DELETE - IT'S IMPLEMENTED ALREADY

DELETE - IT'S IMPLEMENTED ALREADY
Attachments
  • Votes +6
  • Project StrategyQuant X
  • Type Feature
  • Status New
  • Priority Normal

History

k
#1

Karish

31.03.2021 18:02

Task created

k
#2

Karish

31.03.2021 18:03
Voted for this task.
JK
#3

Insanity82007

31.03.2021 23:09
Voted for this task.
AS
#4

alfredsylejmani

01.04.2021 00:57
Voted for this task.
b
#5

bentra

01.04.2021 17:52
To be clear, are you talking about simply making them not appear in the RESULTS tab if they fail the neighbor test? Or you want them to die off from the gene pool and or have lower fitness when they fail the neighbor test? 
b
#6

bentra

01.04.2021 18:59
Also are you talking about checking each variable one at a time? so that 10 steps on TP changes + 10 steps on SL changes + 10 steps on MA changes = 30 iterations total? Or you want all combos checked = 1000 iterations? Or a montecarlo of all combos = x iterations?

In dev4 you can already run a montecarlo crosscheck during building and calculate the gene fitness by  any confidence level of the montecarlo....
So you could already during building do a montecarlo with strategy parameter changes. Somebody could just do the montecarlo script to check each variable one at a time I think if that's what you want.

IH
#7

clonex / Ivan Hudec

01.04.2021 22:48
In dev4 you can already run a montecarlo crosscheck during building and calculate the gene fitness by  any confidence level of the montecarlo....
So you could already during building do a montecarlo with strategy parameter changes. Somebody could just do the montecarlo script to check each variable one at a time I think if that's what you want.


wow didn't knew it. 



k
#8

Karish

01.04.2021 22:50
Bentra,

What i want to achieve is a function that does those specific things and we can manipulate it further:

Lets say the builder came up with those parameters of a strategy you wanted it to find:


If Close[1]>MA 100 == LONG + SL=100 TP=100


So we got 3 parameters that we want to change (lets exclude the Bar Number for this example.)


We can have few ways to check it while it is still DID NOT entered our databank, the builder just came up with those parameters right now,

Method #1:

Check 25% with Steps of 5% of each parameter separately, while keeping the others as original values, so for example first the MA 100 would be changed to 75,80,85,90,95,100,105,110,115,120,125, While the other 2 parameters of SL and TP would stay as they are, which is 100 and 100, if those neighbors passed our criteria and the strategy seems to be the same (in terms of R-Squared for example) then move to the next parameter: SL now the MA would stay as the original value which is 100, and we will now focus on the SL parameter: 75,80,85,90,95,100,105,110,115,120,125, Passed?, great we set it back to the original value and move on to the next parameter which is TP: 75,80,85,90,95,100,105,110,115,120,125, Passed?, this means we have all the neighbors checked and they all passed our criteria, now the engine will add the OUT OF SAMPLE data, Passed?, Save to the Databank.


_ So what we achieved here is we checked all the parameters step by step: MA=CHECKING (75,80,85,90,95,100,105,110,115,120,125) SL=100 TP=100 MA=100 SL=CHECKING (75,80,85,90,95,100,105,110,115,120,125) TP=100 MA=100 SL=100 TP=CHECKING (75,80,85,90,95,100,105,110,115,120,125) Out of sample validation, Save to databank if passed.


Method #2:

We do everything like explained above with ALL parameters optimized at once. MA+SL+TP (75) MA+SL+TP (80) MA+SL+TP (85) MA+SL+TP (90) MA+SL+TP (95) MA+SL+TP (100) MA+SL+TP (105) MA+SL+TP (110) MA+SL+TP (115) MA+SL+TP (120) MA+SL+TP (125) Out of sample validation, save to databank if passed.




Thats it,
We can use it as an MC also, but it would be preferred to have some more in-depth statistics on it,
And maybe some fitness that would stimulate the genetic algorithm to pass this test.

Thank you for your time reading this, this is very impotent and i noticed many known individuals use those methods manually,
if we can do it automatically thats powerful.

Kevin Davey, Andrea Unger, Larry Williams, etc.

here are some examples:

https://youtu.be/q_02_zeFq4w


https://youtu.be/-hRrfnVDo9Y


https://youtu.be/KtUZg0kn7lg


https://youtu.be/PaVIbaTRO7o


https://youtu.be/YLLqU_F7hms


While there are many more example of those, we can even do this with any trading platform that offers backtesting and optimization already... but SQ is just better and missing this feature currently..


Thanks for your attention, hope it gets implemented asap..






k
#9

Karish

01.04.2021 22:54
In dev4 you can already run a montecarlo crosscheck during building and calculate the gene fitness by  any confidence level of the montecarlo....

So you could already during building do a montecarlo with strategy parameter changes. Somebody could just do the montecarlo script to check each variable one at a time I think if that's what you want.



Ill try to play with it a little thanks..., lets see how it goes

IH
#10

clonex / Ivan Hudec

01.04.2021 22:58
k
#11

Karish

01.04.2021 23:37

Attachment e3vty7Vk1p.png added

e3vty7Vk1p.png
(51.52 KiB)
I dont know about that Random forest thing but it looks cool also,

I attached an example of what would be a cool visual feature of this method i wrote about,

and if it can be built visually it can be built automatically with no GUI and make the decision automatically for us if it passed the score for all parameters overall or individually,


this can be done in no time i believe because some of those functions are already built into SQ and can be combined to create it.

k
#12

Karish

01.04.2021 23:53

We can make something like "Optimization Profile and System Parameter Permutation" which is already implemented but rather checking all available parameters,

we will do it as explained in detail, above..


this would be more efficient, faster, logical etc..


https://strategyquant.com/doc/strategyquant/optimization-profile-system-parameter-permutation-strategyquant/


Im sure you got the idea..


b
#13

bentra

02.04.2021 06:10
In dev4 you can already run a montecarlo crosscheck during building and calculate the gene fitness by  any confidence level of the montecarlo....

Apparently it may be bugged though like the calc fitness by retest higher precision. https://roadmap.strategyquant.com/tasks/sq4_7586

We can make something like "Optimization Profile and System Parameter Permutation" which is already implemented but rather checking all available parameters,

we will do it as explained in detail, above..


Could probably petition for a calculate fitness by SPP median...

b
#14

bentra

02.04.2021 06:10
Voted for this task.
b
#15

bentra

02.04.2021 06:49


MA+SL+TP (75) MA+SL+TP (80) MA+SL+TP (85) MA+SL+TP (90) MA+SL+TP (95) MA+SL+TP (100) MA+SL+TP (105) MA+SL+TP (110) MA+SL+TP (115) MA+SL+TP (120) MA+SL+TP (125)

I'd stay away from that one (method #2) it will be severely biased for some strats. The 75 MA if it's a cross for instance will produce more trades than the 125 MA. So way more trades will have smaller TP and SL skewing the results. There's nothing wrong with one at a time (method #1) or monte-carlo is standard.

k
#16

Karish

02.04.2021 11:11

Yes, Method #1 would be what i aim for,

But this should be flexible,


Users can choose MIN/MAX/STEPS in %,

What parameters to choose from, like Integers/Doubles/Bools etc..,

Method #1 or Method #2,

Show Visual Statistics or not,

etc..


But it should be something similar as SPP "Optimization Profile and System Parameter Permutation",

The interface is ready, and have most of the stuff there, but we cannot manipulate it,

it will just search for all combinations of parameters rather than staying with-in the range of the closest neighboring parameters in % and in steps of %..


So i would say this method is like more than 50% built.

k
#17

Karish

02.04.2021 14:37
m
#18

Matze

13.04.2021 22:45
Voted for this task.
NS
#19

nicksim80

15.04.2021 00:56
Voted for this task.
k
#20

Karish

17.03.2024 21:28

Subject changed from More Accurate & Robust Genetic Engine to DELETE - IT'S IMPLEMENTED ALREADY

Description changed:

DELETE - IT'S IMPLEMENTED ALREADY

DELETE - IT'S IMPLEMENTED ALREADY

Votes: +6

Drop files to upload

or

choose files

Max size: 5MB

Not allowed: exe, msi, application, reg, php, js, htaccess, htpasswd, gitignore

...
Wait please