Gekko Forum
VmaPredict2.js (new) - Printable Version

+- Gekko Forum (https://forum.gekko.wizb.it)
+-- Forum: Gekko (https://forum.gekko.wizb.it/forum-13.html)
+--- Forum: Strategy Development (https://forum.gekko.wizb.it/forum-12.html)
+--- Thread: VmaPredict2.js (new) (/thread-58001.html)

Pages: 1 2 3 4 5 6 7


RE: VmaPredict2.js (new) - PGTART - 09-10-2019

Ok the first version i call now : VmaPredictOriginal
What i created later is now named : VmaPredict2

The latest though is VmaPredict 4 ( https://github.com/PGTBoos/GekkoStrategies )
It's a slight improvement upon the original version VmaPredictOriginal , and it has better console output for debugging, showing date and crypto price at that buy moment. On estimate, it will do 5% better (due to small optimization in MediumOut and ShortSight handling).


RE: VmaPredict2.js (new) - Jupiterix - 09-11-2019

(09-10-2019, 06:21 PM)PGTART Wrote: Ok the first version i call now : VmaPredictOriginal
What i created later is now named : VmaPredict2

The latest though is VmaPredict 4  ( https://github.com/PGTBoos/GekkoStrategies )
It's a slight improvement upon the original version VmaPredictOriginal  , and it has better console output for debugging, showing date and crypto price at that buy moment. On estimate, it will do 5% better (due to small optimization in MediumOut and ShortSight handling).

Thanks, I'm testing it, for now v.4 is giving good results with 1 hour candles, even if in the bear market it is in great difficulty, If you need I can do collaborative tests.


RE: VmaPredict2.js (new) - anon.e.mous - 09-11-2019

I've repeated the tests that I've done before and the original is beating the v4 by 10% using the same toml configurations (posted on earlier posts) and the same time-frame (from 01-01-2019 until NOW).


RE: VmaPredict2.js (new) - PGTART - 09-11-2019

(09-11-2019, 07:45 AM)Jupiterix Wrote:
(09-10-2019, 06:21 PM)PGTART Wrote: Ok the first version i call now : VmaPredictOriginal
What i created later is now named : VmaPredict2

The latest though is VmaPredict 4  ( https://github.com/PGTBoos/GekkoStrategies )
It's a slight improvement upon the original version VmaPredictOriginal  , and it has better console output for debugging, showing date and crypto price at that buy moment. On estimate, it will do 5% better (due to small optimization in MediumOut and ShortSight handling).

Thanks, I'm testing it, for now v.4 is giving good results with 1 hour candles, even if in the bear market it is in great difficulty, If you need I can do collaborative tests.

Well i cannot do two thing at once, that is heavily test code and write code  Wink
So please be my geust, and share your findings here

Currently the code is focused around 1 hour time frames.
But there might be settings that work with 5 to 10 minutes trading I'm real keen about knowing that, as i would prefer such trading speed.

On the other hand If you know settings that worked great with several bear markets then it might be an idea that I build in some kind of switch
but i might need a more advanced indicator for longer term bearmakets.


RE: VmaPredict2.js (new) - anon.e.mous - 09-11-2019

Please don't get me wrong, I appreciate your hard work!

I'm just posting results comparing the versions that you've released. I'm always testing with 1 hour candles with 30 warmup and inside the time-frame that I declared above.

To changes that I've done are only on the TOML file, this is what I'm using:

[Swings]
Magica = 280
MediumOut = 280
ShortSight = 380
MaxPeekPred = 580

[ForcastTrend]
Short =  6
Medium = 17
Long = 100   #nan

[PeriodAverage]
Short = 6
Medium = 25  
Long = 100    #nan

[VolumePeriodAverage]  
Short  = 6    #nan
Medium = 30  
Long   = 100  #nan

I've also tested with smaller candles (5 and 10 min) but the results are not as good as with 1 hour.


RE: VmaPredict2.js (new) - anon.e.mous - 09-11-2019

I continue to test the first version and tweaking the TOML parameters. I think I've maxed out the tweaking I can do. We this lasteste TOML using a long trade session (from : 2019-01-01 00:00 UNTIL : 2019-09-11 09:45) we could get a very very decent profit. At least on backtesting Smile

Results: 

[Image: rlhcuxr.jpg]


RE: VmaPredict2.js (new) - PGTART - 09-11-2019

Wow, that's an amazing result ... I wouldn't think such high results where possible by tweaking alone.
Writing code is one thing, but finding optimal settings for it is also an art, for strategies
curiously that it doesn't trade for a long time from feb to may.
please share the used settings, so we can improve it more.

oh BTW version 5 has a small protection against continous repeated stoplosses (draining the wallet) but not sure how well it behaves
the stoploss wont trigger another one after a price goes up for a brief moment (3 candles) => i can make 3 a toml setting too.
but i'm not so sure how often it happens... and maybe other tricks should be tought off (watch the console when loosing money if you have an insight let me know).

Also, version 6 will be released, I think tomorrow.
As I also uploaded the candlestick trader, version 6 will have knowledge of candles as well !! (usable inside the code!! to optimize logic)
It is able to detect some candle types Hangman, Hammer, ShootingStar
As I think those are the most useful (but I can add more if you people want).
It keeps a record of the past last 10 known stick names if any was found.

BTW the trading logic is around line 150 the line that reads
check: function (candle) {
The if (! this.Gotcoins ) "!" means not in coding so thus there it start the buy logic if we have no coins between { }. The else { } after it sells.

Maybe if you observe those rules (where Swingxxxx stands for TOML input variables), you might get ideas of how to use candle logic with it.
I'm open for suggestions


RE: VmaPredict2.js (new) - anon.e.mous - 09-12-2019

(09-11-2019, 04:57 PM)PGTART Wrote: Wow, that's an amazing result ... I wouldn't think such high results where possible by tweaking alone.
Writing code is one thing, but finding optimal settings for it is also an art, for strategies
curiously that it doesn't trade for a long time from feb to may.
please share the used settings, so we can improve it more.

oh BTW version 5 has a small protection against continous repeated stoplosses (draining the wallet) but not sure how well it behaves
the stoploss wont trigger another one after a price goes up for a brief moment (3 candles) => i can make 3 a toml setting too.
but i'm not so sure how often it happens... and maybe other tricks should be tought off (watch the console when loosing money if you have an insight let me know).

Also, version 6 will be released, I think tomorrow.
As I also uploaded the candlestick trader, version 6 will have knowledge of candles as well !!    (usable inside the code!! to optimize logic)
It is able to detect some candle types Hangman, Hammer, ShootingStar
As I think those are the most useful (but I can add more if you people want).
It keeps a record of the past last 10 known stick names if any was found.

BTW the trading logic is around line 150 the line that reads
   check: function (candle) {
The    if (! this.Gotcoins )          "!" means not in coding so thus there it start the buy logic if we have no coins between {   }. The else {  } after it sells.

Maybe if you observe those rules (where Swingxxxx stands for TOML input variables), you might get ideas of how to use candle logic with it.
I'm open for suggestions

the settings that I'm using for the TOML are:

[Swings]

Magica = 325
MediumOut = 280
ShortSight = 400
MaxPeekPred = 580

[ForcastTrend]
Short =  6 
Medium = 17
Long = 100   #nan

[PeriodAverage]
Short = 6 
Medium = 25   
Long = 100    #nan

[VolumePeriodAverage]   
Short  = 6    #nan
Medium = 30  
Long   = 100  #nan


Using 1 hour candles with 30 for warmup.


RE: VmaPredict2.js (new) - PGTART - 09-15-2019

Version 6 has just been released !!.
My plans where todo more in version 6 but that be postponed to version 7

MaxPeekPred is now called DownHill  
It still shorts based upon a givven loss value, its based upon predictions it might loose more as well since the future is unknown.


Also the other 2 rules (Magica and ShortSight) will no longer go short if they wouldnt make proffit (only DownHill can do that).

Version 6 is candleaware but the current rulles didnt improve with candle logic, so its not used yet (might be used in the future 7 ?).

Console outputs now also for the shorts shows the moment (date time), in red

2 new toml  indicators have been added :   (old toml files can not be used anymore).


- HighShot =xxxx
Especially in shorter time intervals, but in other ranges as well, there are regularly sudden price jumps.
I have no ideas why those occure but they do happen, if last 2 candles averaged for a certain amount are xxxx higher as compared to their previous 3 candles, then go short (take profit from that uprise by xxxx)

- DayStop =xxxx
This will stop trading if a day has lost more then xxxx , the reasoning is wait till next day till markets might have recovered from bad signaling to this strategy



---- some comments
I'm over 300 lines of coding now  Smile

Highshots gets often followed by a sideway trend market.
Maybe version 7 will make use of that, ea delay buy when in a side market, since highshots often drop in price as well (short price bubles).
In 7 i might address that (maybe), but tests by you people might allready give some good results.
And i need such test results to allow new math to work, so i depend on you as well guys.

For version 7 i would like to alter some constants also so that they will work in percentages.
That might make it easier for using this with AltCoins.

Notice you need the toml version that came with the strategy (since variable names have changed in version 6 ).



The included toml file is not optimal, higher gains can be made,
Its just the toml that i used while i was writing code to test new settings out

The new options where added to create strategies that would work also in shorter time ranges
I hope you people can find some good settings showing that and share them overhere.
We know now this math works pretty good in 1 hour math,
But i'm eager to optimize this for shorter time ranges as well that's why i need your help to test things out
And find those spots, share your results here, then can I can check the behaviour and improve the code upon it even more.



RE: VmaPredict2.js (new) - anon.e.mous - 09-16-2019

Hi, where was it published?