Great Video @NeuralMarketTrends!! In my opinion though, when it comes to information-science knowing Python or R language is more or less mandatory.. Think of having a chauffeur who drives only automatic. Sure driving automatic is the in-thing nowadays but what happens when you get a stick-shift vehicle? Exactly, no movement whatsoever. Rapidminer is the automatic shift in this case. Really comes in handy for quick touches once you know Python or R (I personally recommend Python, easier to learn)
I know I catch up & down moves into one trend using sine-waves but I'm pretty sure if you don't actually do something to calculate a sin wave the neural net and/or genetic program won't just stumble on it by accident. It's just too complicated. If you put one in, it can evolve perhaps the frequency & amplitude but even so, if it has to follow a funny curve rather than a line, be re-scaled in log-form ... so many things can screw it up. I was thinking of having log & powers computed & loaded
Great video! It really helped in building my basics. However, I wanted to know the exact use of label in windowing operator? And I got a prediction accuracy of .33, I increased my horizon and the accuracy was .5 . How come?
Trend accuracy depends on how much training data you give it in a window. For example, imagine trying to forecast auto production for the year on only two months. The model would be really bad. Train the forecast on twelve months and your accuracy will go up.
I was able to follow until... I got to 11.15 secs and the "Forecasting Performance" operator which is missing. I have RM 9.7 and "Apply Forecast" or "Forecast Validation" but no "Forecasting Performance." So what do I use as I've tried looking this up on google with no luck. At this point I get that we're training/testing but you don't really explain clearly what this Performance operator is for and what all the port connections abbreviations mean, or why we're using them? A little help would be appreciated as this is my first use of RM with a time series. Thanks.
You gotta understand that these videos are from RapidMiner version 5.0, they're like 10 years old. The new version of RM 9+ has incorporated the Time Series extension into their base platform and they likely rename the Forecasting Performance operator to something else or they changed the whole Validation scheme because it was quite confusing. The performance operator is used to assess the performance of how well your model training is fitting to your training data, so it's very important. I would review the online RM documentation for more guidance .
@@NeuralMarketTrends Thanks Thomas, right I just thought you might know the name of the modern equivalent of that RM 5.0 performance operator or process so I could finish the tutorial. Did you get better results by de-trending and using superior signal processing indicators (John Ehlers), Volatility and Random Forests, I've seen research papers that suggest this is the best way to go for financial time series data prediction?
@@allwarsarebankerwars7089 detrending time series would probably work well in things like 'weekly sales' of products. Financial time series (i.e. stock prices) usually need to be processed differently. Like I would look at a time series of the historical volatility of the asset itself, not necessarily the asset price.
@@NeuralMarketTrends Volatility yes, as a trader I’ve studied it extensively. Implied volatility is best. So just to conclude there’s no way for anyone with RM 9.7 to conclude this tutorial above because the operator no longer exists or RM changed the validation scheme for training data? That's a pity, I was enjoying the video.
@@allwarsarebankerwars7089 I don't know, I don't use RapidMiner anymore. You should check their community.rapidminer.com for any answers on the Time Series updates/changes.
When you get the prediction_trend_accuracy, what does the +/- 0.085 mean? I can't figure out if that's the standard deviation or something like confidence interval.
I don't think this video is very helpful. It's obvious that you are guessing when applying operator parameters. It would help to teach people to learn about the algorithms and parameters instead of memorizing steps. "let's put a 5 here" is not helpful. Also, have you considered any pre and post production with your videos? Do we really need to know that you are about to go to work? Do you think that 30 seconds is typical for training a neural network??
