ADCMEKit.jl

Make Inverse Modeling Easy!
Author kailaix
Popularity
1 Star
Updated Last
4 Years Ago
Started In
February 2020

ADCMEKit

ADCMEKit is used to facilitate inverse modeling using ADCME.

Basic Functions

linedata(a, b): generate (1-α)a+αb as a function of α

meshview(a): generate a m×n matrix in the neighborhood of a

loss1 = x->sum(x.^2-0.1*x)
v = linedata(zeros(10), rand(10))
l = loss1.(v)
lineview(l)

close("all")
v = landscapedata(zeros(10))
l = loss1.(v)
meshview(l)
lineview meshview

Use with ADCME

lineview(sess, pl, loss, θ1)

pl = placeholder(Float64, shape=[2])
l = sum(pl^2-pl*0.1)
sess = Session(); init(sess)
lineview(sess, pl, l, rand(2))

meshview(sess, pl, loss, θ1)

pl = placeholder(Float64, shape=[2])
l = sum(pl^2-pl*0.1)
sess = Session(); init(sess)
meshview(sess, pl, l, rand(2))

gradview(sess, pl, loss, u0): use with ADCME.jl to test gradients

pl = placeholder(Float64, shape=[2])
l = sum(pl^2-pl*0.1)
sess = Session(); init(sess)
gradview(sess, pl, l, rand(2))
lineview meshview gradview

Used By Packages

No packages found.