In JiaxiangBU/add2deployment: R functions for translating R to SQL for deployment.

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)

There are three kind of variables

1. factor variable
2. continous variable
3. binary variable : data type is dbl in data.frame but there are only two distinct value (like 1,0).

library(tidyverse)
set.seed(123)
sample_data <- 
cbind(
    matrix(
        c(
            1:100
        )
        ,nrow = 10
        ,byrow = F
    )
    ,matrix(
        c(
            letters[1:20]
        )
        ,nrow = 10
        ,byrow = T
    )
    ,matrix(
        c(
            rep(0:1,5)
        )
        ,nrow = 10
        ,byrow = T
    )
) %>% 
    `colnames<-`(paste0('v',1:13)) %>% 
    `rownames<-`(1:10) %>% 
    as.data.frame() %>% 
    mutate_at(vars(1:10,13),as.integer) %>% 
    mutate_at(vars(1:10),~cut_number(.,3))
sample_data
library(caret)
sample_data_after_onehot <- 
    sample_data %>% 
    predict(dummyVars(~.,data=.),newdata =.) %>%
    as.data.frame()
sample_data_after_onehot
library(irlba)
pca_score <- prcomp_irlba(sample_data_after_onehot,n=2,center = T,scale. = F)

1. sample_data is discretized for all continuous variable by cut_number.
2. Here I use prcomp_irlba instead of prcomp partly because prcomp is hard to train big data.
3. dummyVars function I use is to one-hot encoding the category variables.

r_output <- 
    pca_score$rotation %>% 
    `rownames<-`(sample_data_after_onehot %>% names) %>% 
    as.data.frame() %>% 
    rownames_to_column('variable')
r_output

r_output is what I get when I finish the pca model.

Now, we recode it in SQL style.

library(add2deployment)
trans_inSQL(input=r_output)

JiaxiangBU/add2deployment documentation built on May 21, 2019, 10:16 a.m.