Bitcoin Trading Software for MtGox - With Neural Networks.

[statdude]

I registered. How long to receive my key via email?

[fox19891989]

Any BTCE trading bot? TKS

[yucca]

I registered. How long to receive my key via email?

You should have it by now, my apologies for the slight delay.

[yucca]

Any BTCE trading bot? TKS

I'm still working on the trade engine for MtGox. I have coded with adding exchanges in mind, so I will be adding further exchanges, BTCE is on the list, It might not be added until autumn though.

I was going to add bitcoin24 next. My wife even got an account there and she bought a few bitcoins and had them in there for a while, and then.... "pooof.. aaaand it's gone!"

[daybyter]

It seems to me, that the system needs an abstraction layer between the trading logic and the exchange interface?

[yucca]

It seems to me, that the system needs an abstraction layer between the trading logic and the exchange interface?

Indeed. In the finished version the neural net signals will be just one trade signal that the user can use as visual reference or can mix with other generated signals for use in autotrading. This way the neural net signal can be used as a confidence bias instead of an absolute signal.

Other trade signals will be generated by a simple scripting interface giving access to all of the charted information, including: all technical overlays and underlays, user drawn trendlines, open/high/low/close/vol, exchange-lag.

The scripting will initially just be able to spot crossings between any pair of signals mentioned and set boolean flags. These flags can be logically mixed with AND,OR.

This is a few months down the road. First off is to actually get the trading interface coded and tested.

Eventually I would like to open it up further and give user simple script API to access entire array history of signals mentioned and perform their own math etc.

[Schrankwand]

It seems to me, that the system needs an abstraction layer between the trading logic and the exchange interface?

Indeed. In the finished version the neural net signals will be just one trade signal that the user can use as visual reference or can mix with other generated signals for use in autotrading. This way the neural net signal can be used as a confidence bias instead of an absolute signal.

Other trade signals will be generated by a simple scripting interface giving access to all of the charted information, including: all technical overlays and underlays, user drawn trendlines, open/high/low/close/vol, exchange-lag.

The scripting will initially just be able to spot crossings between any pair of signals mentioned and set boolean flags. These flags can be logically mixed with AND,OR.

This is a few months down the road. First off is to actually get the trading interface coded and tested.

Eventually I would like to open it up further and give user simple script API to access entire array history of signals mentioned and perform their own math etc.

Hi, quick question for understanding: I buy the 12 months now, i get the 6 month beta PLUS 12 months membership?

And: Do you have a good explanation somewhere that gets me an understanding of the mathematics behind NN trading?

[StarenseN]

Can't wait !

[yucca]

Hi, quick question for understanding: I buy the 12 months now, i get the 6 month beta PLUS 12 months membership?

And: Do you have a good explanation somewhere that gets me an understanding of the mathematics behind NN trading?

Yes, When you buy the 12 month subscription you get a 12 month key for use on the final polished release. You also receive free monthly beta keys for as long as Cortex7 is in beta phase, this is expected to last ~6 months, possibly more.

The artificial neural network (ANN) tries to predict if the chart is about to go up or down.

The ANN is first trained by showing it alot of history data and also telling it what points in that history would have been good buy/sell points.

After training the ANN is able to predict when to buy and sell with better accuracy than random walk.

Of course one must realise that this is not a dead cert indicator, but it is much better than random and can give further confidence to technical charting that you may do.

The ANNs in cortex7 are multilayer perceptrons with backpropogation training:

http://en.wikipedia.org/wiki/Artificial_neural_network

[Schrankwand]

Hi, quick question for understanding: I buy the 12 months now, i get the 6 month beta PLUS 12 months membership?

And: Do you have a good explanation somewhere that gets me an understanding of the mathematics behind NN trading?

Yes, When you buy the 12 month subscription you get a 12 month key for use on the final polished release. You also receive free monthly beta keys for as long as Cortex7 is in beta phase, this is expected to last ~6 months, possibly more.

The artificial neural network (ANN) tries to predict if the chart is about to go up or down.

The ANN is first trained by showing it alot of history data and also telling it what points in that history would have been good buy/sell points.

After training the ANN is able to predict when to buy and sell with better accuracy than random walk.

Of course one must realise that this is not a dead cert indicator, but it is much better than random and can give further confidence to technical charting that you may do.

The ANNs in cortex7 are multilayer perceptrons with backpropogation training:

http://en.wikipedia.org/wiki/Artificial_neural_network

Thanks for that!

I just read a study on artificial intelligence, where neural networks were taught to include technical analysis as well. Will this be implemented in the feature? It did increase their ROI from 11,X% to over 15% p.a. in Forex trading. That is quite the number there.

[yucca]

Thanks for that!

I just read a study on artificial intelligence, where neural networks were taught to include technical analysis as well. Will this be implemented in the feature? It did increase their ROI from 11,X% to over 15% p.a. in Forex trading. That is quite the number there.

That's an interesting idea, thanks for sharing.

The technical indicators are derived solely from the raw OHLCV price data. So if we show the neural network the same raw data then it should model internally any functions that might improve prediction. But I suppose presenting it with chosen indicators might give it a head start and reduce training time. It would need extra input nodes to read these indicators and we would need to feed it quite a few data points from each indicator line, and so iterating and training time would increase. Of course we might only feed it the indicator signals and this might reduce training time? interesting.

I plan to make a dedicated neural net "lab" window giving the user more control over training, what signals are used as input, and the desired output function we wish to train to. In this new lab window I will include options to include data from any indicator on the chart.

[Schrankwand]

Thanks for that!

I just read a study on artificial intelligence, where neural networks were taught to include technical analysis as well. Will this be implemented in the feature? It did increase their ROI from 11,X% to over 15% p.a. in Forex trading. That is quite the number there.

That's an interesting idea, thanks for sharing.

The technical indicators are derived solely from the raw OHLCV price data. So if we show the neural network the same raw data then it should model internally any functions that might improve prediction. But I suppose presenting it with chosen indicators might give it a head start and reduce training time. It would need extra input nodes to read these indicators and we would need to feed it quite a few data points from each indicator line, and so iterating and training time would increase. Of course we might only feed it the indicator signals and this might reduce training time? interesting.

I plan to make a dedicated neural net "lab" window giving the user more control over training, what signals are used as input, and the desired output function we wish to train to. In this new lab window I will include options to include data from any indicator on the chart.

Hey, i thought it might be cool to actually share the study instead of just talking about it:

http://www.dabi.temple.edu/~zoran/papers/chenoweth96.pdf

[chunglam]

Registered and paid through Bitpay, hope to receive key soon.

[StarenseN]

Registered and paid through Bitpay, hope to receive key soon.

Same here.

[yucca]

Registered and paid through Bitpay, hope to receive key soon.

Just sent you key out.

Bitpay keys I need to cut by hand and I did it as soon as I woke up, before I brushed my teeth!

[yucca]

Registered and paid through Bitpay, hope to receive key soon.

Same here.

I think you should also now have your beta key and 12mth key.

[theblazehen]

Tried running it in wine. Could not find KERNEL32.dll.

Time to start a windows VM...

[ewibit]

it would be nice if someone would share how good/bad the software works and how much profit/loss (%) in which time was made...

[StarenseN]

Hello,

- Will you implement a visual view of the order book depth (like a wall visualizer)?
- Will it be possible to draw some basic retracements (fibonacci at least) on the chart?
- I'm not sure to have properly trained my ANN. The tutorial is perfectible.

Anyway, that's a good work!

[yucca]

Hello, - Will you implement a visual view of the order book depth (like a wall visualizer)?

Yes, in time. First of all I need to get the order book straightened out. The current release is not correct because it does not modify the book based on the incoming trade and order stream. It just uses market depth query every few seconds. It should query market depth once and then use incoming stream.

- Will it be possible to draw some basic retracements (fibonacci at least) on the chart?

Yes, definitely.

- I'm not sure to have properly trained my ANN. The tutorial is perfectible.

Anyway, that's a good work!

There are some more tips on the forum, but yes the tuturial is very basic.

Things will be changing quite a bit during beta phase but when finished it will have a proper PDF manual, I promise!