Machine learning in trading: theory, models, practice and algo-trading - page 284
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
Half a branch shedding light: predictors have no predictive power and are noise to the target variable. That's why the model is retrained, and the retrained model has NOTHING to do with its future use. NOISE IS NOISE ALL THE SAME, IN ONE APPLICATION THERE IS ONE RESULT, AND IN ANOTHER THERE IS ANOTHER.
Y Well, actually it was about the classifier. So what. Generally speaking, in order to prognosticate, we need to build an indicator of price changes over 10 bars. Shift it back one bar. This will be the target function. It is sufficient to train the network so that the error between the network output and the target function is minimal, i.e. the network should react to the input data as a lead from persentence10%. What is useful for me is that I work with the whole indicator and there is an opportunity to make infinite attachments of an indicator from an indicator. I have such an idea. The thing is that there is a very interesting classification network. What if to train some such grids in NS on a certain area, and then to unload these values to Reshetov's optimizer, and to see, whether it will be possible to increase the level of generalization. Because here, as I understand it, deep learning is obtained, when the network input is not the input itself, but the result of the network on these inputs. Folks, do I understand the concept of deep learning correctly?
When first we train a network on input data, then the result of several networks trained on the same data is fed to the input of another network, thereby obtaining a better level of generalization. Is this true, Guys????
Vladimir, please take a look at my post about Twitter, a few pages earlier I wrote..... Maybe you can help me with it
Y Well, actually it was about the classifier. So what. Generally speaking, in order to prognosticate, we need to build an indicator of price changes over 10 bars. Shift it back one bar. This will be the target function. It is sufficient to train the network so that the error between the network output and the target function is minimal, i.e. the network should react to the input data as a lead from persentence10%. What is useful for me is that I work with the whole indicator and there is an opportunity to make infinite attachments of an indicator from an indicator. I have such an idea. The thing is that there is a very interesting classification network. What if to train some such grids in NS on a certain area, and then to unload these values to Reshetov's optimizer, and to see, whether it will be possible to increase the level of generalization. Because here, as I understand it, we get deep learning, when the network input is not the input itself, but the result of the network on these inputs. Folks, do I understand the concept of deep learning correctly?
When first we train a network on input data, then the result of several networks trained on the same data is fed to the input of another network, thereby obtaining a better level of generalization. Is this true, Guys????
=========================================================
No it isn't. What you are describing is stacked NN. And deep learning is something else entirely...
Good luck
Vladimir, please take a look at my post about Twitter, a few pages earlier I wrote..... Maybe you can help me with that
So you could explain in two words what it is, at least approximately...?????
I read your post, I can not help, as I have not dealt with text data. I have seen many examples. If I find the link, I will drop it.
The problem is that I myself can not run the package, I can not connect with Tweeter, and even easier I can not set up a connection, at least example is given how to do it, but there is confusion with the pin code, I am stupid can not understand where he should enter it
Vladimir, please take a look at my post about Twitter, a few pages earlier I wrote..... Maybe you can help me with that
Look at https://github.com/maxbbraun/trump2cash
PS It's a very big task, with lots of pitfalls, to make a machine-readable analysis of news streams by yourself. I recommend https://www.accern.com/ to try it, I use it, very satisfied.
scale() is not suitable it with its tricky normalization constantly makes different ranges...
Thanks to everyone who tried to help
Needed to map to a range myself. Since I did it earlier, I looked it up. I gave the wrong link. There is a scales package (not a function) and it is full of all kinds of scales. For you, I think recsale is the right one for you. For example.
Displays in the specified range. And a bunch of other similar functions in the above package
Are you talking about deep learning or stackedNN?