more on data io...specially for images.

[AndreGuerra123]

Hello, Chin Lin.
First of all, nicely structured guide you have here.
( I'm starting with mxnet and there is not so much help in R).
I don't know if this is the correct place (since it is not an issue) to put it but here it goes:
I would like to see more on data io...specially how to build datasets efficiently from images.
The main topics I'm dealing with are:
-convert images to Xdataset (most data efficient way (space and speed));
-optimal structure the final X dataset (HeightxWidthxChannelxObservation) or other;
-normalized the data?from 0 to 255? from 0 to 1?other?
-is there any reference guide or format that should be used(data.matrix or array)?
Hope my feedback help your guide in the future, as your answer will certainly help me.
Kind regards, Andre

[xup6fup]

Dear Andre,

You can try following codes:

library(mxnet)

CustomCSVIter <- setRefClass("CustomCSVIter",
fields=c("iter", "data.csv", "data.shape", "batch.size"),
contains = "Rcpp_MXArrayDataIter",
methods=list(
initialize=function(iter, data.csv, data.shape, batch.size){
feature_len <- data.shape*data.shape + 1
csv_iter <- mx.io.CSVIter(data.csv=data.csv, data.shape=c(feature_len), batch.size=batch.size)
.self$iter <- csv_iter
.self$data.csv <- data.csv
.self$data.shape <- data.shape
.self$batch.size <- batch.size
.self
},
value=function(){
val <- as.array(.self$iter$value()$data)
val.x <- val[-1,]
val.y <- val[1,]
val.x <- val.x/255
dim(val.x) <- c(data.shape, data.shape, ncol(val.x))
random.x = sample(0:2, 1) + 1:26
random.y = sample(0:2, 1) + 1:26
val.x <- val.x[random.x,random.y,]
dim(val.x) <- c(data.shape - 2, data.shape - 2, 1, ncol(val))
val.x <- mx.nd.array(val.x)
val.y <- mx.nd.array(val.y)
list(data=val.x, label=val.y)
},
iter.next=function(){
.self$iter$iter.next()
},
reset=function(){
.self$iter$reset()
},
num.pad=function(){
.self$iter$num.pad()
},
finalize=function(){
.self$iter$finalize()
}
)
)

batch.size <- 50
train.iter <- CustomCSVIter$new(iter = NULL, data.csv = "mnist_train.csv", data.shape = 28, batch.size = batch.size)

lenet.model <- function(){
data <- mx.symbol.Variable('data')
conv1 <- mx.symbol.Convolution(data=data, kernel=c(5,5), num_filter=20) #first conv
tanh1 <- mx.symbol.Activation(data=conv1, act_type="tanh")
pool1 <- mx.symbol.Pooling(data=tanh1, pool_type="max", kernel=c(2,2), stride=c(2,2))
conv2 <- mx.symbol.Convolution(data=pool1, kernel=c(5,5), num_filter=50)# second conv
tanh2 <- mx.symbol.Activation(data=conv2, act_type="tanh")
pool2 <- mx.symbol.Pooling(data=tanh2, pool_type="max", kernel=c(2,2), stride=c(2,2))
flatten <- mx.symbol.Flatten(data=pool2)
fc1 <- mx.symbol.FullyConnected(data=flatten, num_hidden=100) # first fullc
tanh3 <- mx.symbol.Activation(data=fc1, act_type="tanh")
fc2 <- mx.symbol.FullyConnected(data=tanh3, num_hidden=10) # second fullc
network <- mx.symbol.SoftmaxOutput(data=fc2) # loss
network
}
network <- lenet.model()

n.cpu <- 4
device.cpu <- lapply(0:(n.cpu-1), function(i) {mx.cpu(i)})

model <- mx.model.FeedForward.create(symbol=network,
X=train.iter,
ctx=device.cpu,
num.round=2,
array.batch.size = batch.size,
learning.rate =0.1,
momentum=0.9,
eval.metric=mx.metric.accuracy,
wd=0.00001,
batch.end.callback = mx.callback.log.speedometer(batch.size, frequency = 100)
)