He barely mentiones R in that video, in only a few words. He mostly talks about HFT and even as an HFT guy says that HFT are often bad for the investor. Amazing isn't it. He also mentions not being able to figure out information flow Options to Securities. Not sure how old this (edit, it's May 2012, very new) is but I've read one or two very interesting papers about how Options information can be used in reverse to price Securities. It's just that the statistical models required to measure the validity are not the same standard ones used in Securities > Options path. Interesting vid anyway. As for R, i'm not sure what you mean "for neural networks". They (or any other system or data mining or whatever) should give you data that you should know how to analyze if you know what you want. There isn't a standard "output" for neural networks or SVMs that would warrant a separate package. If there was, it would probably be very limiting due to the designer's idea of application, so it's probably a good thing there isn't one. R is all about using the broader toolset to work on your own models, not coerce you into models with assumptions that you might not agree with. Still, maybe I'm wrong, i use Matlab more often so i'm no expert on R.
No I think you're probably right. I was wondering if there was something similar to what they have in Mathworks "Neural Network Toolbox" http://www.mathworks.co.uk/products/neural-network/index.html
I do not use NNs. I have used SVMs in the past (this is probably what I was referring to in a rather old post of mine) but this was not within the R environment. I prefer Matlab, Python, C# and F# these days. Mainly Matlab and Python.
You may be looking for something like the e1071 package... Description: Functions for latent class analysis, short time Fourier transform, fuzzy clustering, support vector machines, shortest path computation, bagged clustering, naive Bayes classifier,... Godspeed on your journey... Cordially, -DD-
+1000. Wish I saw your warning last week. I had cobbled together a nice little script for merging and resampling that worked flawlessly for the last year. Upgraded R and packages last week and no longer working (known bug in XTS package). I have found R to be infuriating at times, so have switched over to Python. For anyone who likes the idea of working with data frames, have a look at this 10 minute overview of the pandas package, it is pretty cool: http://wesmckinney.com/blog/?p=647
What were the exact tasks/points where python out-shined R in your experience? Thanks for the link. Very cool. Pandas seem good both on speed and functionality. It blows R away at least on speed alone.
Funny, I love python as an infrastructure/glue language but I did not like pandas at all. It just felt like an attempt to fit a square peg (vectorized logic) into a round hole (a regular OO language that Python is). My main gripe with R (ability to seamlessly interact with larger-then-memory data sets) is not resolved there either, so I saw no reason to switch.
I'm only just getting into it, but it just seems more intuitive/logical for me. I remember with R stuffing around for hours trying to load a CSV file dealing with date/time correctly, whereas in Python it worked seamlessly. What I do miss about Rstudio is the pop viewer for data frames so you can quickly check if your calculations look right. Don't know if there is an equivalent feature in a Python IDE.
Just in case you or anyone comes back to R, there is a new fread in data.table 1.8.7 on rforge (not yet in the default distribution but you can get it from rforge) that figures almost everything by itself and is lightening fast. Date format transformations can be done too in the other readers once you figure out how - I do a lot of them.