warp_fitted_model: Create a warped view of a fitted machine-learning model

In alexanderbrenning/wiml: Interpreting machine-learning models in transformed space

Create a warped view of a fitted machine-learning model

Description

This function creates the composition function \hat{f}\circ T^{-1} from a fitted machine-learning model \hat{f} and a warper object representing the transformation function T.

Usage

warp_fitted_model(x, warper)

Arguments

x	A fitted model, for example from a randomForest call. This model was fitted to the original, untransformed data.
warper	A warper object, representing an (invertible) transformation from the model's feature space to a transformed space whose axes (or base vectors) define a new perspective on the feature space and model.

Details

This function does not re-fit the model, it simply creates a data structure that jointly stores the fitted model and the warper object. The warped_model's predict (and other) methods allow the user to use a warped_model as if it had been fitted using the transformed data.

Value

An object of class warped_model, which can be used to make predictions using input data from the transformed space thanks to the predict method for this object type.

Examples

### Create principal components warper for Maipo data set
### from Peña et al. (2015):
xvars <- c(paste("ndvi0", 1:8, sep = ""), paste("ndwi0", 1:8, sep = ""),
           paste("b", outer(1:8,2:7,paste,sep = ""), sep = ""))
fo <- as.formula(paste("class ~", paste(xvars, collapse=" +" )))
d <- maipofields

sel <- seq(1, nrow(d), by = 2)
#sel <- sample(1:nrow(d), size = floor(nrow(d)/2))
dtrain <- d[sel, ]
dtest <- d[-sel, ]

wrp <- pca_warper(dtrain, xvars = xvars, yvar = "class")

fit <- randomForest::randomForest(formula = fo, data = dtrain)
fit
warped_fit <- warp_fitted_model(fit, warper = wrp)
warped_fit$fit # just a copy of fit

# Create data frame with transformed test data:
wdtest <- warp(dtest, warper = wrp)

# Prediction using the transformed data and the warped model
# using the predict.warped_model method:
wpred <- predict(warped_fit, newdata = wdtest)
# ...should give the same results as prediction using the
# untransformed features and the original model using
# randomForest's predict method:
pred <- predict(fit, newdata = dtest)
all.equal(wpred, pred)

# Misclassification error rate on the test set:
mean(wdtest$class != wpred)

# Use this warped model to visualize main effects in a tidy way:
library(iml)
predictor <- Predictor$new(warped_fit, data = wdtest, y = "class",
                           type = "prob", class = "crop3")
effs <- FeatureEffects$new(predictor, features = c("PC1","PC2","PC3","PC4"),
                           method = "ale")
plot(effs)
# What's the meaning of these principal components?
# Interpretation aid:
# https://landsat.gsfc.nasa.gov/landsat-8/landsat-8-bands
round(wrp$pca$rotation[,1:4], digits = 2)
# PC1: mid- to late season vegetation vigour and moisture
# PC2: late minus early season vegetation vigour and moisture contrast
# PC3: earliest season vegetation vigour (image date 1)
# PC4: all-season SWIR (moisture) signal (except last image date)
# See help(strucpca_warper) for a more structured approach!


# For comparison: fit RF to transformed data -> often not a good idea,
# Repeat this several times to see that OOB error is not only larger
# but especially more variable than using the untransformed data:
pcfo <- as.formula(paste("class ~", paste(wrp$wvars, collapse="+" )))
wdtrain <- warp(dtrain, warper = wrp)
pcfit <- randomForest::randomForest(formula = pcfo, data = wdtrain)
pcfit
mean(wdtest$class != predict(pcfit, newdata = wdtest))
# (but the rotation forest method makes more intelligent use of transformed
# predictors...)

alexanderbrenning/wiml documentation built on Sept. 29, 2023, 4:45 a.m.