Extract Rules from Generative Grammar primitives

[AleksandrNevskiy]

Hi all, guys!!!1

I have an idea for collaboration here. We can use distributed processing for combinatorial search all together.

We have an EA, which has rule generator inside; each rule is used as a buy/sell signal and tested on history; the rule in EA is coded as an abstract form and exact definition depends on parameters.

here's a simple (first starter) EA attached, it's built up abstract rule of the form:

(V1 C1 V2 ) L1 (V3 C2 V4), where

V1,V2,V3,V4 are of set { 0O,0C,0H,0L, 1O, 1C, 1H, 1L }

in which O,C,H,L mean Open,Close,High,Low and 0 or 1 prefixes indicate price for today and price for next day respectively.

C1, C2 are of comparative operation set { >, <, =, = }

L1 is of logic operators set { AND , OR}

so we have V,C,L for variable, comparative and logic sets.

the values are coded like that

V = (from 1 to 8) ==> 8 combinations

C = (1..5) ==> 5 combinations

L = (1,2) ==> 2 combinations

and then we use parameter combinatorics to test all rule generated from digit combinations from

(1 1 1) 1 (1 1 1)meaning

(todayOpen > todayOpen) AND (todayOpen > todayOpen)

.... to ...

(8 5 8) 2 (8 5 8) meaning

(yesterdayLow = yesterdayLow) OR (yesterdayLow = yesterdayLow)

which in these concrete examples abviously doesn't make much logic sence, BUT possible useful rules are supposed to be found in the space of about 200,000 variants. Which takes about a day calculation time on my computer. That is the problem. WE ALL NEED TO COLLABORATE TO EXTRACT MORE FANCY RULES WITH NEW EAs THAT WOULD HAVE MORE ADVANCED CYNTAX RULES

having rule structure of (CVC)L(CVC) we get 8*5*8*2*8*5*8= 204,800 combinations. The next parameter I will add next is TIMEFRAME, and make it from 5 min to 500 min with step of 5, so we have

1 * 5

2 * 5 = 10

3 * 5 = 15

...

100 * 5 = 500 ==========>>> 100 combinations

204800 * 100 = 20 480 000 is the number of combinations to calculate above rule space on 100 timeframes as well, this is a huge amount of hours.

the next thing I'm gonna do is to introduce even greater cyntax change, it would be like:

1) add more historic bars : num_bars

2) make comparison members consists of more expressions : num_exp,

so we have something like: expr(1) L exp(2) ... L exp(num_expr)

and this all to bruteforce will take months, but if you, personally, can help with fast servers and interest, we can all mine super amount of useful rules!

ALL INTERESTED PEOPLE WITH ACCESS TO FAST COMPUTERS WILLING TO HELP PLS GIVE YOUR COORDINATES HERE. THANK YOU!

here's a described above demo rule-miner is attached.