machine learning for stock data as image processing

machine learning for stock data as image processing

Cancelled

Job Description

Hi!

I have 1 year stock data for 1500 US equties.

My aim is to to train an algorithm to learn to identify them before they rise a lot.

I have generated images of the stock charts.
The negative folder contains images of the stock before it did not rise.
The positive folder contains images of the stock before it rose.

The images have been generated with the following script:

setwd("/home/nk/rtrading/")
stocklist <- read.csv("USD_volumefiltered_small_stocks.csv")
setwd("/home/nk/rtrading/data/small volume stocks 60min") #directory for the data files

for (s in 1:1) {
#for (s in 1:nrow(stocklist)) {
symbol <- stocklist[s,]
setwd("/home/nk/rtrading/data/small volume stocks 60min") #directory for the data files

getSymbols.csvnk4(symbol) #Loads and assigns the data to symbol. for comma delimited-files

#stock(symbol,currency="USD",multiplier=1) # This one fails
#contract <- twsEquity(symbol,"SMART") # This one warns me - neither appear necessary for this example
data <- get(symbol)
data <- to.daily(data)
#colnames(data) <- "Close"
data$roc <- ROC(Cl(data))

movsum <- function(x,n=10){filter(x,rep(1,n), sides=1)}
data$cumsum <- sapply(data$roc, movsum)

# Skip any symbols with less than 41 days of records
if (nrow(data) < 61) {
next}

for (i in 51:(nrow(data)-10)) {
if(data$cumsum[i] < 0.2) {
setwd("/home/nk/rtrading/machineplots/negatives")
file <- paste(symbol, i, ".png", sep="_")
png(filename=file)
plot(chart_Series(data[(i-50):(i-10),]))
dev.off()
Sys.sleep(0)
}#if

if(data$cumsum[i] > 0.2) {
setwd("/home/nk/rtrading/machineplots/positives")
file <- paste(symbol, i, ".png", sep="_")
png(filename=file)
plot(chart_Series(data[(i-50):(i-10),]))
dev.off()
Sys.sleep(0)
}#if

} #nrow data

}#symbolloop



The images are attached in a rar file. And they are only for one stock, I have 1500 more.

My suggestion is we generate images for the rest of the 1500 stocks, use 1000 to train on and 500 to test on.
It is very possible the rules and criteria to generate images has to be altered, also some other features can be engineered if needed. Im not an expert in machine learning although I understand a few things about it.

When applying, please specify which machine learning algorithms, programming-languages or implementations you propose to use, just so I see you know your field.

To summarise:
1. Use machine learning and image processing on stock charts to identify stocks before they rise.
2. Other rules are quite flexible, maybe more history has to be in the stock-charts or something else has to be altered.

I suggest 30% payment for just completing the work and 100% payment if we achieve the goal of having the algorithm trained to identify rising stocks before it happens.

Open Attachment

Other open jobs by this client