Tensorflow - single placeholder or list of them

Vote count: 0

i'm working in a code using Tensorflow, and i came to this problem. the neural network i have to model, has many inputs ( lets say something like a image 512x512 ) annd those inputs are connected to other neurons in the network ( in a feedforward architecture ) , i have to test many of these networks and in each case the connections between de input neurons and the other neurons is different.

So far i have done this by creating a placeholder( one dimensional) for each input, store them in a python list, and save in a graph structure the relationships between neurons, but i'm concerned about performance, considering that in each training try, i will have to create something like 100 or more networks. So i end up with 512x512x100 placeholder objects. if is ok do it, but i'll have memory issues, running the code in Amazon Web Services or Google Cloud may help?

So i'm wondering if there is a better way to do it. maybe a way to create a single placeholder object for each network, and access the components in a "array like" way. but so far, i can't figure it out.

Any advise in how to approach this, is welcome :) Thanks.

asked 52 secs ago

RolandDeschain

Let's block ads! (Why?)

Tensorflow - single placeholder or list of them