search

Mri-monitoring / Mri-app

Monitor Caffe with Mri
Apache License 2.0
10 stars 1 forks source link

Test loss in "Final Extremes" #12

Closed haoyangz closed 8 years ago

haoyangz commented 8 years ago

In some cases for example siamese convolutional network, the accuracy is not easily defined and we can only evaluate the performance of a hyper-param set by the testing loss. In current version, it seems that only training loss and testing accuracy are calculated in the "Final Extremes". I wonder if it is possible that the testing loss can also be included? Much appreciated!

nharada1 commented 8 years ago

This actually highlights a more fundamental problem in Mri, namely that the system is not aware of the testing loss at any point. I will implement this soon.

nharada1 commented 8 years ago

I have fixed this in the latest commit. Mri now also outputs "test_loss" in addition to loss and accuracy.