Uniform sampling

regress_with_uniform.R

# Get a subset of values from the training set where days spent is zero
temp <- subset(train, DaysInHospital_Y2 == log1p(0))

# Get 100 samples from it with replacement and add it to a new dataframe
uniform_set <- temp[sample(nrow(temp), 100, replace=TRUE),]

# Keep on doing this for other values as well 
uniform_set <-  merge(uniform_set, temp[sample(nrow(temp), 100, replace=TRUE),], all=T)

# Train a tree on it 
uniform_tree <- tree(DaysInHospital_Y2 ~ ., uniform_set)
plot(uniform_tree,type="uniform"); text(uniform_tree,pretty=0)

# Sample a smaller set from the old training set
uniform_test <- train[sample(nrow(train), 400, replace=T),]

# Predict values
result <- predict(uniform_tree, uniform_test, type="vector")

# Append the result
uniform_test$predicted <- result

# Calculate the RMSE
sqrt(mean(log1p(uniform_test$DaysInHospital_Y2) - test$predicted)^2))