group-by_benchmark.Rmd

---
title: 'Benchmark: Group-by'
author: "Marianna Foos"
date: "8/6/2019"
output: html_document
---

```{r setup, include=FALSE}
library(microbenchmark)
library(reticulate)
library(data.table)
library(vroom)
library(readr)
library(dplyr)
library(tidyr)
library(ggplot2)
library(forcats)
library(scales)
```

This file comes from the GTEx project of the Broad Institute. The file is 14G unzipped, and breaks my laptop R install immediately. I've truncated it to 20M rows, or 1.6G to save myself some frustration.

##### readr
```{r readr, cache=TRUE}
readr_tab <- read_delim("Brain_Amygdala.truncated.txt", delim = "\t")
readr_bench <- microbenchmark(readr_tab2 <- readr_tab %>%
                                group_by(gene_id) %>%
                                summarise(mean(tss_distance), sum(maf)),
               times = 5,
               unit = "s",
               setup = NULL)

print(readr_bench)
rm(readr_tab, readr_tab2)
```

##### data.table
```{r datatable, cache=TRUE}
data.table_tab <- fread("Brain_Amygdala.truncated.txt", sep = "\t")
data.table_bench <- microbenchmark(data.table_tab2 <- data.table_tab[, .(v1 = mean(tss_distance), v2 = sum(maf)), by = gene_id],
               times = 5,
               unit = "s",
               setup = NULL)
print(data.table_bench)
rm(data.table_tab, data.table_tab2)
```

##### vroom
```{r vroom, cache=TRUE}
vroom_tab <- vroom("Brain_Amygdala.truncated.txt", delim = "\t")
vroom_bench <- microbenchmark(vroom_tab2 <- vroom_tab %>%
                                group_by(gene_id) %>%
                                summarise(mean(tss_distance), sum(maf)),
               times = 5,
               unit = "s",
               setup = NULL)

print(vroom_bench)
rm(vroom_tab, vroom_tab2)
```


```{r}
dividend = 1000000000
times <- data.table("readr plus dplyr" = mean(readr_bench$time)/dividend,
           "data.table" = mean(data.table_bench$time)/dividend,
           "vroom plus dplyr" = mean(vroom_bench$time)/dividend,
           stringsAsFactors = F)
times <- times %>% 
  gather(method, time_seconds, everything())
```

```{r}
ggplot(times, aes(x = fct_reorder(method, time_seconds, .desc = TRUE), y = time_seconds)) +
  geom_col() +
  coord_flip() +
  xlab("method") +
  theme(axis.text = element_text(size = 18), axis.title = element_text(size = 16, face = "italic"))
```

```{r}
sessionInfo()
```