Good afternoon. Could you, please, recommend me packages and functions in R, which can help me to fit neural networks (especially LSTM networks) to time series data? Particularly, time series represent
1 - minute logarithmic returns of securities. The dataset is rather large and involves about 25000 values.
In addition, I`d be very pleased if you provide me with some training manual or examples of such type of work.
Thank you for your help.
I really don't have too much experience with neural networks, but from what little I have dabbled in I can definitely recommend the keras package and the book Deep Learning with R by Francois Chollet and J.J. Allaire.
Here's a nice walkthrough from the RStudio AI blog:
To get up and running, there is excellent info on:
https://keras.rstudio.com/ and https://tensorflow.rstudio.com/
Thank you for your help!
Are keras and tensorflow the only packages, aimed at creating neural networks in R?
No definitely not, nnet and neuralnet are two other popular packages, although I don't have any experience with them. nnet can actually be used as the computation engine for both caret and parsnip if you have experience with either package 
To add to that, according to the CRAN Task View (which is a great resource to browse packages by what you want to do), there is also RSNNS and deepnet.
https://cran.r-project.org/web/views/MachineLearning.html
Also, looking at Task View for machine learning, nnet is listed as a "core" package which means that a lot of other packages depend on it. Perhaps nnet could then be of interest for you 
Thank you for your advice!
Thanks for sharing, this is super interesting.
This topic was automatically closed 21 days after the last reply. New replies are no longer allowed.