Logistic Regression - Doing Data Science - page 131

fold <- c ()
# make a formula object
f = as.formula ( paste ( "Y" , "~" , paste ( vars ,
collapse = "+" )))
for ( i in c ( 1 : folds )) {
train <- data [( data $ fold != i ), ]
test <- data [( data $ fold == i ), ]
mod_x <- glm ( f , data = train , family = binomial ( logit ))
p <- predict ( mod_x , newdata = test , type = "response" )
# Get wMAE
wmae <- c ( wmae , getmae ( p , test $ Y , mae_bins ,
"dummy" , 0 ))
fold <- c ( fold , i )
auc <- c ( auc , get_auc ( p , test $ Y ))
}
return ( data.frame ( fold , wmae , auc ))
}

###############################################################
########## MAIN: MODELS AND PLOTS ##########
###############################################################
# Now build a model on all variables and look at coefficients
and model fit
vlist <- c ( "AT_BUY_BOOLEAN" , "AT_FREQ_BUY" ,

"AT_FREQ_LAST24_BUY" ,
"AT_FREQ_LAST24_SV" , "AT_FREQ_SV" , "EXPECTED_TIME_BUY" ,
"EXPECTED_TIME_SV" , "LAST_BUY" , "LAST_SV" , "num_checkins" )
f = as.formula ( paste ( "Y_BUY" , "~" , paste ( vlist ,
collapse = "+" )))
fit <- glm ( f , data = train , family = binomial ( logit ))
summary ( fit )

# Get performance metrics on each variable

vlist <- c ( "AT_BUY_BOOLEAN" , "AT_FREQ_BUY" ,

"AT_FREQ_LAST24_BUY" ,
"AT_FREQ_LAST24_SV" , "AT_FREQ_SV" , "EXPECTED_TIME_BUY" ,
"EXPECTED_TIME_SV" , "LAST_BUY" , "LAST_SV" , "num_checkins" )

# Create empty vectors to store the performance/evaluation met
rics
auc_mu <- c ()
auc_sig <- c ()
mae_mu <- c ()
mae_sig <- c ()

for ( i in c ( 1 : length ( vlist ))) {
a <- getxval ( c ( vlist [ i ]), set , 10 , 100 )
auc_mu <- c ( auc_mu , mean ( a $ auc ))
auc_sig <- c ( auc_sig , sd ( a $ auc ))
mae_mu <- c ( mae_mu , mean ( a $ wmae ))