Ch-9: Text Classification

# Deep network for text classification
library(keras)
library(syuzhet) 

# Sep 12, 2017 tweets after iphone X release
t1 <- "I'm not a huge $AAPL fan but $160 stock closes down $0.60 for the day on huge volume isn't really bearish"
t2 <- "$AAPL $BAC not sure what more dissapointing: the new iphones or the presentation for the new iphones?"
t3 <- "IMO, $AAPL animated emojis will be the death of $SNAP."
t4 <- "$AAPL get on board. It's going to 175. I think wall st will have issues as aapl pushes 1 trillion dollar valuation but 175 is in the cards"
t5 <- "In the AR vs. VR battle, $AAPL just put its chips behind AR in a big way."
tweets <- c(t1, t2, t3, t4, t5)

# Tokenize
token <- text_tokenizer(num_words = 10) %>%    
         fit_text_tokenizer(tweets)

token$word_counts %>% head()
token$index_word %>% head()
token$index_word[1:3]

token$word_index %>% head()
token$word_index[1:3]
length(token$word_index)

seq <- texts_to_sequences(token, tweets)

z <- NULL
for (i in 1:5) {z[i] <- print(length(seq[[i]]))}
summary(z)

# Padding
pad_seq <- pad_sequences(seq, maxlen = 5, padding = 'post', truncating = 'post')

# Sentiment
s <- get_nrc_sentiment(tweets) 
get_nrc_sentiment('animated') 
get_nrc_sentiment('death') 

get_sentiment(tweets, method="syuzhet")
get_sentiment(tweets, method="bing")
get_sentiment(tweets, method="afinn")

get_sentiment('bearish', method="syuzhet")
get_sentiment('bearish', method="bing")
get_sentiment('bearish', method="afinn")

# IMDB data
imdb <- dataset_imdb(num_words = 500)  
c(c(train_x, train_y), c(test_x, test_y)) %<-% imdb
z <- NULL
for (i in 1:25000) {z[i] <- print(length(train_x[[i]]))}
hist(z,
     xlab = 'Sequence Length',
     col = 'green',
     main = '')

# Padding
train_x <- pad_sequences(train_x, maxlen = 200, padding = 'post') 
test_x <- pad_sequences(test_x, maxlen = 200, padding = 'post')

# Model
model <- keras_model_sequential()
model %>% layer_embedding(input_dim = 500, 
                          output_dim = 16, 
                          input_length = 200) %>%
         layer_flatten() %>% 
         layer_dense(units = 16, activation = 'relu') %>%
         layer_dense(units = 1, activation = "sigmoid")

# Compile
model %>% compile(optimizer = "adamax",  
                  loss = "binary_crossentropy",
                  metrics = c("acc"))

# Fit model
model_1 <- model %>% fit(train_x, train_y,
                         epochs = 10,
                         batch_size = 512,
                         validation_split = 0.2)
plot(model_1)

# Prediction
model %>% evaluate(train_x, train_y)
pred <- model %>%   predict_classes(train_x)
table(Predicted=pred, Actual=imdb$train$y)

model %>% evaluate(test_x, test_y)
pred1 <- model %>%   predict_classes(test_x)
table(Predicted=pred1, Actual=imdb$test$y)