Beta version of new Bitcoin price prediction model!

[davidgdg]

I don't much believe in TA, whether it's done by a human or a computer. But I must admit I'm impressed by the prediction made eight hours ago of a fall from 670 to 635 which was almost spot on both in terms of timing and degree.

[samurai1200]

I don't much believe in TA, whether it's done by a human or a computer. But I must admit I'm impressed by the prediction made eight hours ago of a fall from 670 to 635 which was almost spot on both in terms of timing and degree.

Were you looking at the chart here:
http://www.btcpredictions.com/Historic.php

or here?
http://www.btcpredictions.com/NewPredictor.php

For a few hours after i posted last, I checked in with the NewPredictor every once in a while, which said we'd be going through a rising swell today. The whales apparently weren't going to have that happen today. The insane thing is that the supposedly simpler predictor at Historic.php forsaw the crash 6 hours before it happened.

Got me sitting up straight and watching intently now.....

[davidgdg]

I don't much believe in TA, whether it's done by a human or a computer. But I must admit I'm impressed by the prediction made eight hours ago of a fall from 670 to 635 which was almost spot on both in terms of timing and degree.

Were you looking at the chart here:
http://www.btcpredictions.com/Historic.php

or here?
http://www.btcpredictions.com/NewPredictor.php

For a few hours after i posted last, I checked in with the NewPredictor every once in a while, which said we'd be going through a rising swell today. The whales apparently weren't going to have that happen today. The insane thing is that the supposedly simpler predictor at Historic.php forsaw the crash 6 hours before it happened.

Got me sitting up straight and watching intently now.....

The historic (3 layer once an hour). Uncanny.

[K128kevin2]

Yeah the 24 hour predictor on the home page has been doing really well recently... I'm almost tempted to just scratch the new predictor. The cool thing about the way this works is that the longer I keep it running, the more accurate it will become. As more data becomes available to train the neural network, it can learn more and the average error becomes lower.

[Raystonn]

Got a neural network to tell folks which neural network they should follow at any particular moment?  Looks like they give conflicting calls.

[K128kevin2]

Got a neural network to tell folks which neural network they should follow at any particular moment?  Looks like they give conflicting calls.

The 24 hour one - it has the lowest average error. The 20 day one is unreliable, as it says above the chart. The 5 day one does OKAY but it's not as good (as is reflected in the higher average error)

[Gingermod]

Got a neural network to tell folks which neural network they should follow at any particular moment?  Looks like they give conflicting calls.

The 24 hour one - it has the lowest average error. The 20 day one is unreliable, as it says above the chart. The 5 day one does OKAY but it's not as good (as is reflected in the higher average error)

You should honestly remove the 5-20 day ones, they just mislead noobs.

Is your 24 hour one updating in real time? I imagine it would be more effective if it corrected mistakes on the spot.

http://puu.sh/9eM83/b90ecc457d.png

As you can see above there is an increasing disparity between prediction and actuality, but no correction or timing adjustment. However, you did more or less see that recent drop coming, very inspiring.

[K128kevin2]

You should honestly remove the 5-20 day ones, they just mislead noobs.

Is your 24 hour one updating in real time? I imagine it would be more effective if it corrected mistakes on the spot.

http://puu.sh/9eM83/b90ecc457d.png

As you can see above there is an increasing disparity between prediction and actuality, but no correction or timing adjustment. However, you did more or less see that recent drop coming, very inspiring.

I think I want to keep those ones still because they are interesting and a lot of people were asking me for longer term predictions when the 20 day one didn't exist. However, I am definitely going to keep trying to improve them.

And yeah, I'm hoping to solve the issue of it taking a bit of time to respond to changes with the new predictor. That's why I'm having it update every 5 minutes. I just uploaded a new neural network for the beta so we'll see how that one works.

[Gingermod]

You should honestly remove the 5-20 day ones, they just mislead noobs.

Is your 24 hour one updating in real time? I imagine it would be more effective if it corrected mistakes on the spot.

http://puu.sh/9eM83/b90ecc457d.png

As you can see above there is an increasing disparity between prediction and actuality, but no correction or timing adjustment. However, you did more or less see that recent drop coming, very inspiring.

I think I want to keep those ones still because they are interesting and a lot of people were asking me for longer term predictions when the 20 day one didn't exist. However, I am definitely going to keep trying to improve them.

And yeah, I'm hoping to solve the issue of it taking a bit of time to respond to changes with the new predictor. That's why I'm having it update every 5 minutes. I just uploaded a new neural network for the beta so we'll see how that one works.

I'll be keeping an eye on it.

[natewelt]

According to the historic 24 hour predictor...we should start seeing a ramp-up in price here in the next couple hours.

Let's see what happens... 

[K128kevin2]

According to the historic 24 hour predictor...we should start seeing a ramp-up in price here in the next couple hours.

Let's see what happens... 

It's been pretty accurate about the major price movements over the past several days... I would not be surprised at all if it is right again here. We'll see, hopefully it is right!

[MOB]

No neural network is required to see that we are about to either go back up to the 680 range (or higher finally), or way down.  At a bit of a crossroads.

That said, I hope your network is correct on this one.  I have some of those 623 coins to unload!


Also, why not have the network predict Huobi instead?  if Bitstamp is tracking it, then you are just compounding an error.  Do you think that Bitstamp is not tracking Huobi?

[knightcoin]

is there white paper or source code available ?

[K128kevin2]

No neural network is required to see that we are about to either go back up to the 680 range (or higher finally), or way down.  At a bit of a crossroads.

That said, I hope your network is correct on this one.  I have some of those 623 coins to unload!


Also, why not have the network predict Huobi instead?  if Bitstamp is tracking it, then you are just compounding an error.  Do you think that Bitstamp is not tracking Huobi?

I just chose Bitstamp because it's a major exchange and there is a lot of historic data available. I don't think it would make a huge difference which exchange I used.

is there white paper or source code available ?

The source code is proprietary for the time being, sorry!

[knightcoin]

No neural network is required to see that we are about to either go back up to the 680 range (or higher finally), or way down.  At a bit of a crossroads.

That said, I hope your network is correct on this one.  I have some of those 623 coins to unload!


Also, why not have the network predict Huobi instead?  if Bitstamp is tracking it, then you are just compounding an error.  Do you think that Bitstamp is not tracking Huobi?

I just chose Bitstamp because it's a major exchange and there is a lot of historic data available. I don't think it would make a huge difference which exchange I used.

is there white paper or source code available ?

The source code is proprietary for the time being, sorry!

faith is required then ... ok, wish lucky then...

[K128kevin2]

awww it changed its mind

[Raystonn]

So according to the 24-hour prediction chart, we're supposed to be at 687.62 in 6 minutes?

[MOB]

Sweet, time to buy!

[knightcoin]

to be honest you need to show your method, the secret if you have one such as special setup, or secret number you can keep but ... it's like have a scientific approach to data analysis ...

like professor David Aronson says on
Scientific Approaches to Algorithmic Trading, Part 1 of 4
http://youtu.be/AgOxoMMillM?t=4m15s  

simulate + backtest + cross validation

if you just says that the future price is "X" because is based on neural network simulations sounds very vague to me ... I used to play a lite bit with Jeff Heaton's Encog framework .. and boy ... so much ways to simulate something within neural networks ...

of course in the end of the day in a imperfect information game ... we need some faith ... but to sell a product for financial market .... is better backup you product with a good maths and strong hypotheses  ..

[Gingermod]

to be honest you need to show your method, the secret if you have one such as special setup, or secret number you can keep but ... it's like have a scientific approach to data analysis ...

like professor David Aronson says on
Scientific Approaches to Algorithmic Trading, Part 1 of 4
http://youtu.be/AgOxoMMillM?t=4m15s  

simulate + backtest + cross validation

if you just says that the future price is "X" because is based on neural network simulations sounds very vague to me ... I used to play a lite bit with Jeff Heaton's Encog framework .. and boy ... so much ways to simulate something within neural networks ...

of course in the end of the day in a imperfect information game ... we need some faith ... but to sell a product for financial market .... is better backup you product with a good maths and strong hypotheses  ..

yes ... but .... maybe .... his .... neural ... network ... is .... in ... too ... early ... of ... a ... stage ... to ... show ... off