"New Neural" is an Open Source neural network engine project for the MetaTrader 5 platform. - page 68
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
Let me add a little.
I as the end user of the black box need the following.
I put into it the last 20-1000 bars, a few symbols.
In response the black box says . Stable state of the cluster (flat) is observed on the last 15 bars .
These clusters are in the history period from January 1, 1995 to January 20, 95 etc. We can highlight the graph.
Minimal lifetime of the cluster is 20 bars, maximal 74 bars, average 47 bars, observed 125 times throughout the history.
The recommended strategy is to trade from the border of the channel, the channel levels are 1.2567-1.2687.
Or.
Steady state of the cluster (flat) is observed on the last 65 bars.
These clusters are in the history period from 1.01.1995 till 20.01.95 etc, we can illuminate the chart.
The minimum life time of the cluster is 20 bars, maximum 74 bars, the average 47 bars, the entire history was observed 125 times.
The recommended strategy is to break through the channel, the channel levels are 1.2567-1.2687.
I am puzzling over it now. It can be done on the NS.
"Reverse" is needed to find partial derivatives of a fitness function by adaptive arguments, i.e. any gradient methods (for example, BackProp in any modifications) require "reverse"
The other methods do not.
Am I correct in assuming that there is no such thing as a forward move of adjusting weights?
and algorithms that use a forward move in training actually use several (concurrently existing) weight arrays,
By choosing the best one among them.
or are there still such algorithms that use the direct stroke of the weights adjustment?
Am I correct in assuming that there is no such thing as a straightforward stroke of adjusting the scales?
Algorithms that use a direct path in training actually use several (parallel existing) weight arrays,
by choosing the best one among them.
Yes, that's correct.
I dare say even harder - there is no such thing as a backward move of the scales adjustment either
"The reverse course is our visual perception of the process of finding a derivative of a complex function, which in essence is a neural network
Learning is a process external to the network itself
Different methods of training impose different requirements to the topology of the network and to the form of the estimated function
Gradient methods are the most demanding, stochastic methods are omnivorous
Thoughts aloud...
It occurred to me that a project in this format is almost doomed.
First, because everyone is pulling in their own direction. Someone wants the project to do everything for him, someone wants the project to assemble the topology itself, someone wants everything to fly in the third space as a target.
Second, because of the confusion with the controls.
Third, and probably most importantly, because of the lack of clear goals of the project so far.
___________________________
So, I figured out, do I really need it? I'd better split off and promote echo networks as a separate project.
Not a plan to take over the world, but functional and effective.
So good luck and good luck. Hope the project doesn't get buried completely.
___________________________
Sorry if anything is wrong.
Thoughts aloud...
Thoughts aloud...
It occurred to me that a project in this format is almost doomed.
First, because everyone is pulling in their own direction. Someone wants the project to do everything for him, someone wants the project to assemble the topology itself, someone wants everything to fly in the third space as a target.
Second, because of the confusion with the controls.
Third, and probably most importantly, because of the lack of clear goals of the project so far.
___________________________
So, I figured out, do I really need it? I'd better split off and promote echo networks as a separate project.
Not a plan to take over the world, but functional and effective.
So good luck and good luck. Hope the project doesn't get buried completely.
___________________________
Sorry if anything is wrong.
In general, how are Open-Source projects implemented?
Everything usually comes from one evangelist (notional Torvalds), who writes the engine and develops the basic ideology of the project. Then, as the project expands, new forces interested in the end result join. And here it turns out that you have to start with a leaf, and therefore the direction is not visible, because there is none. The best option - to try to develop the project alone, or in a very cohesive team, not separated geographically. And then, if the engine will be interesting, people will pull up. I think this is the only option that can survive in these conditions.
Thoughts aloud...
Andrew, no one prevents you from doing three projects in one branch.
Since all the branches in fact deal with the same thing, it would be useful to publish and exchange solutions within one branch.
Now we have three directions, code generation, connection of different implementations, and a universal engine.
Connecting different implementations (that's what you want to do) would be very useful for both the code generator and the universal engine.
The universal engine would be useful for the code generator. And the code generator (as the MQL master) will not help either direction, although it combines the simplicity for the end user and the speed (those are all the best qualities of parallel branches).
To avoid confusion we can use abbreviations, GC RR UD, and put an abbreviation for each post in its title or, for example, to distinguish posts by category by GC RR UD colors
I wish they would spit, but they are ignoring me. You were asked for advice - yes or no.
If (YES) I have to go read some smart books;
else go and get another one, and kick you in the right direction;
Urain:
Now we have three directions, code generation, connecting different implementations, and a universal engine.
I'd rather deal with one implementation and try to improve it to perfection.
I may not be able to compete with genetics in scale (those classes of tasks, which don't have clearly specified target function), but in terms of efficiency of using and learning...
Thanks :)
I'd rather take care of one implementation and try to bring it to perfection.
I may not be able to compete with genetics in scale (those classes of problems that do not have a clearly defined target function), but in terms of efficiency of use and training...
Very correct decision, until some project appears here you will already have done a lot of groundwork, and if the project does not appear, you will give a result.
Them with this approach is more likely that the project will not stutter.
Just do not disappear, leave developments, ask questions, so that someone else will buckle.