fabian-s
commented
2 years ago you can get running medians or averages from tf_smooth, and then only keep the arg-vals you care about.
EDIT: but in that version, we do need a function to put the tf-object on a different domain measured in different units --> tidyfun/tf#6....
library(tidyfun)
#> Registered S3 method overwritten by 'GGally':
#> method from
#> +.gg ggplot2
#>
#> Attaching package: 'tidyfun'
#> The following objects are masked from 'package:stats':
#>
#> sd, var
library(tidyverse)
hour_data <- chf_df %>% filter(day == "Mon") |>
select(id, activity) |>
tf_unnest(activity) %>%
mutate(hour = floor((activity_arg -1) / 60)) %>%
group_by(id, hour) %>%
summarize(act_hour = mean(activity_value)) %>%
tf_nest(.id = id, .arg = hour)
#> `summarise()` has grouped output by 'id'. You can override using the `.groups`
#> argument.
hour_data2 <- chf_df %>% filter(day == "Mon") |>
select(id, activity) |>
mutate(act_hour2 =
tf_smooth(activity, method = "rollmean", k = 60, align = "right") |>
tfd(arg = seq(60, 1440, by = 60))) |>
select(-activity)
#> setting fill = 'extend' for start/end values.
ggplot(hour_data) + geom_spaghetti(aes(y = act_hour))
ggplot(hour_data2) + geom_spaghetti(aes(y = act_hour2))
Created on 2022-04-26 by the reprex package (v2.0.1)
jeff-goldsmith
if you have functions measured over a rich grid, you might want to downsample (e.g. go from minute-level wearable device data to 5 minute or one-hour increments). if that's your goal, you might prefer to average over bins -- but i don't think there's a good way to do that right now, is there?
tf_evaluatelets you evaluate on a new domain, but uses interpolation rather than averaging. andtf_integratecould work, sort of, in that you get average value betweenlowerandupper-- but it produces a scalar, and you'd have to do some kind of loop.for what it's worth, my current work around is to unnest, aggregate, then nest and re-merge. something like: