Char-LSTM example fails on eval_metric

[colbec]

The provided example will run completely only if I remove the eval_metric argument from the mx.fit() process. Using:

julia> versioninfo()
Julia Version 0.6.0
Commit 9036443 (2017-06-19 13:05 UTC)
Platform Info:
  OS: Linux (x86_64-pc-linux-gnu)
  CPU: Intel(R) Core(TM) i5-4460  CPU @ 3.20GHz
  WORD_SIZE: 64
  BLAS: libopenblas (USE64BITINT DYNAMIC_ARCH NO_AFFINITY Haswell)
  LAPACK: libopenblas64_
  LIBM: libopenlibm
  LLVM: libLLVM-3.9.1 (ORCJIT, haswell)

julia> Pkg.status("MXNet")
 - MXNet                         0.0.8+             master (dirty)

The provided example code shows:

mx.fit(model, optimizer, data_tr, eval_data=data_val, n_epoch=N_EPOCH,
       initializer=mx.UniformInitializer(0.1),
       callbacks=[mx.speedometer(), mx.do_checkpoint(CKPOINT_PREFIX)], eval_metric=NLL())

which produces on my machine:

ERROR: LoadError: MethodError: no method matching _update_single_output(::NLL, ::MXNet.mx.NDArray{Float32,1}, ::MXNet.mx.NDArray{Float32,2})
Closest candidates are:
  _update_single_output(!Matched::MXNet.mx.MSE, ::MXNet.mx.NDArray{T,N}, !Matched::MXNet.mx.NDArray{T,N}) where {T, N} at /home/colin/.julia/v0.6/MXNet/src/metric.jl:237
Stacktrace:
 [1] _update!(::NLL, ::Array{MXNet.mx.NDArray{Float32,1},1}, ::Array{MXNet.mx.NDArray{Float32,2},1}, ::Val{true}) at /home/colin/.julia/v0.6/MXNet/src/metric.jl:45
 [2] update!(::NLL, ::Array{MXNet.mx.NDArray{Float32,1},1}, ::Array{MXNet.mx.NDArray{Float32,2},1}) at /home/colin/.julia/v0.6/MXNet/src/metric.jl:34
 [3] #fit#10000(::Array{Any,1}, ::Function, ::MXNet.mx.FeedForward, ::MXNet.mx.ADAM, ::CharSeqProvider) at /home/colin/.julia/v0.6/MXNet/src/model.jl:522
 [4] (::MXNet.mx.#kw##fit)(::Array{Any,1}, ::MXNet.mx.#fit, ::MXNet.mx.FeedForward, ::MXNet.mx.ADAM, ::CharSeqProvider) at ./<missing>:0
 [5] include_from_node1(::String) at ./loading.jl:569
 [6] include(::String) at ./sysimg.jl:14
 [7] process_options(::Base.JLOptions) at ./client.jl:305
 [8] _start() at ./client.jl:371
while loading /home/colin/courses/deeplearn/mxnet/trainlstm.jl, in expression starting on line 39

Removing the eval_metric to rely on the default allows the iteration to perform to a conclusion.

I also get many warnings:
WARNING: Task iteration is now deprecated. Use Channels for inter-task communication. A for-loop on a Channel object is terminated by calling close on the object.

[iblislin]

diff --git a/examples/char-lstm/lstm.jl b/examples/char-lstm/lstm.jl
index d930240..46f36c2 100644
--- a/examples/char-lstm/lstm.jl
+++ b/examples/char-lstm/lstm.jl
@@ -123,7 +123,7 @@ mutable struct NLL <: mx.AbstractEvalMetric
   NLL() = new(0.0, 0)
 end

-function mx.update!(metric :: NLL, labels :: Vector{mx.NDArray}, preds :: Vector{mx.NDArray})
+function mx.update!(metric::NLL, labels::Vector{<:mx.NDArray}, preds::Vector{<:mx.NDArray})
   @assert length(labels) == length(preds)
   nll = 0.0
   for (label, pred) in zip(labels, preds)

It failed because of this change: #331