library(mxnet)
# Data preparation
train <- read.csv('https://github.com/ozt-ca/tjo.hatenablog.samples/raw/master/r_samples/public_lib/jp/mnist_reproduced/short_prac_train.csv')
test <- read.csv('https://github.com/ozt-ca/tjo.hatenablog.samples/raw/master/r_samples/public_lib/jp/mnist_reproduced/short_prac_test.csv')
train <- data.matrix(train)
test <- data.matrix(test)
train.x <- train[,-1]
train.y <- train[,1]
train.x <- t(train.x/255)
test_org <- test
test <- test[,-1]
test <- t(test/255)
test_org[97:103,1:7]## label pixel0 pixel1 pixel2 pixel3 pixel4 pixel5
## [1,] 0 0 0 0 0 0 0
## [2,] 0 0 0 0 0 0 0
## [3,] 0 0 0 0 0 0 0
## [4,] 0 0 0 0 0 0 0
## [5,] 1 0 0 0 0 0 0
## [6,] 1 0 0 0 0 0 0
## [7,] 1 0 0 0 0 0 0data <- mx.symbol.Variable("data")
fc1 <- mx.symbol.FullyConnected(data, name="fc1", num_hidden=128)
act1 <- mx.symbol.Activation(fc1, name="relu1", act_type="relu")
fc2 <- mx.symbol.FullyConnected(act1, name="fc2", num_hidden=64)
act2 <- mx.symbol.Activation(fc2, name="relu2", act_type="relu")
fc3 <- mx.symbol.FullyConnected(act2, name="fc3", num_hidden=10)
softmax <- mx.symbol.SoftmaxOutput(fc3, name="sm")
devices <- mx.cpu()
mx.set.seed(0)
model <- mx.model.FeedForward.create(softmax, X=train.x, y=train.y, ctx=devices, num.round=10, array.batch.size=100,
learning.rate=0.07, momentum=0.9, eval.metric=mx.metric.accuracy,
initializer=mx.init.uniform(0.07),
epoch.end.callback=mx.callback.log.train.metric(100))## Warning in mx.model.select.layout.train(X, y): Auto detect layout input matrix, use colmajor..## Start training with 1 devices## [1] Train-accuracy=0.470204081632653## [2] Train-accuracy=0.8326## [3] Train-accuracy=0.9052## [4] Train-accuracy=0.9278## [5] Train-accuracy=0.9466## [6] Train-accuracy=0.9568## [7] Train-accuracy=0.9646## [8] Train-accuracy=0.9756## [9] Train-accuracy=0.9888## [10] Train-accuracy=0.9926 preds <- predict(model, test)## Warning in mx.model.select.layout.predict(X, model): Auto detect layout input matrix, use colmajor.. dim(preds)## [1] 10 1000 dim(test)## [1] 784 1000 t(preds[1:8,1:7])## [,1] [,2] [,3] [,4] [,5]
## [1,] 0.9999969 1.317755e-14 1.589962e-06 6.074838e-09 3.611948e-11
## [2,] 0.9999995 5.126884e-15 2.002695e-07 1.661002e-07 3.718969e-13
## [3,] 0.9999769 5.292092e-15 1.876923e-07 3.734210e-09 7.665295e-11
## [4,] 0.1416395 5.842936e-12 3.163636e-02 7.424699e-07 7.118647e-04
## [5,] 0.9999815 1.539962e-11 4.302275e-06 1.112681e-06 1.971993e-09
## [6,] 0.9999206 8.039203e-12 8.970006e-06 2.368191e-07 5.628424e-08
## [7,] 0.9990907 1.000745e-10 3.315137e-05 3.895680e-07 2.098697e-04
## [,6] [,7] [,8]
## [1,] 3.478055e-09 9.787780e-07 2.249146e-09
## [2,] 2.376122e-09 3.812215e-08 1.035085e-10
## [3,] 1.806618e-08 5.179310e-08 1.256901e-05
## [4,] 1.657250e-07 8.259764e-01 2.155527e-08
## [5,] 2.141903e-07 6.556791e-06 1.561373e-06
## [6,] 7.062349e-07 3.770629e-05 1.259197e-07
## [7,] 2.745748e-04 3.264020e-04 4.216875e-06## get max.col only for the 7 first rows
max.col(t(preds))[1:7]## [1] 1 1 1 7 1 1 1## rea-ttribute labels to tested images
pred.label <- max.col(t(preds)) - 1
table(pred.label)## pred.label
## 0 1 2 3 4 5 6 7 8 9
## 95 101 101 94 104 104 99 107 102 93Normally, there are 100 images by digital number. for example 0 gets 95, there are 5 image not classed as zero. -1: I think because R starts conting position of matrices by 1, C++ starts counting latruces by 0 max.col returns the position (column number/index ) of higher number in each row.
## predicted labels
table(pred.label)## pred.label
## 0 1 2 3 4 5 6 7 8 9
## 95 101 101 94 104 104 99 107 102 93## initial labels
table(test_org[,1])##
## 0 1 2 3 4 5 6 7 8 9
## 100 100 100 100 100 100 100 100 100 100## compare predicted label with initial labels
table(test_org[,1], pred.label)## pred.label
## 0 1 2 3 4 5 6 7 8 9
## 0 94 0 1 0 0 0 3 0 1 1
## 1 0 100 0 0 0 0 0 0 0 0
## 2 0 0 97 1 1 0 0 1 0 0
## 3 0 0 2 91 0 3 0 1 2 1
## 4 0 0 0 0 95 0 2 1 0 2
## 5 0 1 0 1 0 96 0 0 2 0
## 6 1 0 0 0 1 3 94 0 1 0
## 7 0 0 0 0 0 0 0 100 0 0
## 8 0 0 1 1 1 1 0 0 96 0
## 9 0 0 0 0 6 1 0 4 0 89## get means of identical labels, located in the diagonal
sum(diag(table(test_org[,1],pred.label)))/1000## [1] 0.952To predicted an other dataset of image
# results <- predict(model, new.dataset)
## same as
# results <- mxnet:::predict.MXFeedForwardModel(model, new.dataset)