Noxa indicators for Neuroshell

[Noxa]

Hi,

Post 180 - training range was set incorrectly so future leak and repaint occured later
so those results are not valid unfortunatelly

Post 211 - all settings were correct ===>very profitable for 1st 12h

So any solution (except falling or flattering equity curve)to dedect that conditions of market are changing and it's time to reoptimize CSSA or simple to stop trade with it ??

Regarding your optimization tool. Any special reason that you keep it secret ??

Krzysztof

>> So any solution (except falling or flattering equity curve)to dedect that conditions of market are changing and it's time to reoptimize CSSA or simple to stop trade with it ??
Our technique works well in ETFs and stocks. We have not tried it in currencies, so you may want to test it if it works for you.
We don't reoptimize very often. Instead we collect cycles that worked in the past (up to 3 or 4 per issue) so that we can jump on them quickly. Our decision to reoptimize is based on whether the market regime changed or not; to this end we use various indicators (entropy-based and others)... But the important question is when we should not trade with cycles (congestion ranges, trend swamping, trend persistence and snr ratio).

>> Regarding your optimization tool. Any special reason that you keep it secret ??
There is no optimization tool really; just excel. The method however is proprietary to Noxa but why not considering a release in some future...

Noxa

 

[Noxa]

Hi,

IF it can make on CHIRP with noise than it is very powerfull. All other systems like MESA/Fourier/Goertzel will have a big problem here because of lack of spectrum.

Krzysztof

Notice that I used about -16dB snr at its lower (Gaussian noise) and this is already pushing the envelope for a causal indicator. The detection limit for CSSA is more about -13dB. As for uniform noises I believe it is more about -8dB but I'll have to double check that; in fact we did not conduct much tests with uniform noises as we believe they are not representative of random walks.

Noxa

 

[arryex]

Inserting external dll indicator

Hi Krzys,

As per your PM request, here I put my personal guide how to insert any external dll indicator.

We can do the same thing for any dll indicator as long as you know what are the required parameters.


Good luck,
Arry

 

[Krzysiaczek99]

without knowing the source ??

Hi Krzys,

As per your PM request, here I put my personal guide how to insert any external dll indicator.

We can do the same thing for any dll indicator as long as you know what are the required parameters.


Good luck,
Arry

Hi,

So did you insert those parameter without knowing the source code ?? Because I thought we need to know exact parameter types and order. Here is an example from NS SDK

// This is a very simple example of a DLL function which is
// easily coded and used as an indicator in the NeuroShell
// Trader Professional. .
_declspec(dllexport) double Spread (double high, double low)

//Returns the spread (difference) between two
// "input" variables, such as the high and the low

{return high - low; }

Krzysztof

 

[arryex]

Hi Krzys,

Based on NS SDK we can call dll function with two methods by value or by array. The one you shown is function call by value. Most of indicator use the second method (function call by array).

Refer to the chart shown on Traders Tips - November 2008, I knew that coronacharts.dll use two variables High and low, and by assuming that the method is call by array i use the method that I explained recently on the ebook.

You may try to call coronacharts.ddl using function call by value, you will see the result...
So I can inserting the indicator dll without knowing the source code (most of the time but not always). Lucky if you get the template (tpl) file provided by some one/creator. On the given ebook it shown how to create your own template.

Arry

 

[Krzysiaczek99]

comparative analysis

Comparative analysis of systems based on MESA, GOERTZEL and NOXA CSSA starts here.

Trading Strategies Based On Digital Filters - Page 53 - Forex Trading

post 521 and down

Some comments are welcome.

Krzysztof

 

[Noxa]

Results are not conclusive

Comparative analysis of systems based on MESA, GOERTZEL and NOXA CSSA starts here.

Trading Strategies Based On Digital Filters - Page 53 - Forex Trading

post 521 and down

Some comments are welcome.

Krzysztof

My comments regarding your posts #531 and #536 on forex.tsd

With all due respect, your results are not conclusive and VERY MISLEADING to say the least. You’ll have to explain the readers of this thread why (I quote you) CSSA is “definitelly worse” than MESA!!!

Your post#531:
Trading Strategies Based On Digital Filters - Page 54 - Forex Trading
>> Here is comparision to NOXA results on the same signal. They hit 100% on
>> clean CHIRP signal, 72.2% on _2 and 90.9% on _1.

>> So definitelly worse even that it is curve fitter type. Can you make it also on GOLD5 ??”


100% HITS IS NOISE FITTING:

1. You definitively cannot use %profitable trades as you cannot use a change in PF as you did in the past to compare indicators and assess the quality of a system unless it is statistically correct. My point relates to the noise artifact I already touched on (read post#166):
http://www.trade2win.com/boards/software/29736-noxa-indicators-neuroshell-21.html#post670462
As you can see in the graph below (left graph), a trade from a perfect entry signal can be a looser when snr is lower than 0dB (0dB means as much noise as signal). In fact the odds that it looses are 50% when the noise is uniform. Since we are dealing with -16dB in this example (at its lowest, -16dB means 6.3 times more noise than signal) and only 11 trades, it is very unlikely that they can be all winners unless there is noise fitting. I have attached another example where all trades loose even though they are nearly perfect (GOLD15 + noise, snr = 13dB). They loose because of noise artifact, not because of bad entry signals.

Another point I want to make: you certainly cannot make such bold statement based on only a couple of noise trials. You need a full factorial DOE.


YOU COMPARE APPLES AND ORANGES

2. You do not compare MESA with CSSA but digital filtering + MESA with CSSA (correct me if I am wrong). We know that because MESA performs poorly under high noise condition this additional filtering step was necessary which unfortunately introduces lags.

3. MESA results were obtained in-sample while CSSA results were given out-of-sample.


YOU TOOK CSSA RESULTS OUT OF THEIR CONTEXT

4. MESA results were obtained on noisy signal directly while CSSA was trained on noise free signal first and tested against noise after adding noise to it.


BACKWARD CURVE FITTING OCCURRED WITH MESA

5. The setting for MESA was adjusted during noise comparison: max_period was changed from 140 to 800 to accommodate for the increased noise (noise 2). The setting with Noxa remained unchanged. Wasn’t that your preferred argument when trying to bash CSSA? Backward curve fitting!! I am very surprised that you did not report it!!

Your post #536:
Trading Strategies Based On Digital Filters - Page 54 - Forex Trading
>> so it looks that in this competition MESA won, 2nd NOXA and third GOERTZEL as a clear underperformer.”

Let me redo the test the way it was done with MESA; no training, just manual setting on noisy signal directly (third and bottom graphs, chart attached). See what happens! Let me use your colored terms:innocent:: Results IMMEDIATELY jump to 100% on both noises. I did not even need to change the parameters between the two noise levels. Should I conclude that CSSA is now as good as MESA or better?

More seriously, the only information I can use from these experiments to compare MESA with CSSA is the quality of the cycles they both produced. With MESA, cycles lag and can be noisy whereas cycles from CSSA remain smooth and timely.

Noxa

 

[Krzysiaczek99]

My comments regarding your posts #531 and #536 on forex.tsd

With all due respect, your results are not conclusive and VERY MISLEADING to say the least. You’ll have to explain the readers of this thread why (I quote you) CSSA is “definitelly worse” than MESA!!!

Your post#531:
Trading Strategies Based On Digital Filters - Page 54 - Forex Trading
>> Here is comparision to NOXA results on the same signal. They hit 100% on
>> clean CHIRP signal, 72.2% on _2 and 90.9% on _1.

>> So definitelly worse even that it is curve fitter type. Can you make it also on GOLD5 ??”


100% HITS IS NOISE FITTING:

1. You definitively cannot use %profitable trades as you cannot use a change in PF as you did in the past to compare indicators and assess the quality of a system unless it is statistically correct. My point relates to the noise artifact I already touched on (read post#166):
http://www.trade2win.com/boards/software/29736-noxa-indicators-neuroshell-21.html#post670462
As you can see in the graph below (left graph), a trade from a perfect entry signal can be a looser when snr is lower than 0dB (0dB means as much noise as signal). In fact the odds that it looses are 50% when the noise is uniform. Since we are dealing with -16dB in this example (at its lowest, -16dB means 6.3 times more noise than signal) and only 11 trades, it is very unlikely that they can be all winners unless there is noise fitting. I have attached another example where all trades loose even though they are nearly perfect (GOLD15 + noise, snr = 13dB). They loose because of noise artifact, not because of bad entry signals.

Another point I want to make: you certainly cannot make such bold statement based on only a couple of noise trials. You need a full factorial DOE.


YOU COMPARE APPLES AND ORANGES

2. You do not compare MESA with CSSA but digital filtering + MESA with CSSA (correct me if I am wrong). We know that because MESA performs poorly under high noise condition this additional filtering step was necessary which unfortunately introduces lags.

3. MESA results were obtained in-sample while CSSA results were given out-of-sample.


YOU TOOK CSSA RESULTS OUT OF THEIR CONTEXT

4. MESA results were obtained on noisy signal directly while CSSA was trained on noise free signal first and tested against noise after adding noise to it.


BACKWARD CURVE FITTING OCCURRED WITH MESA

5. The setting for MESA was adjusted during noise comparison: max_period was changed from 140 to 800 to accommodate for the increased noise (noise 2). The setting with Noxa remained unchanged. Wasn’t that your preferred argument when trying to bash CSSA? Backward curve fitting!! I am very surprised that you did not report it!!

Your post #536:
Trading Strategies Based On Digital Filters - Page 54 - Forex Trading
>> so it looks that in this competition MESA won, 2nd NOXA and third GOERTZEL as a clear underperformer.”

Let me redo the test the way it was done with MESA; no training, just manual setting on noisy signal directly (third and bottom graphs, chart attached). See what happens! Let me use your colored terms:innocent:: Results IMMEDIATELY jump to 100% on both noises. I did not even need to change the parameters between the two noise levels. Should I conclude that CSSA is now as good as MESA or better?

More seriously, the only information I can use from these experiments to compare MESA with CSSA is the quality of the cycles they both produced. With MESA, cycles lag and can be noisy whereas cycles from CSSA remain smooth and timely.

Noxa

Hi,

I'm very happy you replied and I also think that results are inconclusive. See my post 549 at forex-tsd.
I'm not an author of MESA test, i just provided data set from your chart, I will ask author
of MESA system to comment if we have backward curve fitting here and if all conditions for out of sample test are fullfilled.

Krzysztof

 

[arryex]

Tips using MT4NSDT Data feed

First, thanks to Kryzstof who has provide me the correct files.

I could not found any complete guide using MT4NSDT Datafeed & TradePut, hope the following tips are useful.
1. used File and Folder location
In NSDT
C: \ NeuroShell Trader 5\Servers\MT4NSDTDataFeed.dll
C: \ NeuroShell Trader 5\Servers\MT4NSDTDataFeed.DAT
C: \ NeuroShell Trader 5\Servers\ MT4NSDTDataFeed \ Forex.lst
C: \ NeuroShell Trader 5\Template\Trade put.dll
C: \ NeuroShell Trader 5\Template\Trade put MT4.tpl
Edit file MT4NSDTDataFeed.DAT as the correct MT4 folder (some brokers has different folder name)
In MT4
C:\Program Files\Meta Trader\experts\NSDT-MT4Feed.ex4
C:\Program Files\Meta Trader\experts\libraries\MT4NSDTLib.dll

I assumed here that you have no problem on program registration. I insert the original manual from the developer.

2. Which Run first MT4 or NSDT
When I run NSDT first, MT4 with MT4NSDT Data feed will be blank (no commentary), even the data transmitted to NSDT. See att 1.
When I closed NSDT then the commentary will be displayed in MT4, see att 2.

3. Trade Put
I found two methods we can use to send the signals:
a. Using prediction signal (see att. 3), with this configuration, signals will be directly transmitted if no open position in MT4.
b. Using Trading Strategy output (see att. 4), with this configuration, only new signal will be transmitted no matter existing open position in MT4.

The configuration is shown in att. 5.

4. Chart not synchronize in NSDT and MT4.
After closing both MT4 and NSDT for a while, when restart them, there will be missing bars in NSDT since the data timing stored has different with new incoming data. There are two files involved for the historical data storing in MT4 (extension hst), e.g. EURUSD5.hst and e.g. nEURUSD5.hst. nEURUSD5.hst files is used by NSDT.
If both file has different size then you will get different bar on NSDT.
If both file are same but you get missing bars either in MT4 and NSDT, it means hst data is stored by MT4 not properly. You need to close MT4 program and NSDT opened chart, then restart MT4 first then reopen the chart on NSDT. See att 6a/b for missing bar sample and 7 after chart rebuild.

5. Precaution running the addons
If you use the method 3a above, you will be possible in swing condition, example if NSDT Prediction Strategy generate sell signal while temporary the price in uptrend. You will be in loss condition if afterward NSDT generate new reversal signal.

Personally, using method 3b is more safe, since the signal will be generated only by new signal. Preferable for me to use this method and enter the trade manually. Of course combination manual and auto trade may profitable.

Last but not least, whatever any trading methods we build, discussion even fightings, but should come to our original target,i.e. PROFITABLE TRADE !!! Truly it is not an easy thing.
Here is the sample of my recent trade in demo account. See att. 8. At the moment I use Noxa-CSSA trendline as one of my prediction strategy input (see att 5).

Any advice are appreciated.

Regards,
Arry

 

[kimkim]

Hello arry,

You can enhance your prediction and strategy using multiple time frame indicator, available here.

You can attach different time frame on ONE chart. When you optimize parameter for noxa strategies, you can also optimize the timeframe (!). So the signal can be generated like this: Buy Long signal from noxa oscillator cross below on time frame 5 minutes, Sell Short signal from noxa oscillator cross above on time frame 15 minutes.

Below chart is what I have done with GBPUSD with the base chart of 1 hour. I optimize time frame from 1 to 4 hour. Out of sample results is 744 pips (net profit $0.0744) from Feb 24th until today. Please note I made this chart last month and I am testing it until end of this month.



Cheers

 

[kimkim]

This is trading strategy for GBPUSD using base chart 15 minutes, optimized for 13 days and has been running for 3 days since last Monday.



Out of sample result is excellent:



I think there is no need find thousands % of return. I prefer highest % profitable trades as we can manage size of trade rather than frequency of trades. But it's just my opinion.

Cheers

 

[arryex]

Hi Kim Kim,

Thank you to share your idea, would you please posted the chart with saved data.

Personally, setting the m-history is the first door prior using noxa to build trading strategy.

I use my own formula to get raftly m-history, measure each cycle length (peak-to-peak or valley-to-valley), calculate average, std dev, min and max. I get almost same values with min+stdev/2 or (max-average)/2, average-min/2, (max-stdev-min)/2. From the chart posted by noxa above #227, I get 250 to 260 values, then I use it for m-history value.

Afterward, on trading strategy with manual adjustment the smoothing factor i found better equity.
You can verify the impact of enlargement/reducing the smoothing factor.

Hope noxa can advice where he can get the value 0.128 and 0.12 on his trading strategy.

Here attached chart (with saved data), you can compare noxa (using Qphase cross over 0) with my cycle crossover.

Thank you,
Arry

 

[arryex]

due to slow internet connection, here I attached the chart.

Cheers

 

[Krzysiaczek99]

perfectlly rigth !!

My comments regarding your posts #531 and #536 on forex.tsd

With all due respect, your results are not conclusive and VERY MISLEADING to say the least. You’ll have to explain the readers of this thread why (I quote you) CSSA is “definitelly worse” than MESA!!!

Your post#531:
Trading Strategies Based On Digital Filters - Page 54 - Forex Trading
>> Here is comparision to NOXA results on the same signal. They hit 100% on
>> clean CHIRP signal, 72.2% on _2 and 90.9% on _1.

>> So definitelly worse even that it is curve fitter type. Can you make it also on GOLD5 ??”


100% HITS IS NOISE FITTING:

1. You definitively cannot use %profitable trades as you cannot use a change in PF as you did in the past to compare indicators and assess the quality of a system unless it is statistically correct. My point relates to the noise artifact I already touched on (read post#166):
http://www.trade2win.com/boards/software/29736-noxa-indicators-neuroshell-21.html#post670462
As you can see in the graph below (left graph), a trade from a perfect entry signal can be a looser when snr is lower than 0dB (0dB means as much noise as signal). In fact the odds that it looses are 50% when the noise is uniform. Since we are dealing with -16dB in this example (at its lowest, -16dB means 6.3 times more noise than signal) and only 11 trades, it is very unlikely that they can be all winners unless there is noise fitting. I have attached another example where all trades loose even though they are nearly perfect (GOLD15 + noise, snr = 13dB). They loose because of noise artifact, not because of bad entry signals.

Another point I want to make: you certainly cannot make such bold statement based on only a couple of noise trials. You need a full factorial DOE.


YOU COMPARE APPLES AND ORANGES

2. You do not compare MESA with CSSA but digital filtering + MESA with CSSA (correct me if I am wrong). We know that because MESA performs poorly under high noise condition this additional filtering step was necessary which unfortunately introduces lags.

3. MESA results were obtained in-sample while CSSA results were given out-of-sample.


YOU TOOK CSSA RESULTS OUT OF THEIR CONTEXT

4. MESA results were obtained on noisy signal directly while CSSA was trained on noise free signal first and tested against noise after adding noise to it.


BACKWARD CURVE FITTING OCCURRED WITH MESA

5. The setting for MESA was adjusted during noise comparison: max_period was changed from 140 to 800 to accommodate for the increased noise (noise 2). The setting with Noxa remained unchanged. Wasn’t that your preferred argument when trying to bash CSSA? Backward curve fitting!! I am very surprised that you did not report it!!

Your post #536:
Trading Strategies Based On Digital Filters - Page 54 - Forex Trading
>> so it looks that in this competition MESA won, 2nd NOXA and third GOERTZEL as a clear underperformer.”

Let me redo the test the way it was done with MESA; no training, just manual setting on noisy signal directly (third and bottom graphs, chart attached). See what happens! Let me use your colored terms:innocent:: Results IMMEDIATELY jump to 100% on both noises. I did not even need to change the parameters between the two noise levels. Should I conclude that CSSA is now as good as MESA or better?

More seriously, the only information I can use from these experiments to compare MESA with CSSA is the quality of the cycles they both produced. With MESA, cycles lag and can be noisy whereas cycles from CSSA remain smooth and timely.

Noxa

You were perfectly right - result were no out of sample and with backward curve fitting so we can not compare apples to oranges. Than of course those results are disqualifying MESA system results in this comparative test. Sorry for confusion and thanks for help.

Krzysztof

 

[richard0099]

hi every one,i am new to neuroshell trader (1 day) can any one give me the intruction on how to connect nts to mt4 and what files to down load and where thank you very much

 

[arryex]

Option for NSDT data feed from MT4

Hi Richard,

There are some options for you:
1. Buy the addons for MT4 data feed from Neuroshell, Addons for NeuroShell Trader / DayTrader - Home, pay $40 to $90 monthly (sorry I never try this)
2. Purchase addon from from www.fxreal.ru, one time fee about $75 (only applicable on XP not for vista) or Try this link RapidShare: Easy Filehosting
3. Purchase SocketNSDT contact the owner [email protected], download demo http://myfolder.nm.ru/progr/socketnsdt.rar, price about $100 (one time fee)
4. Find free version with search engine

Some software will requesting user and password name based on hardware ID, it will not work on Vista since it always having different Hadrware ID number.

good luck
Arry

 

[richard0099]

thank you very much for your infromation ,i will try it ,let see how it work thank agian

 

[akift]

is it over?

Last post is from 01-04 !! Is this wonderful thread over?
So guys [arreyx,Krzyos](not counting NOXA he is the owner) what is your final thoughts?
Did we get to a conclusion for a final step-by-step guideline how to use CSSA.
I think we definitely need tips for NOXA&Neuroshell.
Thanks.

Akif,

 

[acj28]

Last post is from 01-04 !! Is this wonderful thread over?
So guys [arreyx,Krzyos](not counting NOXA he is the owner) what is your final thoughts?
Did we get to a conclusion for a final step-by-step guideline how to use CSSA.
I think we definitely need tips for NOXA&Neuroshell.
Thanks.

Akif,

Akif,

Why not NOXA? They seem to know what they talk about.

 

[akift]

Of course they know, they are the creators!
But I want to learn the `user` perspective. If something is wrong in their comments then NOXA can correct it.

Akif,

Why not NOXA? They seem to know what they talk about.