drake - 映射到 ggplot 目标以输出它们
drake - map over ggplot targets to output them
首先，drake 太神奇了。我喜欢设计依赖图然后一举执行的工作流程。
可是，我运行进了一个路障。
我的工作流程是在大参数网格上进行模拟，然后汇总所述网格的不同切片。我想为每个这样的切片创建一个图。如果我理解正确，我应该使用某种形式的 cross->combine->map 来实现这一点。
这是我的：
sim_data <- function(mean, sd) {
tibble(r = rnorm(1000, mean, sd))
}
plot_dis <- function(lg, title) {
ggplot(lg) +
geom_histogram(aes(x=r, fill=sd), binwidth = 0.25) +
labs(title = str_glue("x = {title}")) +
ggsave(str_glue("{title}.pdf")) # side-effect
}
plan <- drake_plan(
data = target(
sim_data(mean = x, sd = sd),
transform = cross(x = c(10, 20, 30), sd = c(1, 2))
), # awesome
s_x = target(
bind_rows(data, .id = "sd"),
transform = combine(data, .by=x)
), # great
plot = target(
plot_dis(s_x, x),
transform = map(s_x)
) # how to add a `file_out` to this target?
)
所以我的 plot 目标有保存情节的副作用。
有一个更好的方法吗？像 plot 目标的正确 file_out 吗？
谢谢。
好问题。考虑这个实际上帮助我解决了 drake + keras.
的一些问题
如何添加file_out()s
您就快完成了，您只需要进行一些整洁的评估 (!!) 以确保每个文件名都是计划中的文字字符串。
library(drake)
drake_plan(
data = target(
sim_data(mean = x, sd = sd),
transform = cross(x = c(10, 20, 30), sd = c(1, 2))
),
s_x = target(
bind_rows(data, .id = "sd"),
transform = combine(data, .by=x)
),
plot = target(
plot_dis(s_x, file_out(!!sprintf("%s.pdf", x))),
transform = map(s_x)
)
)
#> # A tibble: 12 x 2
#> target command
#> <chr> <expr>
#> 1 data_10_1 sim_data(mean = 10, sd = 1)
#> 2 data_20_1 sim_data(mean = 20, sd = 1)
#> 3 data_30_1 sim_data(mean = 30, sd = 1)
#> 4 data_10_2 sim_data(mean = 10, sd = 2)
#> 5 data_20_2 sim_data(mean = 20, sd = 2)
#> 6 data_30_2 sim_data(mean = 30, sd = 2)
#> 7 s_x_10 bind_rows(data_10_1, data_10_2, .id = "sd")
#> 8 s_x_20 bind_rows(data_20_1, data_20_2, .id = "sd")
#> 9 s_x_30 bind_rows(data_30_1, data_30_2, .id = "sd")
#> 10 plot_s_x_10 plot_dis(s_x_10, file_out("10.pdf"))
#> 11 plot_s_x_20 plot_dis(s_x_20, file_out("20.pdf"))
#> 12 plot_s_x_30 plot_dis(s_x_30, file_out("30.pdf"))
由 reprex package (v0.2.1)
于 2019-03-26 创建
通过多一点元编程，您可以使用整个目标名称。
library(drake)
drake_plan(
data = target(
sim_data(mean = x, sd = sd),
transform = cross(x = c(10, 20, 30), sd = c(1, 2))
),
s_x = target(
bind_rows(data, .id = "sd"),
transform = combine(data, .by=x)
),
plot = target(
plot_dis(s_x, file_out(!!sprintf("%s.pdf", deparse(substitute(s_x))))),
transform = map(s_x)
)
)
#> # A tibble: 12 x 2
#> target command
#> <chr> <expr>
#> 1 data_10_1 sim_data(mean = 10, sd = 1)
#> 2 data_20_1 sim_data(mean = 20, sd = 1)
#> 3 data_30_1 sim_data(mean = 30, sd = 1)
#> 4 data_10_2 sim_data(mean = 10, sd = 2)
#> 5 data_20_2 sim_data(mean = 20, sd = 2)
#> 6 data_30_2 sim_data(mean = 30, sd = 2)
#> 7 s_x_10 bind_rows(data_10_1, data_10_2, .id = "sd")
#> 8 s_x_20 bind_rows(data_20_1, data_20_2, .id = "sd")
#> 9 s_x_30 bind_rows(data_30_1, data_30_2, .id = "sd")
#> 10 plot_s_x_10 plot_dis(s_x_10, file_out("s_x_10.pdf"))
#> 11 plot_s_x_20 plot_dis(s_x_20, file_out("s_x_20.pdf"))
#> 12 plot_s_x_30 plot_dis(s_x_30, file_out("s_x_30.pdf"))
由 reprex package (v0.2.1)
于 2019-03-26 创建
但是你真的需要文件吗？
ggplot2 对象与 drake 的缓存配合得很好。
library(drake)
library(tidyverse)
sim_data <- function(mean, sd) {
tibble(r = rnorm(1000, mean, sd))
}
plot_dis <- function(lg) {
ggplot(lg) +
geom_histogram(aes(x=r, fill=sd), binwidth = 0.25) +
labs(title = deparse(substitute(lg)))
}
plan <- drake_plan(
data = target(
sim_data(mean = x, sd = sd),
transform = cross(x = c(10, 20, 30), sd = c(1, 2))
),
s_x = target(
bind_rows(data, .id = "sd"),
transform = combine(data, .by=x)
),
plot = target(
plot_dis(s_x),
transform = map(s_x)
)
)
make(plan)
#> target data_10_1
#> target data_10_2
#> target data_20_1
#> target data_20_2
#> target data_30_2
#> target data_30_1
#> target s_x_10
#> target s_x_20
#> target s_x_30
#> target plot_s_x_10
#> target plot_s_x_20
#> target plot_s_x_30
readd(plot_s_x_10) # see also loadd()
由 reprex package (v0.2.1)
于 2019-03-26 创建
首先，drake 太神奇了。我喜欢设计依赖图然后一举执行的工作流程。
可是，我运行进了一个路障。
我的工作流程是在大参数网格上进行模拟，然后汇总所述网格的不同切片。我想为每个这样的切片创建一个图。如果我理解正确，我应该使用某种形式的 cross->combine->map 来实现这一点。
这是我的：
sim_data <- function(mean, sd) {
tibble(r = rnorm(1000, mean, sd))
}
plot_dis <- function(lg, title) {
ggplot(lg) +
geom_histogram(aes(x=r, fill=sd), binwidth = 0.25) +
labs(title = str_glue("x = {title}")) +
ggsave(str_glue("{title}.pdf")) # side-effect
}
plan <- drake_plan(
data = target(
sim_data(mean = x, sd = sd),
transform = cross(x = c(10, 20, 30), sd = c(1, 2))
), # awesome
s_x = target(
bind_rows(data, .id = "sd"),
transform = combine(data, .by=x)
), # great
plot = target(
plot_dis(s_x, x),
transform = map(s_x)
) # how to add a `file_out` to this target?
)
所以我的 plot 目标有保存情节的副作用。
有一个更好的方法吗？像 plot 目标的正确 file_out 吗？
谢谢。
好问题。考虑这个实际上帮助我解决了 drake + keras.
如何添加file_out()s
您就快完成了，您只需要进行一些整洁的评估 (!!) 以确保每个文件名都是计划中的文字字符串。
library(drake)
drake_plan(
data = target(
sim_data(mean = x, sd = sd),
transform = cross(x = c(10, 20, 30), sd = c(1, 2))
),
s_x = target(
bind_rows(data, .id = "sd"),
transform = combine(data, .by=x)
),
plot = target(
plot_dis(s_x, file_out(!!sprintf("%s.pdf", x))),
transform = map(s_x)
)
)
#> # A tibble: 12 x 2
#> target command
#> <chr> <expr>
#> 1 data_10_1 sim_data(mean = 10, sd = 1)
#> 2 data_20_1 sim_data(mean = 20, sd = 1)
#> 3 data_30_1 sim_data(mean = 30, sd = 1)
#> 4 data_10_2 sim_data(mean = 10, sd = 2)
#> 5 data_20_2 sim_data(mean = 20, sd = 2)
#> 6 data_30_2 sim_data(mean = 30, sd = 2)
#> 7 s_x_10 bind_rows(data_10_1, data_10_2, .id = "sd")
#> 8 s_x_20 bind_rows(data_20_1, data_20_2, .id = "sd")
#> 9 s_x_30 bind_rows(data_30_1, data_30_2, .id = "sd")
#> 10 plot_s_x_10 plot_dis(s_x_10, file_out("10.pdf"))
#> 11 plot_s_x_20 plot_dis(s_x_20, file_out("20.pdf"))
#> 12 plot_s_x_30 plot_dis(s_x_30, file_out("30.pdf"))
由 reprex package (v0.2.1)
于 2019-03-26 创建通过多一点元编程，您可以使用整个目标名称。
library(drake)
drake_plan(
data = target(
sim_data(mean = x, sd = sd),
transform = cross(x = c(10, 20, 30), sd = c(1, 2))
),
s_x = target(
bind_rows(data, .id = "sd"),
transform = combine(data, .by=x)
),
plot = target(
plot_dis(s_x, file_out(!!sprintf("%s.pdf", deparse(substitute(s_x))))),
transform = map(s_x)
)
)
#> # A tibble: 12 x 2
#> target command
#> <chr> <expr>
#> 1 data_10_1 sim_data(mean = 10, sd = 1)
#> 2 data_20_1 sim_data(mean = 20, sd = 1)
#> 3 data_30_1 sim_data(mean = 30, sd = 1)
#> 4 data_10_2 sim_data(mean = 10, sd = 2)
#> 5 data_20_2 sim_data(mean = 20, sd = 2)
#> 6 data_30_2 sim_data(mean = 30, sd = 2)
#> 7 s_x_10 bind_rows(data_10_1, data_10_2, .id = "sd")
#> 8 s_x_20 bind_rows(data_20_1, data_20_2, .id = "sd")
#> 9 s_x_30 bind_rows(data_30_1, data_30_2, .id = "sd")
#> 10 plot_s_x_10 plot_dis(s_x_10, file_out("s_x_10.pdf"))
#> 11 plot_s_x_20 plot_dis(s_x_20, file_out("s_x_20.pdf"))
#> 12 plot_s_x_30 plot_dis(s_x_30, file_out("s_x_30.pdf"))
由 reprex package (v0.2.1)
于 2019-03-26 创建但是你真的需要文件吗？
ggplot2 对象与 drake 的缓存配合得很好。
library(drake)
library(tidyverse)
sim_data <- function(mean, sd) {
tibble(r = rnorm(1000, mean, sd))
}
plot_dis <- function(lg) {
ggplot(lg) +
geom_histogram(aes(x=r, fill=sd), binwidth = 0.25) +
labs(title = deparse(substitute(lg)))
}
plan <- drake_plan(
data = target(
sim_data(mean = x, sd = sd),
transform = cross(x = c(10, 20, 30), sd = c(1, 2))
),
s_x = target(
bind_rows(data, .id = "sd"),
transform = combine(data, .by=x)
),
plot = target(
plot_dis(s_x),
transform = map(s_x)
)
)
make(plan)
#> target data_10_1
#> target data_10_2
#> target data_20_1
#> target data_20_2
#> target data_30_2
#> target data_30_1
#> target s_x_10
#> target s_x_20
#> target s_x_30
#> target plot_s_x_10
#> target plot_s_x_20
#> target plot_s_x_30
readd(plot_s_x_10) # see also loadd()
由 reprex package (v0.2.1)
于 2019-03-26 创建