They call him everyday on the phone and confess to him how they trully achieve their returns (or losses). C'mon man, don't you see you are dealing with a troll?
Machine learning to me implies models with a training phase and what I meant was its too slow for high-frequency because there's no time to train "in the moment"...one can create various "signatures" and use pattern matching, I've done that in real-time facial recognition and radar/sonar systems, which use custom hardware, the concept is applicable and doable with software only
intradaybill, I believe you are a successful trader and have a good grasp of statistics, but I think you are wrong on this one. I think APS is just an implementation of Genetic Algorithms, which is an optimization/search technique over a high-dimensional space (in this case, that space is the n relationships between the past m OHLC bars), using domain-specific (trading-specific) fitness functions (profit factor, MAE, etc. I think what APS does can be reproduced using other off-the-shelf GA software/packages.
I appreciate the advice, but my question is not as much about whether to normalize my inputs as it is about how to feed my inputs (indicators) into the machine learning algorithm. For every sample, I will have the current values from my indicators as well as my target, the result I am trying to predict. For training, my target is calculated by looking ahead at my data and determining the reward:risk ratio for buying or (inversely, shorting) at the current time. In some cases, the current value of the indicator will be all I would need. But say I was using an SMA. In that case, the current value of that indicator would not be enough. I'd also want to send in a measure of how the indicator has changed over the past, say 10 periods. Some do this by sending in the same indicator, but with a longer period. Others take the momentum or slope of the current indicator and feed that into the system. But are there other common approaches? I encountered this problem as well. I think that normalization has to be done for many (most?) indicators, but in general this doesn't tell you about their direction or momentum. I thought about this first in the context of MACD and ended up trying a bunch of things, e.g. the number of periods it was moving in the same direction, the total and average normalized change over those periods (slope), the most recent (one period) normalized change, etc. Not very fancy, but all I really wanted to capture was where the indicator stood relative to the entire time frame I was analyzing, and where it had moved most recently.