The objectives for this week are:
Chapter 6, exercise 1.
Chapter 6, exercise 7a. Take a look at the solution at http://www.jeffreylimbacher.me/ISLR-exercises/chap6.html
library(tidyverse) set.seed(20200515) n <- 1000 b <- c(-2,-1) x <- rnorm(n) y <- b+b*x+rnorm(n) df <- data.frame(x,y) ggplot(df, aes(x, y)) + geom_point()
What would you expect the values for \(\beta_0, \beta_1\) be (or close to) from maximum likelihood?
What are the least squares estimates for these quantities?
Change the values for \(\beta_0, \beta_1\) (within the range of -3, 3 only) and the sample size, and examine the resulting maxima.
This exercise is investigating neural network model fitting. A neural network model was fitted to the
wiggle.csv using the
nnet package in R, from many random starts, and using 2-4 nodes in the hidden layer. The best model is in the data object
nnet_best.rda, and the full set of model fits are stored in
nnet_many.rda. We are going investigate the best fit, and the complete set of fits. The data set is actually a test set, it was not used to fit the model, but it was simulated from the same process as the training set, which we don’t have.
In this problem we will replicate what was done in Section 6, of https://vita.had.co.nz/papers/model-vis.pdf. Please read this first! It documents a struggle with getting the neural network to fit to a reasonably simple problem.
Read in the data, and make an appropriate plot, that communicates the relationship between the two variables and the two groups.
Read in the best model. Take a look at the object. There are three components:
nnet. The best model uses \(s=4\). The
nnet component has the estimated model coefficients, fitted values and residuals. The
hidden component has information related to the models used in the 4 nodes of the hidden layer, and the
output has the same information for the second layer. These latter two contain a grid of values for the predictors, \(x\), \(y\) and the predicted values for each grid point.
Plot the grid of predicted values for the second layer, using node 1. Overlay the data. How well has the model captured the class structure?
Plot the grid of predicted values for each node in the hidden layer, with the data overlaid. Explain how the models at each node would combine to make the final model predictions, which we have already seen are extremely good.
How many parameters are there in this model? Check that your answer matches the number of values in the
wgts element of the
Write down the equation corresponding to the model at first node of the hidden layer. You need to look at the
wgts element of the
nnet component. There are 6 sets of linear model coefficients.
OPTIONAL ADVANCED: See if you can compute the combination of the prediction on each hidden node, to get final prediction.
Read in the complete set of models fitted. There were 600 models fitted, 200 random starts for each \(s = 2, 3, 4\). The
nnet function has its own measure of the goodness of fit, which is used to determine when to stop minimising RSS, which is called
value in this data. (You can think of this like it is training error.) Plot the predictive accuracy against function’s returned value of model fit. Explain how the change in \(s\) affects the predictive accuracy.
Based on the lecture notes, explain whether the keras model can or can’t be applied to the paintings data from the past labs.
Fit the keras model to the handwritten digit data, following the code in the lecture notes. Read section 13.7 of https://bradleyboehmke.github.io/HOML/deep-learning.html#why-dl. Change your fitting code to use:
Examine the validation error vs the training error, like the plots in the chapter to compare the results for each of the changes in model structure.