---
title: "Defining New Scores"
output: rmarkdown::html_vignette
vignette: >
  %\VignetteIndexEntry{Defining New Scores}
  %\VignetteEngine{knitr::rmarkdown}
  %\VignetteEncoding{UTF-8}
---

```{r, include = FALSE}
knitr::opts_chunk$set(
  collapse   = TRUE,
  comment    = "#>",
  fig.width  = 7,
  fig.height = 5
)
```

```{r setup}
library(adoptr)
```



In addition to the already existing ones, `adoptr` allows the user to
implement custom scores. 
Usually, this will be done by defining a new sub-class of `ConditionalScore`.
Assume that one would be interested in the probability of early stopping
for futility. 
First we create a new class as subclass of `ConditionalScore`
```{r}
setClass("FutilityStopping", contains = "ConditionalScore")

# constructor
FutilityStopping <- function() new("FutilityStopping")
```
We only need to implement a method `evaluate()`, all other methods are
inherited from the abstract class `ConditionalScore`.
```{r}
setMethod("evaluate", signature("FutilityStopping", "TwoStageDesign"),
          function(s, design, x1, optimization = FALSE, ...) 
              ifelse(x1 < design@c1f, 1, 0)
)
```
The `optimization` flag here allows to compute scores differently during
the optimization procedure.
This is, e.g., used for the evaluation of conditional power which uses
adaptive Gaussian Quadrature for maximal precision by default but
non adaptive Gaussian Quadrature with the pre-defined integration rule
of the design object during optimization for speed.

The score can now be integrated using the `expected` method for
conditional scores
```{r p-cont}
pr_early_futility <- expected(
  FutilityStopping(), 
  Normal(), PointMassPrior(.0, 1)
)
```

and the resulting integral score can be evaluated as usual.
Consider again, the design
```{r define-design}
design <- TwoStageDesign(
    n1  = 100,
    c1f = .0,
    c1e = 2.0,
    n2_pivots = rep(150, 5),
    c2_pivots = sapply(1 + adoptr:::GaussLegendreRule(5)$nodes, function(x) -x + 2)
)

plot(design)
```

Then the value of the expected score is given by
```{r evaluate-p-cont}
evaluate(pr_early_futility, design)
```

The value is correct since it needs to conform with
```{r check}
pnorm(design@c1f)
```