====== Machine Learning ======


===== Useful Resources =====

http://www-bcf.usc.edu/~gareth/ISL/\\
https://www.r-project.org/about.html\\
https://www.rstudio.com/products/rstudio/\\
https://www.kaggle.com/c/digit-recognizer\\
[[https://www.guru99.com/tensorflow-tutorial.html|TensorFlow and general SW learning website]]

References

[1] C.J. Wu et al., Machine Learning at Facebook: Understanding Inference at the Edge [From Facebook]\\
[2] N.P. Jouppi et. al., In Datacenter Performance Analysis of a Tensor Processing Unit [From Google]\\
[3] Vivien Sze et. al., Efficient Processing of Deep Neural Networks: A tutorial and survey [From MIT]\\
[4] Amr Suleiman et. al., Navion: A 2mW Fully Integrated Real-Time Visual-Inertial Odometry Accelerator for Autonomous Navigation of Nano Drones [From MIT]
====== Handwriting Learning Code Example ======

===== Example 1 =====

<code>
library(h2o)
h2o.init(nthreads = 6)
h2o.removeAll()
traindat=read.csv("train.csv")
traindat[,1] = as.factor(traindat[,1])
traindat[,2:785] = ifelse(traindat[,2:785]>20, 1, 0)
par(mfrow = c(10,10), mar = c(0,0,0,0))
for (i in 1:100) {
  y = as.matrix(traindat[i,2:785])
  dim(y) = c(28,28)
  image(y[,28:1], col = gray(1:0), axes = FALSE)
  text(0.2, 0, traindat[i,1], cex = 3, col = 2, pos = c(3,4))
}
train.h2o = as.h2o(traindat)
nnmodel.h2o = h2o.deeplearning(
  x = 2:785,
  y = 1,
  training_frame = train.h2o,
  activation = "RectifierWithDropout",
  input_dropout_ratio = 0.2,
  balance_classes = TRUE,
  hidden = c(1024,1024,1024),
  l1 = 1e-5,
  classification_stop = 0.001,
  epochs = 100
)
testdat = read.csv("test.csv")
testdat = ifelse(testdat>20, 1, 0)
test.h2o = as.h2o(testdat)
pred.h2o = h2o.predict(nnmodel.h2o, newdata = test.h2o)
pred = as.data.frame(pred.h2o)
par(mfrow = c(10,10), mar = c(0,0,0,0))
for (i in 1:100) {
  y = as.matrix(testdat[i,])
  dim(y) = c(28,28)
  image(y[,28:1],col = gray(255:0/255), axes = FALSE)
  text(0.2, 0, pred[i,1], cex = 3, col = 2, pos = c(3,4))
}
</code>

===== Example 2 =====
<code>
library(h2o)
h2o.init(nthreads = 6)
h2o.removeAll()
train.h2o = h2o.importFile("train.csv")
train.h2o[,1] = as.factor(train.h2o[,1])
nnmodel.h2o = h2o.deeplearning(
  x = 2:785,
  y = 1,
  training_frame = train.h2o,
  activation = "RectifierWithDropout",
  input_dropout_ratio = 0.2,
  balance_classes = TRUE,
  hidden = c(1024,1024,1024),
  l1 = 1e-5,
  classification_stop = 0.001,
  epochs = 100
)
test.h2o = h2o.importFile("test.csv")
pred.h2o = h2o.predict(nnmodel.h2o, newdata = test.h2o)
testdat = as.data.frame(test.h2o)
pred = as.data.frame(pred.h2o)
par(mfrow = c(10,10), mar = c(0,0,0,0))
for (i in 1:100) {
  y = as.matrix(testdat[i,])
  dim(y) = c(28,28)
  image(y[,28:1],col = gray(255:0/255), axes = FALSE)
  text(0.2, 0, pred[i,1], cex = 3, col = 2, pos = c(3,4))
}
</code>