A shiny app that demonstrate model evaluation techniques

The goal is to build an shiny app that user could interact with it. The idea is to create an playground for user model neural net interactively.

model: Species ~ Sepal.Length + Sepal.Width

a 2-12-3 network with 75 weights
inputs: Sepal.Length Sepal.Width 
output(s): Species 
options were - softmax modelling 

With default settings, number of neurons = 12
The Confusion Matrix

pred         setosa versicolor virginica
  setosa         50          0         0
  versicolor      0         37        11
  virginica       0         13        39

So we have fit an vanilla nnet with capability of user interaction for one hyperparameter, number of neurons.

In the future, we would like to make this more interactive on other hyperparameters such as learning rate, selection of loss functions and optimizers.

1. Github Repo that hosts server.R and ui.R
2. Shiny App hosted on shinyapp.io
3. This slides hosted on rpubs