---
title: "Process Information and Channel Ablation"
output: rmarkdown::html_vignette
vignette: >
  %\VignetteIndexEntry{Process Information and Channel Ablation}
  %\VignetteEngine{knitr::rmarkdown}
  %\VignetteEncoding{UTF-8}
---

```{r setup, include=FALSE}
knitr::opts_chunk$set(collapse = TRUE, comment = "#>")
library(eyeprocess)
```

## Posterior uncertainty as measurement value

`process_information()` compares matched posterior targets without assuming
that heterogeneous Fisher information components can simply be added.

```{r}
set.seed(10)
base <- matrix(rnorm(6000, sd=1), ncol=3)
aug  <- matrix(rnorm(6000, sd=.8), ncol=3)
colnames(base) <- colnames(aug) <- c("A","B","C")
info <- process_information(base, aug, metric="entropy_reduction")
info
plot(info)
```

## Channel ablation

```{r}
sim <- simulate_multimodal_irt(n_person=50,n_item=8,seed=5)
abl <- ablate_multimodal_channels(sim$measurement)
names(abl$scenarios)
```

Ablation is an inferential comparison design, not automatically a causal
counterfactual.