example: fix parametric NDArray in lstm example (#370)

and fix some depwarns fix #368
dmlc · Dec 10, 2017 · db09528 · db09528
1 parent 2d7cdc6
commit db09528
Show file tree

Hide file tree

Showing 5 changed files with 34 additions and 35 deletions.
diff --git a/examples/char-lstm/config.jl b/examples/char-lstm/config.jl
@@ -1,24 +1,23 @@
-const DROPOUT       = 0
-const BATCH_SIZE    = 32
-const SEQ_LENGTH    = 32
-const DIM_HIDDEN    = 256
-const DIM_EMBED     = 256
-const LSTM_N_LAYER  = 2
-const N_EPOCH       = 21
-const BASE_LR       = 0.01
-const WEIGHT_DECAY  = 0.00001
-const CLIP_GRADIENT = 1
-const NAME          = :ptb
-const N_GPU         = 4
-const USE_GPU       = true
-const DATA_TR_RATIO = 0.9
-const CKPOINT_PREFIX = joinpath(dirname(@__FILE__), "checkpoints/$NAME")
+const DROPOUT        = 0
+const BATCH_SIZE     = 32
+const SEQ_LENGTH     = 32
+const DIM_HIDDEN     = 256
+const DIM_EMBED      = 256
+const LSTM_N_LAYER   = 2
+const N_EPOCH        = 21
+const BASE_LR        = 0.01
+const WEIGHT_DECAY   = 0.00001
+const CLIP_GRADIENT  = 1
+const NAME           = :ptb
+const N_GPU          = 1
+const USE_GPU        = true
+const DATA_TR_RATIO  = 0.9
+const CKPOINT_PREFIX = joinpath(@__DIR__, "checkpoints/$NAME")
 
-const BATCH_SIZE_SMP= 10
-const SAMPLE_LENGTH = 100
-const SAMPLE_START  = 'a'
-
-const UNKNOWN_CHAR  = Char(0)
-const INPUT_FILE    = joinpath(dirname(@__FILE__), "input.txt")
-const VOCAB_FILE    = joinpath(dirname(@__FILE__), "vocab.dat")
+const BATCH_SIZE_SMP = 10
+const SAMPLE_LENGTH  = 100
+const SAMPLE_START   = 'a'
 
+const UNKNOWN_CHAR   = Char(0)
+const INPUT_FILE     = joinpath(@__DIR__, "input.txt")
+const VOCAB_FILE     = joinpath(@__DIR__, "vocab.dat")
diff --git a/examples/char-lstm/lstm.jl b/examples/char-lstm/lstm.jl
@@ -123,12 +123,12 @@ 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)
     @mx.nd_as_jl ro=(label, pred) begin
-      nll -= sum(log(max(broadcast_getindex(pred, round(Int,label+1), 1:length(label)), 1e-20)))
+      nll -= sum(log.(max.(broadcast_getindex(pred, round.(Int,label+1), 1:length(label)), 1e-20)))
     end
   end
 

diff --git a/examples/char-lstm/sampler.jl b/examples/char-lstm/sampler.jl
@@ -1,6 +1,6 @@
-include(joinpath(dirname(@__FILE__), "config.jl"))
-include(joinpath(dirname(@__FILE__), "lstm.jl"))
-include(joinpath(dirname(@__FILE__), "seq-data.jl"))
+include(joinpath(@__DIR__, "config.jl"))
+include(joinpath(@__DIR__, "lstm.jl"))
+include(joinpath(@__DIR__, "seq-data.jl"))
 
 using StatsBase
 using MXNet

diff --git a/examples/char-lstm/seq-data.jl b/examples/char-lstm/seq-data.jl
@@ -60,7 +60,7 @@ end
 #--/provide
 
 #--eachbatch-part1
-function mx.eachbatch(p :: CharSeqProvider)
+function mx.eachbatch(p::CharSeqProvider)
   data_all  = [mx.zeros(shape) for (name, shape) in mx.provide_data(p)]
   label_all = [mx.zeros(shape) for (name, shape) in mx.provide_label(p)]
 
@@ -73,7 +73,7 @@ function mx.eachbatch(p :: CharSeqProvider)
 
   #--eachbatch-part2
   #...
-  function _text_iter()
+  function _text_iter(c::Channel)
     text = p.text
 
     n_batch = floor(Int, length(text) / p.batch_size / p.seq_len)
@@ -100,11 +100,11 @@ function mx.eachbatch(p :: CharSeqProvider)
         copy!(label_all[i], label_jl[i])
       end
 
-      produce(batch)
+      put!(c, batch)
     end
   end
 
-  return Task(_text_iter)
+  return Channel(_text_iter)
 end
 #--/eachbatch-part2
 

diff --git a/examples/char-lstm/train.jl b/examples/char-lstm/train.jl
@@ -1,6 +1,6 @@
-include(joinpath(dirname(@__FILE__), "config.jl"))
-include(joinpath(dirname(@__FILE__), "lstm.jl"))
-include(joinpath(dirname(@__FILE__), "seq-data.jl"))
+include(joinpath(@__DIR__, "config.jl"))
+include(joinpath(@__DIR__, "lstm.jl"))
+include(joinpath(@__DIR__, "seq-data.jl"))
 
 # build vocabulary
 vocab   = build_vocabulary(INPUT_FILE, VOCAB_FILE)
@@ -29,7 +29,7 @@ data_val  = CharSeqProvider(text_val, BATCH_SIZE, SEQ_LENGTH, vocab, NAME,
 if USE_GPU
   context = [mx.gpu(i) for i = 0:N_GPU-1]
 else
-  context = [mx.cpu()]
+  context = mx.cpu()
 end
 
 #--train