Machine learning in trading: theory, models, practice and algo-trading - page 535

New comment
 
Maxim Dmitrievsky:

Well, the feeling is 4-5 seconds

And I have 25 (on real ticks). Plus some more time to prepare ticks in the first run, but it can be ignored.

 
Dr. Trader:

And I have 25 (on real ticks). Plus some more time to prepare ticks in the first run, but it can be ignored.


that's about97555367 ticks :) not so much

your computer's fast, my real ticks take over a minute, it feels like

 
Dr. Trader:

The topic is interesting in itself, but it has not been tested on forex. There were some articles about it in the thread, even there is a package for R -https://github.com/ahunteruk/RNeat .
NEAT a couple of words - we select neuronkey weights using genetic algorithm instead of conventional training.
Here's an example of the algorithm in action, the neuronka is trained to play a Mario gamehttps://www.youtube.com/watch?v=qv6UVOQ0F44

If with normal neural network training you can sometimes pause training and check overfit on new data to stop training in time, with NEAT it won't work, the genetics will search for weights that best satisfy the fitness function until it reaches its limit, resulting in strong overfit and useless model on new data.

This is not true at all. NEAT(NeuroEvolution of Augmenting Topologies) is a genetic search for optimal neural network architecture. Exactly the architecture, not the NN weights of a given architecture. Unfortunately the package was not continued in recent releases of R. There is a similar package in Python.NEAT is a method developed by Kenneth O. Stanley to develop arbitrary neural networks.NEAT-Python is a pure Python implementation of NEAT without any dependencies other than the standard Python library. You can read in more detail - Our original journal on NEAT (co-authored with Ken Stanley and Risto Miikkulainen), "Evolution of Neural Networks through Complementary Topologies"

A small excerpt from :Evolving Neural Networks through Complementary Topologies (2002)

Kenneth O. Stanley andRisto Miikkulainen
An important question in neuroevolution is how to benefit from evolving neural network topologies along with weights.We present the NeuroEvolution of Augmenting Topologies (NEAT) method, which outperforms the best fixed topology method on the complex artifact learning task.We argue that the improved performance is due to (1) using a principled method of crossover of different topologies, (2) protecting structural innovation using speciation, and (3) gradually increasing from a minimal structure.We test this claim through a series of ablation studies that demonstrate that each component is necessary for the system as a whole and for each other.Which results significantly accelerate learning.NEAT is also an important contribution to GA because it shows how one can evolve to both optimize andcomplicate solutions simultaneously, offering the possibility of evolving increasingly complex solutions over generations and reinforcing the analogy with biological evolution.

Notice the difference? We need to experiment. There would be 26 hours of time in a day...

 
Maxim Dmitrievsky:

the standard MACD sample Expert Advisor with simple logic, 1 minute at 2 opening prices for the year... well it felt like 4-5 seconds... and that for the year on the minutes

for me it's not that slow + he reproduced the trading environment such as floating spreads, drew a chart and showed the report

The guys are growing, what can I say, 4 years ago it was much slower. But anyway 4-5 sec is an eternity for one run, it should be two orders of magnitude faster. 4-5 sec, over a one year interval, this "strategy" should be optimized by genetics or by a burnout of 100-200 runs.

 

I'm a fucking programmer. For four hours I tried to make an AD indicator for MT5 using CDs, but I kind of did it. This is a mess, comrades. I got lost in three lines :-(

It's just difficult when you don't know and you've forgotten :-)

 

You won't believe it, but an idiot's dream has come true, I've run the three main components of the market for optimization. Delta + Volume + Open Interest. I can't wait to see the results of the training...

 
Mihail Marchukajtes:

You won't believe it, but an idiot's dream has come true, I've run the three main components of the market for optimization. Delta + Volume + Open Interest. I can't wait to see the results of the training...

What do you mean "Open Interest"?

"Delta what?

 
SEM:

What did you mean by "Open Interest"?

"Delta what?


Open interest with Forts, delta with KD. I've got a vinigrette like this.... Let's see what comes out of this salad.......

 
Mihail Marchukajtes:

Open interest with Forts, delta with KD. The vinaigrette I have got such.... Let's see what comes out of this salad.......


I have a question. Why these parameters will give you an advantage over other market players, if this data is already known and probably much earlier?

 
Mihail Marchukajtes:

Open interest with Forts, delta with KD. The vinigrette I have got such.... Let's see what comes out of this salad.......

Try adding standard deviations from Bollinger Bands or Envelopes, for channel boundaries, interesting stuff comes out.

"Open interest from Forts", I wonder who is broadcasting the real data on these indicators?

Again, I don't understand what "QD" means?

1...528529530531532533534535536537538539540541542...3492
New comment
Reason: