missing fn and helper var definitions

Author: nmdefries

.gitignore

@@ -28,3 +28,5 @@ _site
 .Rdata
 
 /.quarto/
+
+/_*.local

slides/day1-afternoon.qmd

@@ -2432,4 +2432,4 @@ cowplot::plot_grid(p1, p2)
 * Group built [`{rtestim}`](https://dajmcdon.github.io/rtestim) doing for this nonparametrically.
 
 * We may come back to this later...
--->
+-->

slides/day1-morning.qmd

@@ -401,34 +401,6 @@ weekly_snapshots |>
 
 **Example**: A signal based on insurance claims may take several days to appear as claims are processed
 
-<!--
-## Latency and revision in signals - Example
-
-* Recall the first example of panel & versioned data we've seen...
-
-```{r latency-ca-june1}
-#| echo: false
-lat_ca_june1 <- dv_ca |>
-  filter(time_value == "2020-06-01") |>
-  summarise(min(version - time_value)) |>
-  pull() |> as.numeric()
-```
-
-* On June 1, this signal is subject to [**revision**]{.primary} over time (ex. consider Dec. 1's `percent_cli` across `version`):
-
-```{r revision-ca-ex}
-#| echo: false
-head(x_dt_with_diff) |> as_tibble()
-```
-
-and `r lat_ca_june1` days [**latent**]{.primary}: `min(version - time_value)`
-
-```{r latency-ca-ex}
-#| echo: false
-x_dt_with_diff <- dv_ca |> mutate(version_time_diff = version - time_value)
-head(x_dt_with_diff |> group_by(time_value) |> slice(1) |> ungroup())
-```
--->
 
 <!--
 ## Revision triangle, Outpatient visits in WA 2022
@@ -508,7 +480,7 @@ A forecast that is made today can only use data we have access to today
 
 1. [**Sources that don't revise**]{.primary} (provisional and final are the same)
 
-Facebook Survey and Google symptoms
+Google symptoms
 
 2. [**Predictable revisions**]{.secondary}
 

slides/day2-afternoon.qmd

@@ -647,6 +647,41 @@ pred_arx_hosp |>
   theme(legend.title = element_blank())
 ```
 
+```{r function errors}
+MSE <- function(truth, prediction) {
+  mean((truth - prediction)^2)}
+
+MAE <- function(truth, prediction) {
+  mean(abs(truth - prediction))}
+
+MAPE <- function(truth, prediction) {
+  100 * mean(abs(truth - prediction) / truth)}
+
+MASE <- function(truth, prediction) {
+  100 * MAE(truth, prediction) / mean(abs(diff(truth)))}
+
+getErrors <- function(truth, prediction, type) {
+  return(data.frame(#"MSE" = MSE(truth, prediction), 
+                    "MAE"= MAE(truth, prediction), 
+                    #"MAPE" = MAPE(truth, prediction), 
+                    "MASE" = MASE(truth, prediction), 
+                    row.names = type))
+}
+
+getAccuracy = function(finalized, predictions, row_name = "") {
+  observed = (finalized |> 
+                filter(time_value %in% predictions$target_date))$deaths
+  return(cbind(
+    getErrors(observed, predictions$.pred, ""),
+    "Coverage" = mean(observed >= predictions$`0.1` & observed <= predictions$`0.9`), 
+    row.names = row_name))
+}
+```
+
+```{r error-arx}
+getAccuracy(ca, pred_arx)
+```
+
 Quite different predictions!
 
 ## Customizing `arx_forecaster`
@@ -676,7 +711,7 @@ pred_arx_hosp <- covid_archive_dv |> epix_slide(
 
 ```{r arx-with-more-lags-plot}
 #| fig-align: left
-pred_arx_more_lags |>
+pred_arx_hosp |>
   ggplot(aes(target_date, .pred)) +
   geom_line(data = ca, aes(x = time_value, y = deaths), inherit.aes = FALSE, col = base) +
   geom_line(col = primary) +
@@ -689,7 +724,7 @@ pred_arx_more_lags |>
 ```
 
 ```{r error-arx-more-lags}
-getAccuracy(ca, pred_arx_more_lags)
+getAccuracy(ca, pred_arx_hosp)
 ```
 
 
@@ -768,6 +803,12 @@ Color corresponds to `forecast_date`
 
 ## Geo-pooling
 
+```{r}
+#| echo: false
+h <- 28         # horizon
+w <- 120 + h    # trailing window length
+```
+
 ```{r arx-geo-pooling}
 #| echo: true
 #| code-line-numbers: "1"