Version3Output.rmd

---
title: "R Notebook"
output: html_notebook
---

```{r}
source("dtm2.r")
twittertrain <- read.csv("train.csv")
twittertest <- read.csv("test.csv")

tweetall <- c(twittertrain$tweet,twittertest$tweet)

dtm <- dtmexp(tweetall)
# mean <- apply(dtm, 2, mean)
# std <- apply(dtm, 2, sd)
# dtm <- scale(dtm, center = mean, scale = std)

trainrows <- dim(twittertrain)[1]

training.x <- dtm[1:trainrows, ]
training.x <- data.matrix(training.x)

testrowstart <- trainrows+1
testrows <- trainrows+dim(twittertest)[1]

testing.x <- dtm[testrowstart:testrows, ]
testing.x <- data.matrix(testing.x)

training.y <- as.numeric(twittertrain$sentiment)
#testing.y <-  as.numeric(twittertest$sentiment)
```
```{r}
dim(training.x)
dim(testing.x)


#testing.y
```



```{r}
# library(caTools)
# 
# 
# 
# set.seed(123)
# spl <- sample.split(twitter$sentiment,SplitRatio = 0.7)
# 
# training.x <- subset(dtm,spl==TRUE)
# training.x <- data.matrix(training.x)
# 
# testing.x <- subset(dtm,spl==FALSE)
# testing.x <- data.matrix(testing.x)
# 
# training.y <- as.numeric(subset(twitter$sentiment,spl==TRUE))
# testing.y <-  as.numeric(subset(twitter$sentiment,spl==FALSE))
```

```{r}
library(keras)
model <- keras_model_sequential() %>%
  layer_dense(units = 20, activation = "relu", input_shape = NULL) %>%
  layer_dense(units = 15, activation = "relu") %>%
  layer_dense(units = 4, activation = "softmax")

model %>% compile(
  optimizer = "SGD",
  loss = "sparse_categorical_crossentropy",
  metrics = c("accuracy")
)

history <- model %>% fit(
  training.x,
  training.y,
  epochs = 100,
  batch_size = 100,
)
```


```{r}
testing.predict <- predict_classes(model, x = testing.x, verbose = 1)
twittertest['sentiment'] <- testing.predict
twittertest <- subset (twittertest, select = -tweet)
write.csv(twittertest, "output_dtmmod3.csv", row.names = FALSE)

```