Machine learning in trading: theory, models, practice and algo-trading - page 41
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
to train a neuron which will give the opposite result to what you want, it will turn out like yours, but on purpose. But it is not clear why to do it? :)
I did not do it on purpose, I just trained it and see that )))
The recipe is this.
1) train two networks separately for buy and sell. I took the "deep neural" network. It's important because its output is not 01010101, but the probability of 01010
2) we simply take the outputs of two nets and calculate the difference of cumulative sums and get what is shown on the picture...
In the first picture, probabilities are displayed below (outputs of two networks as is) and the cumulative difference (as is) is displayed above together with the price
on the second picture is the same, just shift the sign of the cumulative difference (i.e. turn it upside down) for better clarity
=======================================
If this process is predicted for a few candlesticks forward, at least for one, it is a grail)) I have not yet succeeded
all day waiting for comments, is no one interested???
The question itself is not clear.
Take caret. Almost 2 hundred models. If you have not started to master, I recommend it. There are not only models, but a lot of other very useful things.
If you want it, it predicts everything several steps ahead.
Regression models accumulate error from previous steps.
For classification models, the error grows, but not due to the number of steps ahead.
Note that for time series it is always possible to enlarge the TF and predict by one step, which will correspond to predicting by several steps on a smaller TF.
Need some kind of tool that could predict a few steps ahead using a set of predictors
mytarmailS:
Need some kind of tool that could predict a few steps ahead using a set of predictors
Check out rminer. Very well organized package
Hello!
I have two questions for you
1) with regard to recurrent networks from the previous page, I threw the package "rnn" it really is some kind of strange and ill-conceived and switched to "RSNNS" took the network "Elman" threw the data all workedI really don't know where and how to set "how far the network remembers itself in the past" in "rnn", but it was immediately clear that each variable turned into a matrix with a number of columns that corresponded to the size of the network. Columns corresponded as if to network memory size, but I haven't found such parameter in"RSNNS", but it's impossible that it wasn't there as the essence of recurrent network is exactly in it.
2) What should I look for in"rminer"? If you mean something like "arima" then it is not suitable already discussed.
I've tried to write such a classifier target to make several step forecasts, but I failed to forecast it, the results were strange, in the first place the quality fell down (it's normal because the forecast is not on the 5th or 10th candle), anti-correlation was still present even if less expressed, but the trick is in the outrunning effect of the rminerIt's not clear, I've got the same indicator reversals where the market reversals were, so I've got something worse, why it's not clear to me?..
Let's talk a little bit about qualitative trait selection. my thoughts on this...
Don't beat yourself up! I'm going to exaggerate, but it's a good example...
All conventional methods of attributes selection (for the market) reduce to putting 100 indicators in a training sample and using different methods of different "degrees" to select, say, 20 indicators that best describe the price, but isn't this method too primitive, despite the power of advanced methods for attributes selection? I think it is coarse and primitive, and the selection methods have nothing to do with it, I'll try to explain why...
Let's take a break and think how a professional trader works(remember, I'm still exaggerating :)) who has only two signs - levels and "RSI" indicator And there is a trading system in which a sell trade sounds like this: if the level is broken through upwards and RSI is greater than 0.9, then the sale...
In this case, the trading system acts as a data filter, a filter that does not let the trader enter in noise, and the share of noise in this example with RSI is not a joke 95% because RSI range from -1 to 1, and the trader only needs what is >0.9 those 5% ...
So what follows from this is that God willing there is 5% of useful information in each characteristic and the rest 95% is noise, noise that is not necessary, that prevents learning, noise that must be removed... In our example with RSI it's a range from 0,9 to 1, all the rest is garbage...
And now let's go back to the beginning of the post, you have 100 attributes (indicators), all you can (however cool the method may be) is to pick the best indicators, realizing that every indicator has 95% of noise, what do you think is the quality of selection of these attributes?
Conclusion: it is necessary to take each trait and to extract from it useful, I have ideas how to do it but before to sound it I would like to hear your thoughts, ideas and offers in this regard