@nswamy The following is the current behavior of the code. We need to reflect this in the "train-accuracy" log in base_module.py
With Speedometer's `auto_reset=True` ``` INFO:root:Epoch[0] Batch [1-100] Speed: 45690.13 samples/sec accuracy=0.772123 INFO:root:Epoch[0] Batch [101-200] Speed: 50611.24 samples/sec accuracy=0.898594 . . . INFO:root:Epoch[0] Batch [801-900] Speed: 52047.39 samples/sec accuracy=0.950625 INFO:root:Epoch[0] Batch [901-938] Train-accuracy=0.944679 INFO:root:Epoch[0] Time cost=1.250 INFO:root:Epoch[0] Validation-accuracy=0.953125 ``` With `auto_reset=False` ``` INFO:root:Epoch[0] Batch [1-100] Speed: 16628.82 samples/sec accuracy=0.759746 INFO:root:Epoch[0] Batch [1-200] Speed: 40806.39 samples/sec accuracy=0.828980 . . . INFO:root:Epoch[0] Batch [1-900] Speed: 42613.81 samples/sec accuracy=0.911470 INFO:root:Epoch[0] Batch [1-938] Train-accuracy=0.912830 INFO:root:Epoch[0] Time cost=1.811 INFO:root:Epoch[0] Validation-accuracy=0.956509 ``` [ Full content available at: https://github.com/apache/incubator-mxnet/pull/12182 ] This message was relayed via gitbox.apache.org for [email protected]
