Having a RAM and CPU overload issue

[dhoule]

Working in the linear branch.

The linear algorithm, just increasing the number of nodes in the single, hidden layer, completely overloads the system I'm working on. This does not bode well for the genetic algorithm.

I need to look more into the Tensorflow API about removing/streamlining the overhead of the code.

• One way was to switch to the GPU version of Tensorflow; I'm currently using the CPU version.

One way around this, maybe, if to have all of the processing of the runs done outside of the code, like I did for the term project.

• The code would need to take command line arguments.
• The argument would be a BASH loop iterator, $i for instance.
• This would take longer, as the images would need to be reloaded every, single loop.
  • Should be able to make each loop a single run for each design version.

[dhoule]

Using command line arguments has proven to be a way to limit RAM usage, but the model; CNN; still overloads the CPU every time. I might have to switch to another system. Using command line arguments does require the image dataset to be reloaded each time the program is restarted, but this appears to be small when compared to the RAM overload of trying to run the linear algorithm within the code itself. It seems that comparisons of the runs are going to be have to be made outside of the CNN code.

I am currently only focussing on the design aspects of the CNN. When I am able to tell what designs performed better, overall, I will start modifying the CNN under those designs.