deepgp (version 1.2.1)

plot: Plots object from deepgp package

Description

Acts on a gp, gpvec, dgp2, dgp2vec, dgp3, or dgp3vec object. Generates trace plots for outer log likelihood, length scale, and nugget hyperparameters. Generates plots of hidden layers for low dimensions or monotonic warpings. Generates plots of the posterior mean and estimated 90% prediction intervals for one-dimensional inputs; generates heat maps of the posterior mean and point-wise variance for two-dimensional inputs.

Usage

# S3 method for gp
plot(x, trace = NULL, predict = NULL, ...)
# S3 method for gpvec
plot(x, trace = NULL, predict = NULL, ...)
# S3 method for dgp2
plot(x, trace = NULL, hidden = NULL, predict = NULL, ...)
# S3 method for dgp2vec
plot(x, trace = NULL, hidden = NULL, predict = NULL, ...)
# S3 method for dgp3
plot(x, trace = NULL, hidden = NULL, predict = NULL, ...)
# S3 method for dgp3vec
plot(x, trace = NULL, hidden = NULL, predict = NULL, ...)

Arguments

x

object of class gp, gpvec, dgp2, dgp2vec, dgp3, or dgp3vec

trace

logical indicating whether to generate trace plots (default is TRUE if the object has not been through predict)

predict

logical indicating whether to generate posterior predictive plot (default is TRUE if the object has been through predict)

...

N/A

hidden

logical indicating whether to generate plots of hidden layers (two or three layer only, default is FALSE)

Details

Trace plots are useful in assessing burn-in. If there are too many hyperparameters to plot them all, then it is most useful to visualize the log likelihood (e.g., plot(fit$ll, type = "l")).

In one dimension, hidden layer plots show 100 evenly distributed samples. In two dimensions (two layer only), hidden layer plots show 3 samples. Hidden layer plots are colored on a gradient - red lines represent earlier iterations and yellow lines represent later iterations - to help assess burn-in of the hidden layers.

Examples

# See ?fit_one_layer, ?fit_two_layer, or ?fit_three_layer
# for examples