I'm trying to record some values with tensorboard writer with the following lines. If I leave the commented section commented, then the code runs fine. But when I uncomment it, I get an error. Any advice would be appreciated.
if (tb_writer != None):
for key, value in qps.items():
tb_writer.add_scalar("QPv2_" + mode + "_Threshold_" + str(key), value, global_step)
# for annotation_idx in range(1, args.num_classes-1):
# annotation_class = annotation_classes[annotation_idx]
# for threshold_idx in range(len(annotation_thresholds)):
# threshold = annotation_thresholds[threshold_idx]
# tb_writer.add_scalar(annotation_class + "_Threshold_" + str(threshold) + "_Precision", str(precision[threshold_idx][annotation_idx].item()), global_step)
# tb_writer.add_scalar(annotation_class + "_Threshold_" + str(threshold) + "_Recall", str(recall[threshold_idx][annotation_idx].item()), global_step)
Traceback (most recent call last):
File "train_transformer_v5.py", line 782, in <module>
main()
File "train_transformer_v5.py", line 763, in main
global_step, tr_loss, metrics_result = trainer.train(args)
File "train_transformer_v5.py", line 451, in train
tb_writer=self.tb_writer, global_step=self.global_step)
File "/home/xianx/model/metrics_v5.py", line 606, in eval_metrics
tb_writer.add_scalar(annotation_class + "_Threshold_" + str(threshold) + "_Precision", str(precision[threshold_idx][annotation_idx].item()), global_step)
File "/opt/conda/lib/python3.6/site-packages/torch/utils/tensorboard/writer.py", line 348, in add_scalar
scalar_value = workspace.FetchBlob(scalar_value)
File "/opt/conda/lib/python3.6/site-packages/caffe2/python/workspace.py", line 379, in FetchBlob
result = C.fetch_blob(StringifyBlobName(name))
RuntimeError: [enforce fail at pybind_state.cc:221] ws->HasBlob(name). Can't find blob: 0.7158585786819458
My issue was that I was using .add_scalar(...)
while converting my value to a string. Removing that conversion fixed the problem.
The technical post webpages of this site follow the CC BY-SA 4.0 protocol. If you need to reprint, please indicate the site URL or the original address.Any question please contact:yoyou2525@163.com.