Visualization.Rmd

---
title: 'Visualization: The Plots Thicken'
author: "Augustina Ragwitz"
output: html_document
params:
  data_folder: "downloads"
---

```{r setup, include=FALSE}
knitr::opts_chunk$set(echo = TRUE)
```

```{r includes}

library(here)
library(keras)
library(tidyverse)

library(lime)
```

# Mission

Explore ways of natively visualizing models in R!

 * Visualize layer content and structure
 * Generate heatmaps to see how the convnet learned

# Generate a Heatmap to see how the Convnet Learned

## Use a 2D Convnent

Let's use a 2d convnent for MNIST instead of dense layers.

```{r}
mnist <- dataset_mnist()
train_images <- mnist$train$x
train_labels <- mnist$train$y

# TODO improve

# TODO show adding a layer from another model 
```

# Show what parameters impacted the prediction with LIME

# Visualize Layer and Model Content

## Greta
Greta is a package to do Bayesian statistics via TensorFlow

https://greta-dev.github.io/greta/

https://cran.r-project.org/web/packages/greta/vignettes/get_started.html

## Diagrammer

http://rich-iannone.github.io/DiagrammeR/

Or Greta


## iGraph

I tried and failed :(