Machine learning in trading: theory, models, practice and algo-trading - page 713

 
Alexander_K2:

Eh, drunkenly my eye didn't figure it out yesterday... I wondered how it could be. Because I have the Grail.

I suggest the moderators clean up the thread.

And yes, people's thirst is unquenchable, no matter how smart they are. Yesterday we all saw that.

That's what it was about, no need to clean it up.

 
Renat Akhtyamov:

But there are envelopes there, and here there are CMAs...

And now neural networks:


How do you compare?

An ordinary CMA machine did it all!!!

There's one thing.

Your indicator with a shifter 1 bar backwards. I.e. a new bar will start, you need the forecast, and there is no forecast, it will appear only in an hour, and it will be too late to trade by it because a new bar will start and there is no forecast for it again.

And Vizard_ has this forecast at the BEGINNING of a new bar. That's what the forecast is.

The difference is huge.

 

Finally I did some regression on the NS.
I decided to check whether the NS really remembers all the examples, if there are a lot of neurons. I fed 1440 examples for 1600 neurons.

Got Train set (best model) RMSE: 1.701 - error in pt.
And here are some teacher and prediction values.

[,1]
[1,] -3.00000000000000
[2,] 9.00000000000690
[3,] 5.99999999999635
[4,] 11.00000000000290
[5,] 17.99999999999269
[6,] 6.99999999999635
[,1]
[1,] -1.98802261182071
[2,] 9.70103773798503
[3,] 5.57447626000446
[4,] 10.58770172551591
[5,] 17.71941468771552
[6,] 6.56410167439433

Really got it all memorized. I'll move on from that to generalization.

 
Dr. Trader:

There is one nuance.

Your indicator with a shifter on 1 bar backwards. I.e. a new bar will start, you need the forecast, but there is no forecast, it will appear only in an hour, and it will be too late to trade by it because there will be a new bar, for which there is no forecast again.

And Vizard_ has this forecast at the BEGINNING of a new bar. That's why it's a forecast.

There's a huge difference, not even close to the CMA.

All right, let's move on.

The indicator may be accurate but it won't help, because I started this "flood" ( as some people thought ) with it:

Forum on trading, automated trading systems and testing trading strategies

Machine Learning in Trading: Theory and Practice (trading and not only)

I am a beginner, I am a professional trader.

The main thing is to understand - if you open a sell, the price will go against the sell, if you open a buy, the price will go against the buy. It is impossible to find any dependence. The market is tough, that's a good place to start.

The neural network is trying to find dependencies, but the price will still go against the higher volume and certainly against the nearest highest risk and nothing else

Forum on trading, automated trading systems and trading strategies testing

Machine learning in trading: theory and practice (trading and not only)

I am a professional trader, Iam very skilled in trading, and I am very good at it.

I mean you're in the wrong place, no more.

good luck!

=======

In Alexander_K's thread I wrote -

H4 forecast - only for 4 hours

On MN1 - only for a month, etc.

Ponder

And that says it all, as a bottom line.

The stack is in the branch.

Gone.


 
Dr. Trader:

I'm for boosting too, but I disagree a bit with his fight against overfit.

For him, defeating overfit = specific tree partitioning criterion + minimum number of trees.

For me: gbm + k-fold. I've stopped at gbm now, but I tried other packages in R - xgboost, catboost - they have built-in overfit protections, but I didn't get as much joy as with gbm.

gbm is the most nice model, most efficient - everything is right in the ideology

 
Renat Akhtyamov:

Okay, let's move on.

No matter how accurate the indicator is, it still does not help, because I started this "flood" (as it seemed to some) with this:

And that says it all, as a result.

The state's in the branch.

I left.


A picture instead of an indicator, a picture instead of the state. And instead of profit - a picture of money? :)

We here believe only in real signals and investment passwords.

 
Dr. Trader:

A picture instead of an indicator, a picture instead of a state. And a picture of money instead of profit? :)

We only believe in real signals and investment passwords.

We will. But the actual straight from the real one.

The indicator was shown above and so were the settings.

Is there any way to feed this MA into a neural network for analysis or is there no point?

 
Renat Akhtyamov:

Gone.


Even here you lie)

 
elibrarius:


Even here you lie)

Just keeping the topic going, questions are coming. It didn't work out. Sorry.