Using Prophet Package to Predict By Group in Dataframe in R

Question

I am using the new package released from Facebook called Prophet. It does time series predictions and I want to apply this function By Group.

Scroll down to R Secti

Answer 1

Here is a solution using tidyr::nest to nest the data by group, fit the models in those groups using purrr::map and then retrieving the y-hat as requested. I took your code, but incorporated it into mutate calls that would compute new colums using purrr::map.

library(prophet)
library(dplyr)
library(purrr)
library(tidyr)

d1 <- df %>% 
  nest(-group) %>% 
  mutate(m = map(data, prophet)) %>% 
  mutate(future = map(m, make_future_dataframe, period = 7)) %>% 
  mutate(forecast = map2(m, future, predict))

Here is the output at this point:

d1
# A tibble: 2 × 5
   group              data          m                future
                                   
1      A   
2      B   
# ... with 1 more variables: forecast 

Then I use unnest() to retrieve the data from the forecast column and select the y-hat value as requested.

d <- d1 %>% 
  unnest(forecast) %>% 
  select(ds, group, yhat)

And here is the output for the newly forecasted values:

d %>% group_by(group) %>% 
  top_n(7, ds)
Source: local data frame [14 x 3]
Groups: group [2]

           ds  group      yhat
             
1  2016-11-30      A 180.53422
2  2016-12-01      A 349.30277
3  2016-12-02      A 288.68215
4  2016-12-03      A 222.33501
5  2016-12-04      A 342.96654
6  2016-12-05      A 203.64625
7  2016-12-06      A 185.37395
8  2016-11-30      B 131.07827
9  2016-12-01      B 222.83703
10 2016-12-02      B 236.33555
11 2016-12-03      B 145.41001
12 2016-12-04      B 228.59687
13 2016-12-05      B 162.49244
14 2016-12-06      B  68.44477