Problem in back propagation

[AHMED IMAM SHAH]

Hello everyone,

I am having problem training my model. My forward pass is working fine. However, the backward pass is giving an error message. The error message is attached. 

Here is the function which is giving the error:

function outputPad(x)

    a, b, c, d = size(x)

    hostx = Array{Float32}(x)

    paddedzeros = zeros(a+2, b+2, c, d)

    paddedzeros[2:a+1, 2:b+1, :, :] .= hostx

    return array_type(paddedzeros)

end

This function pads the output of the deconv4 layer.

I don't understand what is the issue with this function. 

Best,

Ahmed Imam Shah

MS. Computer Science and Engineering

Koç University, Istanbul, Turkey

[Ali Safaya]

Hi AHMED,

I think this is related to the Array type of `hostx` variable. Could you test the following:

function outputPad(x)

    a, b, c, d = size(x)

    hostx = array_type(x)

    paddedzeros = zeros(a+2, b+2, c, d)

    paddedzeros[2:a+1, 2:b+1, :, :] .= paddedzeros + hostx

    return array_type(paddedzeros)

end

--
You received this message because you are subscribed to the Google Groups "COMP541" group.
To unsubscribe from this group and stop receiving emails from it, send an email to COMP541+u...@ku.edu.tr.
To view this discussion on the web visit https://groups.google.com/a/ku.edu.tr/d/msgid/COMP541/CA%2Bsbv-gDWHU0TcmeDWZ9sHLY3qgVGJp8tf4Nhy7wVukhQPENfA%40mail.gmail.com.

[AHMED IMAM SHAH]

Thank you Ali for your reply. I tried that but the problem with this is now the forward pass also does not work.

MethodError: no method matching +(::Array{Float64,4}, ::KnetArray{Float32,4})
Closest candidates are:
  +(::Any, ::Any, !Matched::Any, !Matched::Any...) at operators.jl:538
  +(!Matched::ChainRulesCore.NotImplemented, ::Any) at /kuacc/users/ashah20/.julia/packages/ChainRulesCore/7OROc/src/tangent_arithmetic.jl:24
  +(!Matched::Knet.KnetArrays.Bcasted{Float32}, ::KnetArray{Float32,N} where N) at /kuacc/users/ashah20/.julia/packages/Knet/C0PoK/src/knetarrays/binary.jl:132

Best,

Ahmed

[Ali Safaya]

Please make sure that both arrays have the same type, either Array for CPU or KnetArray GPU.

[AHMED IMAM SHAH]

That is what I was trying to do. Making x a CPU array then to the computation and then back to GPU. It didnt work. Now I made both of them GPU array. Again I get the error in the same function in the backpropagation. 

GPU compilation of kernel broadcast_kernel(CUDA.CuKernelContext, SubArray{Float32,4,CUDA.CuDeviceArray{Float32,4,1},Tuple{UnitRange{Int64},UnitRange{Int64},Base.Slice{Base.OneTo{Int64}},Base.Slice{Base.OneTo{Int64}}},false}, Base.Broadcast.Broadcasted{Nothing,NTuple{4,Base.OneTo{Int64}},typeof(identity),Tuple{AutoGrad.Result{KnetArray{Float32,4}}}}, Int64) failed

KernelError: passing and using non-bitstype argument

Argument 4 to your kernel function is of type Base.Broadcast.Broadcasted{Nothing,NTuple{4,Base.OneTo{Int64}},typeof(identity),Tuple{AutoGrad.Result{KnetArray{Float32,4}}}}, which is not isbits:

  .args is of type Tuple{AutoGrad.Result{KnetArray{Float32,4}}} which is not isbits.

    .1 is of type AutoGrad.Result{KnetArray{Float32,4}} which is not isbits.

      .value is of type Union{Nothing, KnetArray{Float32,4}} which is not isbits.

      .func is of type Any which is not isbits.

      .args is of type Any which is not isbits.

      .kwargs is of type Any which is not isbits.

Stacktrace:

 [1] differentiate(::Function; o::Base.Iterators.Pairs{Union{},Union{},Tuple{},NamedTuple{(),Tuple{}}}) at /kuacc/users/ashah20/.julia/packages/AutoGrad/TTpeo/src/core.jl:148

 [2] differentiate at /kuacc/users/ashah20/.julia/packages/AutoGrad/TTpeo/src/core.jl:135 [inlined]

 [3] trainfunc(::Chain, ::KnetArray{Float32,4}) at ./In[31]:7

 [4] top-level scope at ./In[33]:5

 [5] include_string(::Function, ::Module, ::String, ::String) at ./loading.jl:1091

[Ali Safaya]

Can you please provide a fully reproducible example?