I have 6000 Images to be trained on a Neural Network.

My current PC Specs :-

32GB RAM, i5 2 core Processor, Standard GPU (No work going on GPU), 1TB Hard Disk

My Neural Network Specs :-

3000 Epoch Size, 6 Batch Size

Process is getting killed with these specifications.

How much Hardware is required to trained it well? What would be the ideal Batch Size and Epoch Size?

  • $\begingroup$ possible flavors of resource constraints : CPU (or gpu but not in this case), memory bandwidth, total memory, disc IO access ... you need to learn how to monitor your system to see which is the weakest link ... lower number of images to 1000 and see what which of above is starting to sweat $\endgroup$ – Scott Stensland Aug 24 '18 at 21:19

This is not really a data science specific question, so you might want to ask elsewhere; in any case, you need to provide more information!

What error message are you receiving? Are you running out of memory? What size are the images?

Unless your images are really really large, I would not expect you to be running out of memory, given you have 32Gb. however, if you have built a massive network, where e.g. many ConvNet filters are then passed to an extremely wide FullyConnected layer (Dense layer), you will have a huge number of weights, which may result in a memory error.

Try creating your model (or a similar one) using Keras instead of pure Tensorflow, and then using model.summary on the compiled model. This will show you a nice overview of the number of weights in each layer.

| improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.