Which timeframe is best for SmallCatZzFractal v1.1 ??
Fractals, ZZ & the Pissing Dog
- #227
- Jan 13, 2014 4:32pm Jan 13, 2014 4:32pm
@ gg53
very good.
I performed all the steps.
1 - get set1
2 - get set2
3 - set1 filter to get set3
4 - Calculation of fractal pattern match according to the pdf
5 - In case 4 is accepted, probability calculation in length as you indicated in post 204.
6 - if 5 to be accepted, calculating probability of "next bar" as pdf.
The result:
The last signal is very bad.
I checked the database several times, I tried to apply logic how you indicated in several posts, but signals still appear far from the turning points.
I've missed some step?
I would like to know how to calculate probability "next bar" and what percentages apply to the signal "next bar"
100% next bar
100% actuall bar?
I also like to know if there is any calculation we do differently or do not do and therefore gives us these bad signals. The document assure us post 188 if the logic is correct or have an error.
thank you very much
very good.
I performed all the steps.
1 - get set1
2 - get set2
3 - set1 filter to get set3
4 - Calculation of fractal pattern match according to the pdf
5 - In case 4 is accepted, probability calculation in length as you indicated in post 204.
6 - if 5 to be accepted, calculating probability of "next bar" as pdf.
The result:
The last signal is very bad.
I checked the database several times, I tried to apply logic how you indicated in several posts, but signals still appear far from the turning points.
I've missed some step?
I would like to know how to calculate probability "next bar" and what percentages apply to the signal "next bar"
100% next bar
100% actuall bar?
I also like to know if there is any calculation we do differently or do not do and therefore gives us these bad signals. The document assure us post 188 if the logic is correct or have an error.
thank you very much
- #236
- Edited 3:59am Jan 17, 2014 2:09am | Edited 3:59am
- | Membership Revoked | Joined Oct 2009 | 1,495 Posts
Adaptability of the model:
(Real-time Chart posted in previous thread)
12/15 "next bar opposite" are correct.
Also note that most of the "misses" are after large bars movement (2-4 bars "miss").
Those are signals (1, 10, 12) counting from left to right.
Even the "misses" seems to have some logic according to my algo:
The historical pattern expected more bars to achieve the total movement - but something in the market caused the bars to be larger and achieve the target in less moves (length/bars).
Or this is what I like to think... ;-}
Actual ZZ leg, pattern and length of points/signals (1, 10, 12) are entered into the database of historical data (with their relevant date/time).
The ACTUAL ZZ, fractals and length - not the signaled ones.
You just add a new bar DNA, fractals and length of the "missed" prediction in the "missed" ZZ leg into the historical database in their relevant places.
This is done in hope that it will increase the probability of identifying such patterns/length in the future. Only future predictions will take those changes into accout - because of the date/time signature mentioned earlier).
It's not perfect "adaptability" but that's what I'm currently doing. It might change if and when I'll find some other logic.
I finished (?) testing pseudo real-time updating of the model (i.e. constant updating of historical data).
As I thought, it causes two "bad" side effects:
1. On (rare) ocassions it causes "repainting" or double signals if the previous one stays on screen. -->> B A D
2. In some more cases it causes the logic of alternating signals (which is my base point and purpose) to "fail" - i.e. two sequences of UP's or two DOWN's -->> B A D
I know that the next coming post will be "update/train the model periodically or weekly".
That's the lazy's man way of doing things...
I very much like to be lazy, but I rarely like the outcome...
I went back to the original method of "error" prediction update the database and recalculate.
G.
(Real-time Chart posted in previous thread)
12/15 "next bar opposite" are correct.
Also note that most of the "misses" are after large bars movement (2-4 bars "miss").
Those are signals (1, 10, 12) counting from left to right.
Even the "misses" seems to have some logic according to my algo:
The historical pattern expected more bars to achieve the total movement - but something in the market caused the bars to be larger and achieve the target in less moves (length/bars).
Or this is what I like to think... ;-}
Actual ZZ leg, pattern and length of points/signals (1, 10, 12) are entered into the database of historical data (with their relevant date/time).
The ACTUAL ZZ, fractals and length - not the signaled ones.
You just add a new bar DNA, fractals and length of the "missed" prediction in the "missed" ZZ leg into the historical database in their relevant places.
This is done in hope that it will increase the probability of identifying such patterns/length in the future. Only future predictions will take those changes into accout - because of the date/time signature mentioned earlier).
It's not perfect "adaptability" but that's what I'm currently doing. It might change if and when I'll find some other logic.
I finished (?) testing pseudo real-time updating of the model (i.e. constant updating of historical data).
As I thought, it causes two "bad" side effects:
1. On (rare) ocassions it causes "repainting" or double signals if the previous one stays on screen. -->> B A D
2. In some more cases it causes the logic of alternating signals (which is my base point and purpose) to "fail" - i.e. two sequences of UP's or two DOWN's -->> B A D
I know that the next coming post will be "update/train the model periodically or weekly".
That's the lazy's man way of doing things...
I very much like to be lazy, but I rarely like the outcome...
I went back to the original method of "error" prediction update the database and recalculate.
G.
- #237
- Edited 3:55am Jan 17, 2014 2:19am | Edited 3:55am
- | Membership Revoked | Joined Oct 2009 | 1,495 Posts
Look what happend minutes after last prediction...(previous post (http://www.forexfactory.com/showthread.php?p=7218910#post7218910))
I'm proud of this baby...
It was real-time...
G.
I'm proud of this baby...
It was real-time...
G.