Machine learning in trading: theory, models, practice and algo-trading - page 549

[Maxim Dmitrievsky]

Here is another example of prophet but for R (in python I never got it)

i think that if you transform data instead of quotes, it will predict better than arima

and compare it with arima for interest

https://strategy.doubledoji.com/how-to-use-facebooks-prophet-algorithm-in-forex-trading/

[Maxim Dmitrievsky]

here's another great piece of research:

https://msperlin.github.io/pafdR/research-scripts.html

[СанСаныч Фоменко]

Maxim Dmitrievsky:

Here is another example of prophet but for R (in python I never got it)

i think that if you transform data instead of quotes, it will predict better than arima

and compare it with arima for interest

https://strategy.doubledoji.com/how-to-use-facebooks-prophet-algorithm-in-forex-trading/

It should be tested on the arch. There are timeseries that don't have the arch effect, but if it is present, then it needs garch, and it is necessary to pick up a certain one (there are many of them) and then it is very useful to model distribution.

[Vladimir Perervenko]

Below are two excerpts from my article, which I am finishing up. There's no need to waste your time reinventing the bicycle. You won't make it any better than the professionals. Python and R are already wonderfully integrated. Just use.

"Using the TensorFlow library."

The burgeoning field of deep neural networks has recently been enriched by a number of open source libraries. The widely advertisedTensorFlow(Google),CNTK(Microsoft),Apache MXNet and many others. Due to the fact that all these and other major software developers are part of the R Consortium, all these libraries are provided with an API for R.

All of the above libraries are very low-level. For beginners to learn this area, they are difficult to master. With this in mind, the Rstudio team developed thekeras package for R.

Keras is a high-level neural network API, designed with a focus on being able to experiment quickly. The ability to go from idea to result with the least possible delay is key to doing good research. Keras has the following key features:

• Allowing you to run equally on the CPU or on the GPU.
• Friendly API that makes it easy to prototype deep learning models.
• Built-in support for convolutional networks (for computer vision), recursive networks (for sequence processing), and any combination of both.
• Supports arbitrary network architectures: models with multiple inputs or multiple outputs, layer sharing, model sharing, etc. This means that Keras is suitable for building essentially any deep learning model, from a memory network to a Turing neural machine.
• It's capable of running on top of multiple back-ends, including TensorFlow, CNTK or Theano.

Just install and download the keras R package, then run the keras :: install_keras () function, which installs TensorFlow, Python, and anything else you need, including aVirtualenv orConda environment. It just works! For instructions on installing Keras and TensorFLow on GPUs, see.Here. Seethe article for more details."

"The tfruns package is for experimenting with TensorFlow. Thetfruns package provides a set of tools for managing TensorFlow training and experiments from R:

• Track the hyperparameters, metrics, outputs, and source code of each training cycle.
• Compare hyperparameters and metrics between runs to find the most efficient model.
• Automatically generate reports to visualize individual training runs or comparisons between runs.
• No changes to the source code are required (run data is automatically captured for all Keras andtfestimators models).

The best quality visualization of the process and results of DNN training is provided byTensorBoard.

And, of course, deep-learning connoisseurs have the opportunity to work directly with the low-level TensorFlow library using the tensorflow package.

All of these packages are based on the basic -reticulate- R interface to Python modules, functions and classes. When called in Python, R data types are automatically converted to their equivalent Python types. When values are returned from Python to R, they are converted back to R types. Worthy of careful study.

All of these packages are well documented, which is not surprising given the developer class, are packed with tons of examples, and are constantly evolving. Thus we have a unique opportunity to use in Expert Advisors and Terminal Indicators the most advanced and advanced deep learning models (DNN, RNN, CNN, LSTM, VAE, etc.), reinforcement learning (RL) and many other developments of Python in the field of machine learning, if you have enough knowledge and experience".

Good luck

[Maxim Dmitrievsky]

at the same time, the complexity of strategies and the amount of work becomes too much for 1 person who is not professionally engaged in date satanism :) although this may be only at 1st glance

[Grigoriy Chaunin]

Almost finished integrating Python and MT. I had to write an additional DLL. There are a few small things left. Finish the mqh and write a doc. The library doesn't have a lot of features, but it's enough. It is possible to load and execute a Python script. And also call any function from it. It will work with lists, but only one-dimensional and homogeneous. I.e., the MQL array is converted into a list and back.

[Grigoriy Chaunin]

Machine Learning from Yandex https://habrahabr.ru/company/yandex/blog/333522/

[Maxim Dmitrievsky]

But the tutorial on R is good... from basics and linear models to Garch and Facebook prophet...(it's not for nothing that I'm throwing prophet here, because it arouses interest in certain circles and is mega easy to use) almost the same as for python, but in more detail, from the teacher on R

in general it is logical, of course, to go from econometrics to neural networks, and not vice versa, that is, to study what you already have (if you have not studied at university), to have something to think about, and then to sink into grids

And in general, note that neural networks stand apart in the econometrics and are not the main subject (yet)

https://msperlin.github.io/pafdR/

[СанСаныч Фоменко]

Maxim Dmitrievsky:

In general, it is logical, of course, to go from econometrics to neural networks, and not vice versa, i.e. to study something that already exists (if you have not studied it at university), to have something to think about, and then to get stuck in networks

And in general, note that neural networks stand apart in the econometrics and are not the main subject (yet)

https://msperlin.github.io/pafdR/

Some years ago I was interested in Matlab and, contrary to my ideas about the tools of econometrics, the toolbox called "Econometrics" contained ONLY GARCH models.

Ended up doing MO. What surprised me the most was very poor publications on application of MO to financial markets.

Recently, I returned to GARCH and was astonished of great number of publications on application of GARCH to financial markets: stock exchanges, indices, futures and currency pairs. It's just a ball of fun.

So maybe matlab is right? Maybe all IO instruments, including NS are third-party tools for financial markets?

[Maxim Dmitrievsky]

SanSanych Fomenko:

So maybe matlab is right? Maybe all of the IO tools, including NS are third-party tools for financial markets?

Well, yes, they are not some ready-made econometric models, but just a set of tools universal for any area

Nonparametric econometrics is just about IR and fuzzy logic, but I haven't seen some clear-cut things, maybe because some common approaches haven't been worked out. I don't really know how to work :).