Faster backtest with x second interval/timeframe

Have some interesting ideas about how to improve the software? Let us know in this section!
kekkis
Posts: 12
Joined: Tue Jun 28, 2016 3:41 pm

Re: Faster backtest with x second interval/timeframe

Post by kekkis »

Faster startegy testing, is allways improvment.

I dont mind if seconds, are missing some events, i can live with it , first run as rough approximation with seconds.
sometimes its actually good if the run, miss peak values. better generalasation.
Anyhow have to run best results, again and tweak parameters and do observations, and test they dont curve fit and are not too sensitive

I dont like to use mt4's genetic optimizer, because dont know how it works, i dont even think its solid or robust, example. lack of fitness functions,
those which are availaible are like made by novice programmer who really dont understand the topic of genetic algorithms and eveolutionary based systems.
Even i have now seen some new articles in mql5 forum.

tip for people who have to stop startegy test because of time.
Allways buy intel processors, with top single thread performance/price. if using mt4 for optimisation.

akirabrr
Posts: 4
Joined: Sat Jun 25, 2016 2:49 pm

Re: Faster backtest with x second interval/timeframe

Post by akirabrr »

im completely interested about this option, so fundamental for the backtest on another platform except the standard ones, i would like to be part of the team interested, and give my full support

akirabrr
Posts: 4
Joined: Sat Jun 25, 2016 2:49 pm

Re: Faster backtest with x second interval/timeframe

Post by akirabrr »

im developing a tool for backtesting with .csv format
it will be exactly equal to mt4, any people interested on helping getting .csv format with second timeframe add me, it dont need support from mt4 and the effectivness will be the same, i really need people interested in this, add me on skype torino alvorada

tickstory
Posts: 4883
Joined: Sun Jan 06, 2013 12:27 am

Re: Faster backtest with x second interval/timeframe

Post by tickstory »

Hi Akirabrr,

If you are interested in 1 & 10 second time-frames in CSV, then Tickstory already supports this. At the moment we're discussing specifically a similar modification for MT4.

Thanks.

ticksquick
Posts: 6
Joined: Mon Sep 21, 2015 11:21 pm

Re: Faster backtest with x second interval/timeframe

Post by ticksquick »

I would love to see faster back MT4 testing, and would welcome any speed improvements over the current set up with Tick Story and Birt's softfware. In my opinion the weak link for Meta Quotes MT4 platform has been the back testing, and that's been going on for years now.

George

TJS
Posts: 7
Joined: Sun May 22, 2016 3:24 pm

Re: Faster backtest with x second interval/timeframe

Post by TJS »

Hi,

good idea. I personally prefer using every tick for all tests. Perhaps this is off-topic but my focus is more on hardware. A new skylake i7 6700K is so much faster than my old i7 (first generation) - now can watch "online" for results !
Torsten

experienced_tester
Posts: 18
Joined: Wed Jul 03, 2013 12:38 pm

Re: Faster backtest with x second interval/timeframe

Post by experienced_tester »

Hi everybody

If Tickstory team will implement this feature, I think each one of us could test and select
the suitable EAs to benefit from it.
Thank you again, Tickstory, for all your work!

4EverMaAT
Posts: 29
Joined: Mon Jan 20, 2014 3:55 pm

Re: Faster backtest with x second interval/timeframe

Post by 4EverMaAT »

tickstory wrote:..........

The implementation approach we would like to take is to filter the FXT file. This is a proven method and fits in exactly with the current MT4 FXT generation model so therefore wont easily be broken by future MT4 releases. I think it's worthwhile to note that we are all talking about the same high-level goal here - to speed up back-testing. We just want some input on useful algorithms to filter ticks and you can then leave the implementation detail up to us.

Radu - I'm not sure how your Option 4 differs from Option 2 that was mentioned. I don't see how it would dramatically affect results. Basically you are creating S10, S30 etc candles as Siolibros suggested. Potentially you can go down to a S5, S4, S1 and then tick timeframe to progressively get more accuracy. I would rate this as the most consistent solution at this stage.

Regards.
I think modifying how the fxt file is generated is also the best way to accomplish this. Instead of including every tick movement in the FXT, you can simply create 4 movements with simulated OHLC for each 1/2/5/10/12/15/20/etc second intervals within the 1/5/10/etc minute FXT timeframe. This should reduce the ticks drastically.

I suppose something similar can be done with Birt's csv2fxt file, and I think he provides the full source code for all the related files, including csvreader.dll But you get the advantage of using real spread (I use TDS).

4EverMaAT
Posts: 29
Joined: Mon Jan 20, 2014 3:55 pm

Re: Faster backtest with x second interval/timeframe

Post by 4EverMaAT »

I have an even better idea, and it would be easier to implement.

When generating CSV file or fxt file, why not have an option to "skip every x tick" ? x amount of ticks to evaluate before skipping the next tick.

x=0 // disable; process all ticks
x=1 // skip every tick (not used unless you want empty file)
x=2 // skip every other tick
x=3 // skip every 3rd tick
x=4 // skip every 4th tick
etc, etc

This reduces the accuracy to some degree, but would also reduce total amount of ticks saved in fxt file. It is a tradeoff, but should be quick to program and test. You can also have separate checkbox to ignore duplicate ticks when processing fxt file.

Easy to compare the accuracy when all ticks are used vs when only using 1/2, 2/3, 3/4, 4/5, 5/6, etc ticks.

siolibros
Posts: 13
Joined: Fri May 27, 2016 9:04 am

Re: Faster backtest with x second interval/timeframe

Post by siolibros »

thanks for all comments guys
i hope tickstory team will consider some upgrades about

in the meantime i worked all last 4 days using only my method to backtest, using only openbar method, using slightly different mql4 code
in this way i could backtest all my EA versions on 8 years with original dukascopy data
after that, i retested best results on tick data, with really low differences, in some cases betting best results
if i did the same work directly on tick data, it was impossible to do
with my described method, i'm able to do a full backtest optimization on 8 years in just a couple of days, on a i7 intel computer (using up to 6 cuncurrent mt4 instances)

i personally realized that 1 minute precision is quite enought for most of EA on the market, except for scalpers
any other logic will benefit of my suggested backtest method
i prefer to have my EA optimized to work without problems on 8 years on 1 minute interval rather (1-2 days on i7 processor) than 2 or 4 years on tick data (requires about 1 month on i7 processor)
single optimization on 8 years tick data would require about 2 months... really slower than 1-2 days ;-)))

happy backtesting and optimization to all of you ;-))

Post Reply