Machine learning in trading: theory, models, practice and algo-trading - page 424
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
Give me an example of such a target, just curious.
target = (Close[t+k] - Open[t+1]) / Open[t+1]
feature_n = (Close[t-n] - Open[t]) / Open[t]
1) we build the exit with a peek-ahead
2) all entrances must be without peeking forward
No.
The exit should be built ONLY on bars(ticks, quotes, events) from the FUTURE (t>0), for example (Open[t+1], High[t+2], Low[t+10], Close[t+1000], Volume[t+1000000])
Entry should be built ONLY on bars (ticks, quotes, events) from the past (t<=0), for example (Open[t], High[t-1], Low[t-10], Close[t-1000], Volume[t-10000])
Many indicators have complex calculations with unobvious bar range horizon, in particular ZigZag and other "redrawing" indicators, they "see" both past and future and are not suitable for being used both as features and as targets.
No.
The exit should be built ONLY on bars(ticks, quotes, events) FROM THE FUTURE (t>0) e.g. (Open[t+1], High[t+2], Low[t+10], Close[t+1000], Volume[t+1000000])
Entry should be made ONLY on bars (ticks, quotes, events) from the past (t<=0), for example (Open[t], High[t-1], Low[t-10], Close[t-1000], Volume[t-10000])
And what is the inconsistency? Doesn't "peek-ahead" match your "FROM THE FUTURE" and "without peek-ahead" match your "FROM THE PAST"?
"Peeking forward" is not a strict statement, it is important that the past does not intersect in any way with the future(for features), and the future with the past(for tags)
That is not just"FROM THE FUTURE" but"ONLY FROM THE FUTURE" and not just"FROM THE PAST" but"ONLY FROM THE PAST".
One bar is enough (from past to future and vice versa) to get a grail on random."Peek" is not a strict statement, it is important that the past does not intersect in any way with the future (for features), and the future with the past (for tags)
That is, not just"FROM THE FUTURE" but"ONLY FROM THE FUTURE" and not just"FROM THE PAST" but"ONLY FROM THE PAST".
Nothing to argue about.... both your and my formulations are the same in meaning. Although I agree that yours is more unambiguous.
One bar is enough (from past to future and vice versa) to get a grail on random.
I don't think that's the goal anyone would set for themselves. Why do we need a past bar in the future and why predict it? It is already known. And vice versa, the future can't be known in the past and it's illogical to give it as input.
I don't think anyone would set themselves such a task. Why do we need a past bar in the future and why predict it? It is already known. And vice versa, the future can in no way be known in the past and it is illogical to give it as input.
You are not paying attention:
Many indicators have complex calculations with a non-obvious bar range horizon, in particular ZigZag and other "redrawing" indicators, they "see" both past and future and are not suitable to be used as a fixture or as a target.
target = (Close[t+k] - Open[t+1]) / Open[t+1]
feature_n = (Close[t-n] - Open[t]) / Open[t]
You're not paying attention:
Many indicators have complex calculations with no obvious horizon for using the range of bars, in particular ZigZag and other "redrawing" indicators, they "see" both past and future and are not suitable for use as chips and targets.
For example, we take the zigzag signals, process them, and save them as single reversal signals. Anyway, the Training is performed on a ready dataset, hence, no one will look into the "future". We simply need to review the architecture of the Expert Advisor and the moments of its re-learning. Which is, by the way, a separate big topic.