In yiqiao-yin/YinsKerasNN: Yin's Version of Neural Network through Keras Framework

This is an R Markdown Notebook. When you execute code within the notebook, the results appear beneath the code.

Get Data

Let us download data using Keras package.

# Get data
library(keras)
mnist <- dataset_mnist()
x_train <- mnist$train$x
y_train <- mnist$train$y
x_test <- mnist$test$x
y_test <- mnist$test$y
dim(x_train); dim(y_train); dim(x_test); dim(y_test)

Prep Data

Let us resize the data by array_reshape() function. This way we can feed in the architecture the correct dimensions.

# Reshap & Rescale
x_train <- array_reshape(x_train, c(nrow(x_train), 784))
x_test <- array_reshape(x_test, c(nrow(x_test), 784))
x_train <- x_train / 255; x_test <- x_test / 255
dim(x_train); dim(y_train); dim(x_test); dim(y_test)

# Combine
train <- data.frame(cbind(y_train, x_train))
test <- data.frame(cbind(y_test, x_test))
colnames(test) <- colnames(train)
all <- data.frame(rbind(train, test)); colnames(all)[1] <- "label"

Execute the function YinsKerasNN

Let us call the function YinsKerasNN and this will execute a 3-layer neural network on given data.

# We want to store everything in an object:
# Let us call this **Result**
Begin.Time <- Sys.time()
Result <- YinsKerasNN::YinsKerasNN(x = all[,-1], y = all[, 1], cutoff = 0.8)
End.Time <- Sys.time(); print(paste0("Time spent on training this machine: ", End.Time - Begin.Time))

Results

Let us visualize the results by the following plot.

# Check:
# Training and Validation Plot:
Result$Training.Plot
# Confusion Matrix:
Result$Confusion.Matrix
# Accuracy (percentage of true positive and true negative):
Result$Testing.Accuracy

Implication

The results above indicates good performance on test set. We have ~97\% accuracy for all 10 classes. If we set up a Turing Test, we are likely to fool humans that we have a person guessing the results.

yiqiao-yin/YinsKerasNN documentation built on July 4, 2021, 2:26 a.m.