughput_imgps = (1000.0 * batch_size / timestamps_ms).mean()
def process_performance_stats(timestamps, batch_size, mode):
""" Get confidence intervals
:param timestamps: Collection of timestamps
:param batch_size: Number of samples per batch
:param mode: Estimator's execution mode
:return: Stats
"""
timestamps_ms = 1000 * timestamps
throughput_imgps = (1000.0 * batch_size / timestamps_ms).mean()
stats = {f"throughput_{mode}": throughput_imgps,
f"latency_{mode}_mean": timestamps_ms.mean()}
for level in [90, 95, 99]:
stats.update({f"latency_{mode}_{level}": np.percentile(timestamps_ms, level)})
stats = [("Throughput Avg", str(throughput_imgps)),
('Latency Avg:', str(latency_ms))]
for ci, lvl in zip(["90%:", "95%:", "99%:"],
[1.645, 1.960, 2.576]):
stats.append(("Latency_" + ci, str(latency_ms + lvl * std / n)))
return stats
@ -77,4 +80,3 @@ if __name__ == '__main__':
parse_convergence_results(path=args.model_dir, environment=args.env)
elif args.exec_mode == 'benchmark':
pass
print()