move to dev parsnip

Author: dajmcdon

DESCRIPTION

@@ -44,6 +44,7 @@ VignetteBuilder:
 Remotes:
     dajmcdon/epiprocess,
     tidymodels/hardhat,
+    tidymodels/parsnip,
     tidymodels/recipes
 Config/testthat/edition: 3
 Encoding: UTF-8

musings/updated-example.Rmd

@@ -0,0 +1,71 @@
+---
+title: "Untitled"
+author: "DJM"
+date: '2022-06-06'
+output: html_document
+---
+
+```{r setup, include=FALSE}
+knitr::opts_chunk$set(echo = TRUE)
+library(tidyverse)
+library(tidymodels)
+library(epiprocess)
+library(epipredict)
+
+```
+
+```{r small-data}
+jhu <- jhu_csse_daily_subset %>%
+  filter(time_value > "2021-08-01") %>%
+  select(geo_value:death_rate_7d_av) %>%
+  rename(case_rate = case_rate_7d_av, death_rate = death_rate_7d_av)
+
+jhu_latest <- jhu %>%
+  filter(!is.na(case_rate), !is.na(death_rate)) %>%
+  group_by(geo_value) %>%
+  slice_tail(n = 15) %>% # have lags 0,...,14, so need 15 for a complete case
+  ungroup()
+```
+
+The recipe encodes how to process training/testing data. S3 object.
+
+```{r recipe}
+r <- epi_recipe(jhu) %>%
+  step_epi_lag(death_rate, lag = c(0, 7, 14)) %>%
+  step_epi_ahead(death_rate, ahead = 7) %>%
+  step_epi_lag(case_rate, lag = c(0, 7, 14)) %>%
+  step_naomit(all_predictors()) %>%
+  step_naomit(all_outcomes(), skip = TRUE)
+```
+
+The workflow combines a recipe and a model specification. Fit, estimates
+the model, adds the resulting object to the workflow.
+
+```{r workflow}
+wf <- epi_workflow(r, linear_reg()) %>% 
+  fit(jhu)
+
+wf
+```
+
+The workflow also has slots for post-processing. (Currently unimplemented.)
+
+```{r workflow2}
+names(wf) # 3 lists and a flag
+```
+
+Predict gives a new `epi_df`
+
+```{r predict}
+pp <- predict(wf, new_data = jhu_latest)
+pp
+```
+
+Can add a `forecast_date` (should be a post processing step)
+
+```{r predict2}
+predict(wf, new_data = jhu_latest, forecast_date = "2021-12-31") %>%
+  filter(!is.na(.pred))
+```
+
+