RNN using Tensorflow in R

predict_counts.R

# 2) RNN model 생성----------------
## Refer : https://github.com/hunkim/DeepLearningZeroToAll/blob/master/lab-12-5-rnn_stock_prediction.py
library(tensorflow)
library(reticulate)

contrib <- tf$contrib

tf$reset_default_graph()

# train Parameters
seq_length <- 1
data_dim <- 1
hidden_dim <- 10
output_dim <- 1
learning_rate <- 0.01
iterations <- 100

#
CountVecs <- as.numeric(CountTS)

## 기준시점까지의 데이터로 한달 후 판매건수를 예측
DataForRNN <- data.table(Date = CountPerMonth[-141, Date], 
                         X = CountVecs[-141], 
                         Y = shift(CountVecs, n = 1, type = "lead")[-141])

DataForRNN_matX <- as.matrix(DataForRNN[, X])
DataForRNN_matY <- as.matrix(DataForRNN[, Y])

## train / test split
train_inds = 1:119
test_inds = 120:140

trainX = DataForRNN_matX[train_inds, 1] %>% as.matrix() %>% lapply(., function(x) matrix(x, nrow = 1, ncol = 1))
testX = DataForRNN_matX[test_inds, 1] %>% as.matrix() %>% lapply(., function(x) matrix(x, nrow = 1, ncol = 1))
trainY = DataForRNN_matY[train_inds, 1] %>% as.matrix()
testY = DataForRNN_matY[test_inds, 1] %>% as.matrix()

## input place holders
X = tf$placeholder(tf$float32, shape = shape(NULL, seq_length, data_dim))
Y = tf$placeholder(tf$float32, shape = shape(NULL, 1))

## build a LSTM network
cell <- contrib$rnn$BasicLSTMCell(
  num_units=hidden_dim, state_is_tuple=TRUE, activation=tf$tanh)
OutandState <- tf$nn$dynamic_rnn(cell, X, dtype = tf$float32)
outputs <- OutandState[[1]]
states_ <- OutandState[[2]]
Y_pred <- contrib$layers$fully_connected(outputs[[1]], as.integer(output_dim), activation_fn=NULL) # We use the last cell's output

# cost/loss
loss <- tf$reduce_sum(tf$square(Y_pred - Y)) # sum of the squares
# optimizer
optimizer <- tf$train$AdamOptimizer(learning_rate)
train <- optimizer$minimize(loss)

# RMSE
targets <- tf$placeholder(tf$float32, shape(NULL, 1))
predictions <- tf$placeholder(tf$float32, shape(NULL, 1))
rmse <- tf$sqrt(tf$reduce_mean(tf$square(targets - predictions)))

with(tf$Session() %as% sess, {
  init = tf$global_variables_initializer()
  sess$run(init)

  # Training Step
  for (i in 1:iterations){
    step_tmp <- sess$run(c(train, loss), feed_dict = dict(X = trainX, Y = trainY))
    step_loss <- step_tmp[[2]]
    print(cat("[step: ", i, "] loss: ", step_loss, "\n"))
  }

  # Test step
  test_predict <- sess$run(Y_pred, feed_dict = dict(X = testX))
  rmse_val = sess$run(rmse, feed_dict = dict(
    targets = testY, predictions = test_predict[[1]]))

  print(cat("RMSE: ", rmse_val))
})