README.md
In abarciauskas-bgse/cphtbo: Mashable Popularity Predictions

cphtbo

R package for making classification predictions. Targeted at online news popularity challenge as specified for BGSE Data Science Masters Kaggle Competition

Install the package:

if (!require('devtools')) install.packages('devtools')
devtools::install_github("abarciauskas-bgse/cphtbo")
library(cphtbo)

data('news_popularity_training')
data('news_popularity_test')

train <- remove.extra.cols(news_popularity_training)
test <- news_popularity_test

# Split the data for validating the model
partitions <- training.partitions(train, p = 0.2)
data.train <- partitions$data.train
data.validation <- partitions$data.validation

# Validate the model
model <- train.randomForest(data.train)
preds <- predict(model, data.validation[,1:59])
# Accuracy on validation set
success.rate(preds, data.validation$popularity)

Replicates results from 16032016-2-final-predictions.csv submission @Wed, 16 Mar 2016 22:11:30

NOTE: This assumes you are in the root cphtbo directory, where the 16032016-2-final-predictions.csv is stored.

# Generate predictions for REAL test set
# Replicates results from 16032016-2-final-predictions.csv submission @Wed, 16 Mar 2016 22:11:30
model <- train.randomForest(train)
# Generates a new predictions file using datetime as file prefix
file <- generate.predictions.file(model, test)

Test for equality:

data('d16032016.2.final.predictions')
expected <- d16032016.2.final.predictions
new <- read.csv(file)
assert_that(all(expected[,2] == new[,2]))

cross.val Runs k-fold cross validation via rolling window.

res <- cross.val(
  model.function = train.randomForest,
  model.args = list(data.train = data.train),
  data.train = data.train)

loglik Calculates Log Likelihood from model deviance (model passed as argument must respond to deviance)

if (!require('nnet')) install.packages('nnet')
model <- multinom(popularity ~ ., data = data.train)
loglik(model)

git clone git@github.com:abarciauskas-bgse/cphtbo.git
cd cphtbo

# if loaded
# detach('package:cphtbo', unload = TRUE)

library(devtools)
# Build the package
devtools::install()
# Generate documentation
devtools::document()