---
title: "movie_sentiment"
author: "Prof. Eric A. Suess"
format: 
  html:
    embed-resources: true
---

## Classifying movie reviews: A binary classification example

Load the data.

Note: The multi-assignment operator (%<-%) from the zeallot package to unpack the list into a set of distinct variables.

```{r}
library(keras)

imdb <- dataset_imdb(num_words = 10000)

c(c(train_data, train_labels), c(test_data, test_labels)) %<-% imdb
```

Examine the data.

```{r}
str(train_data)

str(train_labels)
```

decode one of these reviews

```{r}
word_index <- dataset_imdb_word_index()

reverse_word_index <- names(word_index)

names(reverse_word_index) <- as.character(word_index)

decoded_words <- train_data[[1]] %>%
  sapply(function(i) {
    if (i > 3) reverse_word_index[[as.character(i - 3)]]
    else "?"
})

decoded_review <- paste0(decoded_words, collapse = " ")

cat(decoded_review, "\n")
```

Encoding the integer sequences via multi-hot encoding

```{r}
vectorize_sequences <- function(sequences, dimension = 10000){ 
  results <- array(0, dim = c(length(sequences), dimension))
  for (i in seq_along(sequences)) {
    sequence <- sequences[[i]]
    for (j in sequence)
      results[i, j] <- 1
    }
results
}

x_train <- vectorize_sequences(train_data)

x_test <- vectorize_sequences(test_data)
```


```{r}
str(x_train)

y_train <- as.numeric(train_labels)

y_test <- as.numeric(test_labels)
```

Define the model.  Feedforward neural network with **two** hidden layers

```{r}
model <- keras_model_sequential() %>%
  layer_dense(16, activation = "relu") %>%
  layer_dense(16, activation = "relu") %>%
  layer_dense(1, activation = "sigmoid")
```

Compile the model

```{r}
model %>% compile(optimizer = "rmsprop",
                  loss = "binary_crossentropy",
                  metrics = "accuracy")
```

Validation set

```{r}
x_val <- x_train[seq(10000), ]
partial_x_train <- x_train[-seq(10000), ]
y_val <- y_train[seq(10000)]
partial_y_train <- y_train[-seq(10000)]
```

Train the model

```{r}
history <- model %>% fit(
  partial_x_train,
  partial_y_train,
  epochs = 20,
  batch_size = 512,
  validation_data = list(x_val, y_val)
)
```

```{r}
str(history$metrics)

plot(history)
```

Refit the model using 4 epochs

```{r}
model <- keras_model_sequential() %>% 
  layer_dense(16, activation = "relu") %>%
  layer_dense(16, activation = "relu") %>%
  layer_dense(1, activation = "sigmoid")

model %>% compile(optimizer = "rmsprop",
                  loss = "binary_crossentropy",
                  metrics = "accuracy")

model %>% fit(x_train, y_train, epochs = 4, batch_size = 512)
```


```{r}
results <- model %>% evaluate(x_test, y_test)

results
```

```{r}
sentiment <- model %>% predict(x_test)
head(sentiment)
```