machine learning - Manually changing nodes in a TensorFlow network -


i have standard connected feedforward network relu activations, looks like

x -> h1 -> h2 -> h3 -> out 

and i've trained learn simple function (sum inputs , take sin). experiment, want inject different functions of x in network. so, example, if h2 has size [none, 20] want set h2[:,0] = x[:,0]**2 - want first node in layer h2 function takes first coordinate of original input , squares it.

the function scatter_nd_update doesn't work because thing i'm trying assign entry of h2 variable, , i'm pretty sure doesn't make sense assign h2 anyway. else can do?


Comments

Popular posts from this blog

javascript - Create a stacked percentage column -

Optimising Firebase database by automatically overwriting data -

javascript - Angular UI-Grid customTemplate directive causing rows to load slowly/? -