I did the same thing. Computers can't see everything the human eye can. You can't design a quantifiable directional model unless you see what's happening over and over again. Only then will you truly understand market mechanics/price action. Many programmers, never having gambled in their life, come to ET for advice on building algorithmic models without seeing the realities like risk, slippage, scalability, odds of entry, etc.....
Real money trades and visual analysis of the structure allows you to refine system properly. At least that's how I did it. I'm not aware of any substitutes. It sure would have saved me loads of cash if I did
Substitutes.....readily available, try Support Vector Machines which can analyze chart dynamics as non-linear planes in n-dimensional hyperspace with say 2000 dimensions. The human mind can only make decisions with 7 variables at a time...anything more than that it unconsciously summarizes the weights of each measure, 7 at a time. So try looking at 2000 dimensions of measure sometime and see how far you get....even with 3 Red Bulls it is a flop. Plus a SVM model does not have poor judgement on the days it is coming down with the flu or having a fight with the wife.