Better NN EA development - page 28
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
Here's the result i've got from the "open source" version, with the open source indicator.
Init Capital : 10K$
First test : Without MM. 0.1 lot, 6 months (Jan-Today). Buy/Sell trigger at 54/74, SL 200 TS 170.
Second test : Without MM. 1 lot, 6 months (Jan-Today). Buy/Sell trigger at 54/74, SL 200 TS 170.
not bad...
@dswk - I stand corrected - I was entering fractional values for the Buy/Sell triggers which was the problem! Thanks!
PS - did you retrain the indicator for a particular currency before posting it?
The training of the NN is not as easy as it may sound... It involves using the cortex software.. I would suspect a re-trained version could prove to be better then the outdated version we currently have..
deleted....
And I thought that the mimetic ea I posted is obtained from some corner of Internet,Who has already found its trail could share it?
The training of the NN is not as easy as it may sound... It involves using the cortex software.. I would suspect a re-trained version could prove to be better then the outdated version we currently have..
Hmmm, I stand corrected again! Having just spent an hour with Cortex (which is pretty easy to use), I find that while training an network is easy, there is no way to export it from Cortex - I'm pretty sure this is one of the options that is missing with the Shareware version, along with the inability to run scripts (which sucks!)...
btw, interesting exercise - while the opensource EA does seem to produce good recent results for EURUSD H1, running a test on all available data with a monthly chart open, you should notice that the equity curves follows price perfectly - when EUR is appreciating, it makes money, and when it was falling it lost money. This is further borne out when you compare the Buy/Sell percentages. Considering the EUR has been appreciating almost continually for several years now, I reckon its apparent edge is an illusion...
Hmmm, I stand corrected again! Having just spent an hour with Cortex (which is pretty easy to use), I find that while training an network is easy, there is no way to export it from Cortex - I'm pretty sure this is one of the options that is missing with the Shareware version, along with the inability to run scripts (which sucks!)... btw, interesting exercise - while the opensource EA does seem to produce good recent results for EURUSD H1, running a test on all available data with a monthly chart open, you should notice that the equity curves follows price perfectly - when EUR is appreciating, it makes money, and when it was falling it lost money. This is further borne out when you compare the Buy/Sell percentages. Considering the EUR has been appreciating almost continually for several years now, I reckon its apparent edge is an illusion...
This will be fixed by optimizing the buy/sell levels.. As eur/usd begins to fall the sell level should be closer to the 50 level.. The optimizer in mt4 would give the best results for the more recent market..
This is why i said the strategy tester should be run one month at a time after a 3 month optimization.. this would roughly produce the most accurate buys/sells for the market/upcoming market..
Running live i would probably re-optimize these values every weekend..
This will be fixed by optimizing the buy/sell levels.. As eur/usd begins to fall the sell level should be closer to the 50 level.. The optimizer in mt4 would give the best results for the more recent market..
This is why i said the strategy tester should be run one month at a time after a 3 month optimization.. this would roughly produce the most accurate buys/sells for the market/upcoming market..
Running live i would probably re-optimize these values every weekend..That's the way to use it ...
This will be fixed by optimizing the buy/sell levels.. As eur/usd begins to fall the sell level should be closer to the 50 level.. The optimizer in mt4 would give the best results for the more recent market..
This is why i said the strategy tester should be run one month at a time after a 3 month optimization.. this would roughly produce the most accurate buys/sells for the market/upcoming market..
Running live i would probably re-optimize these values every weekend..I have already done an optimisation for the period Jan '07 -> present - the optimisation results look little more than random to me!
The 'kicker' as far as I'm concerned is that the two differently named NN indicators that I have in my library all use the same weights that the author of the NN tutorial generated back in 2003. In fact, considering that Bogie's indicator generates an identical output signal, it looks increasingly likely that this also uses the same NN weighting values! So optimising Buy/Sell levels for 2003 weightings is pretty much a waste of time (imo)...
I dont think you are understanding... The output of the NN is not going to change as long as the NN weights do not change so no matter if EUR/USD is uptrending/downtrending/horizontal the indicator will still be able to give Buy/Sell signals.. But the actuall buy and sell levels will change based on what the market is doing "more".
For example the EUR/USD market has been in a strong uptrend since 2001 that is why default buy level is closer to 50 then the sell level, but for the last couple of months the market has been horizontal. So the Buy/Sell levels would be more even'd out.. Test this youself, re-optimize the buy/sell levels for the more recent "horizontal market". You will see how the buy/sell levels will change based on market trend.
If you optimize for too long of a period you will see more random results like you have seen.. Because you have optimized from jan07 to present how many times have the market changed? Look at the eur/usd 1day since jan07.. How many times has market changed? i can count more then 4. If you optimize too long you will be over-generalizing. If you optimize for just the last month you might be curve fitted to just that months market. The idea is to find just the right amount of timeframe to optimize for.. Not too much, not too little. This is why i said 3 months.. you will then see what im explaining..
BTW i can guarantee the weights/logic (line for line) of the bogie indicator and indicator from the cortex site are the same..
Infact i can guarantee the entry logic from bogie and the open source EA are the same..