I definitely see what I like & need here. The margin bands I produce & the linear correlations I produce can all be determined this way. The margin % spread band might be wrong but the neural-net or genetic-algorithm learning could determine how to adjust it and then it's a done deal. Nice. Now you mentioned the RapidAnalytics server&cron I've got more ideas to put together. last detail, is automatically downloading the datasets.
To calculate de difference between two days in RapidMiner with Generate Attributes, the functions expresses is: date_diff(First_day, Last_day)/3600/1000/24 where 3600 are the seconds, 1000 I can't understant where came from, and 24 are the hours
@MrThomasOtt from there the idea I'm at, partially with suggestions from comments on my videos, is a later handling of data so there's an input for the $ in fees and then the output should suggest the # shares to buy on the price & date and then the output chart from the sheet will show the profit target in $, in % and on the specific dates shown by the future model. It takes forever doing this by hand but my numbers are good, that much I have figured out.
Hi Thomas, really love your effort in teaching the masses. I have one question concerning both SVM's and ANN's, I can create models and use cross validation to prove how good they are at classifying datasets, however I want to use these models to against unseen data to classify them under the rules or models generated. How do I go about this?
@ytgv3fc7 I haven't tried it yet but in theory you could execute a R script (with the quantmod package) in Rapidminer to pull the data down, store it in a Rapidminer Repository and then do your analysis. This can be run via RapidAnalytics Cron. You should be able to generate BUY/SELL signals automatically with RapidAnalytics, Rapidminer and the R extension.
@MrThomasOtt I literally just "made" (not sold yet) $175 on GPR.T O but the basis of this was to correlate its moves (alongside agq / hzu.to) & compare that to the silver future-chart which itself is derived from a correlation with my gold future-chart which itself is derived from the 131-rate of change. It's been rock solid & I have to say I am happy with the result. But none of it is automated (boo). I think SVM/radial kernel-function is related to what I'm looking at as well re:margin bounds
@MrThomasOtt ah, very interesting. I'm not so hot on buy/sell signals because of how I "do things" but I'm "almost" the same. I generate % profit and $ profit outputs based on the price range, buy+sell prices known, which of course requires a future model showing the time-frame & price on that date, which is what I do & wish to automate. If I can get the data, evolve/neural-net the fit-model, same for the margin on close-price around the mid-line on each date, I can save that to spreadsheets
