Plot accuracy of both training and validation set over iterations

[Cuong Duc]

I want to plot a graph of accuracy of the CaffeNet over both training and validation set.

Right now, I do snapshotting the model after each 1000 iterations and run the classification code on each individual snapshot on both training and validation data set.

This takes very long time as my training set contains 150k images and the validation set has 50k images.

So I wonder whether there are any faster way the I can do to achieve what I want?

[Dinesh]

If you want to plot the approximate accuracy (as measured over a batch, or a few batches in the case of test), you could just write a script to read the losses and accuracies from caffe train output. To get accuracy over your entire train/val set, I think what you're doing now is the best way.

[Axel Straminsky]

Have you tried using DIGITS?