nn_evaluate_jordan_ema.R

require("RCurl")
require("quantmod")
require("TTR")
require("jsonlite")
require("RSNNS")



# Read from stdin - need to find a pattern to stop reading?

f <- file("stdin")
open(f)
while(TRUE) {
  line <- readLines(f,n=1)

  streamRow <- fromJSON(line)
  

  historical <- getURL(paste0('http://localhost:8080/gemfire-api/v1/queries/adhoc?q=SELECT%20DISTINCT%20*%20FROM%20/Stocks%20s%20ORDER%20BY%20entryTimestam%20desc%20LIMIT%20100'))

  historicalSet <- fromJSON(historical)
  historicalSet <- historicalSet[order(historicalSet$entryTimestamp),]

  dataset <- subset(historicalSet, select = c("DaysHigh", "DaysLow", "LastTradePriceOnly")) 
  names(dataset) <- c("High","Low","Close")

  #Add new row to the end of historical dataset for computing technical indicators.
  temprow <- matrix(c(rep.int(NA,length(dataset))),nrow=1,ncol=length(dataset))
  newrow <- data.frame(temprow)
  colnames(newrow) <- colnames(dataset)
  dataset <- rbind(dataset,newrow)
  dataset[nrow(dataset),] <- c(as.numeric(streamRow$DaysHigh), as.numeric(streamRow$DaysLow), as.numeric(streamRow$LastTradePriceOnly))
  

  # Computing and adding the change column
  #originalSet <- dataset
  #dataset <- originalSet[-1:-50,]
  #dataset$Change <- diff(originalSet$Close, lag=50) # applies lag to the change calculation (here we're trying to predict the change within 50 iterations)

  # Remove the first X lines (x=3 here)  to avoid NAs due to the lag
  #lag_dataset <- lag_dataset[-1:-3,]

  # include technical indicators
  ema <- EMA(dataset$Close) # lag = n-1 (default=9)
  ema_diff <- dataset$Close - ema # lag = above
  rsi <- RSI(dataset$Close) # lag = n (default=14)
  smi <- SMI(HLC(dataset))     # lag = nSlow+nSig (default=34)
  sar <- SAR(HLC(dataset))     # lag = 0

  high_diff = dataset$High-dataset$Close
  low_diff = dataset$Close-dataset$Low

  inputs <- data.frame(dataset$Close, ema, ema_diff, rsi, smi, sar, high_diff, low_diff)
  names(inputs) <- c("close", "ema", "ema_diff", "rsi", "smi","sar", "high_diff", "low_diff")
 
  #remove extra NAs due to technical indicator lags
  inputs <- inputs[-1:-35,]
  dataset <- dataset[-1:-35,]


  normalized <- normalizeData(subset(inputs, select = c(close, ema)))
  to_predict <- normalized[nrow(normalized),] # we'll predict based on the last value 
 

  load(file='mynet_jordan.RData')
  results <- predict(jordannet, to_predict) # should be an input without response column

  results <- denormalizeData(x=results, getNormParameters(normalized))    

  streamRow$predictedPeak <- results[1,1]

 
#  inputWithPrediction=streamRow[1,]

  predicted_line <- data.frame(streamRow$entryTimestamp, inputs$ema[nrow(inputs)], inputs$close[nrow(inputs)], streamRow$predictedPeak);
  names(predicted_line) <- c("entryTimestamp", "ema", "close","predictedPeak")
  predicted_line <- toJSON(predicted_line);

  cat (predicted_line)

#  cat("\nForecasting for input: ",streamRow$LastTradePriceOnly,"\n")
  #cat("\nForecasted ",results, "\n")
  #write('Done',stdout())


  write('\r\n',stdout())
  #write(jsonStr,stderr())
  # process line
}
close(f)
# Remember to close file