Machine learning in trading: theory, models, practice and algo-trading - page 869
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
I have already learned about 2 NS in practice, and this is the only way to use them. If I have 3 classes (buy, wait, sell), then the middle class slips through very quickly, especially if the output neuron is sigmoid or tangent.
But if regression... Ideally 1 output neuron is needed.
Each NS - 2 classes - long/0, the second NS - short/0.
Just in case, I repeat one of the results of the NS training - this is still a test from 09.17. Something like this comes out.
Say, more than 0.5 - long, less - 0, we stand.
No, I just use scaffolding so far (an ensemble of any number of models, on different features), the output is average
it all works very fast
So, is the tree analogous to the TF? Something tells me it's not.
I have no idea... went to drink bacardi )) I'll think about it later
I plan to try NS, even just logistic regression, the main thing is that it should be fast, because it's multiagent environment, many models are trained at once... but I adapt RL to my problems
I also really want to use clustering and have thought of how, but alglib doesn't allow to get centroid membership on new data, I will have to take a different lib somewhereMm-hmm. The NS input is a normalized time series. Let's say the NS structure -15-20-15-10-5-1 is already doing well.
To determine the longs and shorts one needs two NS.
I wish I could find an algorithm for calculating network structure... so that it would be sufficient and not superfluous.
I don't know. In classical literature, only by intuition or selection.
I did it with about the 3rd or 4th time. By the results of the training you start to feel them).
I have no idea... went to drink bacardi )) I'll think about it later
I'm going to try NS, even just logistic regression, as long as it's quick, because it's multiagent environment, many models are trained immediately... but I adapt RL to my problems
I really want to use clustering and I've found out how, but alglib doesn't allow to get centroid membership on new data, I'll have to take a different lib somewhereTime to switch to R, I tried alglib NS - dozens of times slower counts the same network as in R (like a day against 30-60 minutes). Plus in alglib maximum 2 hidden layers, and according to your observations you need 3 consecutive conversions, i.e. 3 layers.
Plus in alglib maximum 2 hidden layers, and according to your observations you need 3 consecutive conversions, i.e. 3 layers.
3 layers is nothing at all.(
I was a little wrong about the structure of my NS there, by the way
it was15-20-15-10-5-1
Should be - 15-15-20-15-10-5-1. The first figure - the number of inputs.
3 layers is nothing at all.(
I was a little wrong about the structure of my NS there, by the way
it was15-20-15-10-5-1
Should be - 15-15-20-15-10-5-1. The first figure - the number of inputs.
Total of 1030 scales is not weak. And how many input data (lines)?
A total of 1030 scales is not weak. And how many inputs (lines)?
Excuse me? What do you mean? The NS has 15 inputs - the 1st digit of the NS structure. The input is directly normalized BP - 15 counts.
Excuse me? What do you mean? The NS has 15 inputs - the 1st digit of the NS structure.
For example 10000 lines of 15 inputs