diff --git a/src/DINCAE.jl b/src/DINCAE.jl
index 0d09baa..a066136 100644
--- a/src/DINCAE.jl
+++ b/src/DINCAE.jl
@@ -46,10 +46,7 @@ import Base: eltype
 import Random: shuffle!
 
 using Profile
-#import Knet: KnetArray, AutoGrad
-#import Knet
 
-#include("knet.jl")
 include("flux.jl")
 include("types.jl")
 include("data.jl")
diff --git a/src/model.jl b/src/model.jl
index 21da655..4d83070 100644
--- a/src/model.jl
+++ b/src/model.jl
@@ -6,7 +6,6 @@ CatSkip(x) = SkipConnection(x,(mx,y) -> cat(mx, y, dims=Val(3)))
 #using JLD2
 
 # save inversion
-#function sinv(x::Union{AbstractArray{T},KnetArray{T},AutoGrad.Result{<:AbstractArray{T}}}; minx = T(1e-3)) where T
 function sinv(x, ; minx = eltype(x)(1e-3))
     T = eltype(x)
     return one(T) ./ max.(x,minx)
diff --git a/src/points.jl b/src/points.jl
index 0d9aa1c..a199111 100644
--- a/src/points.jl
+++ b/src/points.jl
@@ -24,13 +24,7 @@ function interpnd!(pos::AbstractVector{<:NTuple{N}},A,vec) where N
     #return vec
     return nothing
 end
-#=
-function interpnd!(pos::AbstractVector{<:NTuple{N}},A::KnetArray,vec) where N
-    cuA = CuArray(A)
-    cuvec = CuArray(vec)
-    @cuda DINCAE.interpnd!(pos,cuA,cuvec)
-end
-=#
+
 function interpnd!(pos::AbstractVector{<:NTuple{N}},cuA::CuArray,cuvec) where N
     @cuda DINCAE.interpnd!(pos,cuA,cuvec)
 end
@@ -43,8 +37,6 @@ function interpnd(pos,A)
     return vec
 end
 
-#Knet.AutoGrad.@primitive interpnd(pos,A),dy,y 0 interp_adjn(pos,dy,size(A))
-
 function ChainRulesCore.rrule(::typeof(interpnd), pos::AbstractVector{<:NTuple{N}}, A) where N
     function interpnd_pullback(dy)
         dpos = similar(pos)
@@ -83,14 +75,6 @@ function interp_adjn!(pos::AbstractVector{<:NTuple{N}},values,A2) where N
     return nothing
 end
 
-#=
-function interp_adjn!(pos::AbstractVector{<:NTuple{N}},values::KnetArray,A2) where N
-    cuvalues = CuArray(values)
-    cuA2 = CuArray(A2)
-    @cuda interp_adjn!(pos,cuvalues,cuA2)
-end
-=#
-
 function interp_adjn!(pos::AbstractVector{<:NTuple{N}},cuvalues::CuArray,cuA2) where N
     @cuda interp_adjn!(pos,cuvalues,cuA2)
 end
@@ -380,7 +364,7 @@ function costfun(
     xrec,xtrue::Vector{NamedTuple{(:pos, :x),Tuple{Tpos,TA}}},truth_uncertain;
     laplacian_penalty = 0,
     laplacian_error_penalty = laplacian_penalty,
-    ) where TA <: Union{Array{T,N},CuArray{T,N}#=,KnetArray{T,N}=#} where Tpos <: AbstractVector{NTuple{N,T}} where {N,T}
+    ) where TA <: Union{Array{T,N},CuArray{T,N}} where Tpos <: AbstractVector{NTuple{N,T}} where {N,T}
 
     #@show typeof(xin)
     #@show typeof(xrec)
@@ -479,7 +463,7 @@ end
 Mandatory parameters:
 
 * `T`: `Float32` or `Float64`: float-type used by the neural network
-* `Array{T}` or `KnetArray{T}`: array-type used by the neural network.
+* `Array{T}` or `CuArray{T}`: array-type used by the neural network.
 * `filename`: NetCDF file in the format described below.
 * `varname`: name of the primary variable in the NetCDF file.
 * `grid`: tuple of ranges with the grid in the longitude and latitude direction e.g. `(-180:1:180,-90:1:90)`.
diff --git a/src/vector2.jl b/src/vector2.jl
index db01ade..dc1c698 100644
--- a/src/vector2.jl
+++ b/src/vector2.jl
@@ -55,18 +55,9 @@ function vector2_costfun(xrec,xtrue,truth_uncertain,directionobs)
 
     # interpolate to location of observations
     # output should be the same shape as m_true
-    #obsoper(x::Union{KnetArray{T,2},AbstractArray{T,2}}) where T = repeat(x,(1,1,nsites))
-    #obsoper(x::AbstractArray{T,3}) where T = mapslices(obsoper,x,dims=(1,2))
-    #obsoper(x::AbstractArray{T,3}) where T = x
-    #function obsoper(x::Union{KnetArray{T,3},AbstractArray{T,3},CuArray{T,3},
-    #                          AutoGrad.Result{KnetArray{T,3}},AutoGrad.Result{CuArray{T,3}},AutoGrad.Result{Array{T,3}}}) where T
     function obsoper(x)
         tmp = reshape(x,(size(x)[1:2]...,1,size(x,3)))
         #repeat(tmp,inner=(1,1,nsites,1))
-        # https://github.com/denizyuret/Knet.jl/issues/635
-        #tmp[:,:,ones(Int,nsites),:]
-        #cat(tmp,tmp,tmp,dims=3)
-        #cat((tmp for i = 1:nsites)...,dims=3)
         cat([tmp for i = 1:nsites]...,dims=3)
     end