Caffe2 Quick Start Guide by Ashwin Nanjappa

Author:Ashwin Nanjappa [Ashwin Nanjappa]
Language: eng
Format: epub
Tags: COM044000 - COMPUTERS / Neural Networks, COM004000 - COMPUTERS / Intelligence (AI) and Semantics, COM016000 - COMPUTERS / Computer Vision and Pattern Recognition
Publisher: Packt
Published: 2019-05-31T11:52:08+00:00

We can monitor the health of the training by looking at the raw values, or importing them into a spreadsheet, or plotting them in a graph. Figure 3.7 shows a graph plotted from the values of one such training session:

Figure 3.7: Loss and accuracy of training our model

We can see that the loss is high at the beginning. This is expected, as we typically initialize a network with zero or random weights. As the training proceeds, we see that the loss decreases and, correspondingly, the accuracy of the network increases. If you do not see the loss decreasing or accuracy increasing, then that indicates a problem with our training parameters or training data. If the training pace is slow or is causing values to blow off, then you might need to tweak the learning rate and such parameters.

We can typically stop training at any iteration where the loss curve is leveling off and the accuracy is suitable. To aid the export of a model at a particular iteration, it is a good idea to export the model to disk (demonstrated in Chapter 5, Working with Other Frameworks) after each iteration. That way, you can pick the model at the best iteration after the training is complete.

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.