The Training

Good to see this traing section!

Hi Skander, Thank you for your comment. Usually a network would work on my little laptop normally and start training, but the training process is very slow (several hours for 1 epoch cycle), that is when I move to a dedicated cloud based GPU since I do not have my own setup (also I have the freedom to choose different GPU types). In the case of this neural network, it is too big to even start on my computer, so I had to move to a cloud GPU directly. Luckily google provides a free GPU (for only 12 hours), so it is great to test the workings of the network and make sure the script has no errors. I have already passed this step. I also have access to a supercomputer. Though it only has CPUs to perform computation, it is large enough to emulate a GPU, but a GPU is stills better. I do not have my own system built since up to this point cloud services were enough to generate the data I need and train the networks I use, and cheaper tha. Building my own system.

This is fantastic, thanks for sharing! You mentioned not having enough compute power on your laptop for this example - do you have your own development machine for other ML/DL research or do you generally stick with cloud-based GPU resources?