In craiggrabowski/predictr: Objects For Fast Predictions Of Common Model Types

devtools::load_all()
knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  fig.path = "README-"
)

predictr

The goal of predictr is to allow for model predictions which minimize memory usage and computing time. This package provides a new type of object and adaptors for various common model types.

Installation

You can install predictr from github with:

# install.packages("devtools")
devtools::install_github("craiggrabowski/predictr")

Example

This is a basic example which illustrates predictor(), the main method defined in this package:

## basic example code
v <- seq(1024)
df <- data.frame(x = v, y = v + 1)
model <- lm(y ~ x, df)
x <- predictor(model)
predict(x, data.frame(x = c(2, 3)))

We create a model object using linear regression and then convert to a new object using predictor(). This new object implements the predict() method, but its implementation does not produce the same results as the original regression object. In particular, the output is now expressed as a matrix instead of a numeric vector.

The new object created by predictor() has a few advantages over the object produced using a call to lm(). First, the new object is significantly smaller than the existing model object:

pryr::object_size(model)
pryr::object_size(x)

Second, the predict() method is now faster for large data:

a <- data.frame(x = seq(100000))
microbenchmark::microbenchmark(predict(model, a), predict(x, a))

Third, the new object supports a conversion to a function using as.function(). This return value of this converion is a function which implements predict() as well, only now the compute time is much smaller:

f <- as.function(x)
microbenchmark::microbenchmark(predict(x, a), f(a))

craiggrabowski/predictr documentation built on May 28, 2019, 7:48 p.m.