5 ms·

TensorFlow for R

115 points by martgnz 10y ago

baldfat 10y ago
This is a big leap for Machine Learning in R. R keeps moving forward in so many ways. We now have native dplyr into Spark and now a TensorFlow library from RStudio.
- blahi 10y ago
  "Leap" is quite the exaggeration IMO, especially talking about ML. This is improvement wrt to neural networks which had some support in R already (mxnet). In ML, in general, R is quite a bit ahead of the competition. You have been able to use dplyr on spark dataframes for a while now (at least a year).
  - thatcat 10y ago
    >In ML, in general, R is quite a bit ahead of the competition. Could you elaborate on this? I'm just starting out in R.
  - baldfat 10y ago
    > You have been able to use dplyr on spark dataframes for a while now (at least a year) How? I didn't get it to work till sparklyr???? https://blog.rstudio.org/2016/09/27/sparklyr-r-interface-for-apache-spark/ https://blog.rstudio.org/2016/09/27/sparklyr-r-interface-for...
    - blahi 10y ago
      https://github.com/RevolutionAnalytics/dplyr-spark https://github.com/RevolutionAnalytics/dplyr-spark
      - baldfat 10y ago
        > not yet endowed with a through test suite. Nonetheless we expect it to inherit much of its correctness, scalability and robustness from its main dependencies, dplyr and spark. > we don't recommend production use yet I was a bit disappointed by that package, though it is great to see it in the ecco-system of R. The RStudio sparklyr is 100% native and Production ready. You should check it out I was happily surprised.
cs702 10y ago
Great news for R users who largely have been left behind by the deep learning community. The main downside to this is that we will now start seeing all sorts of experimental models from R users who know a lot more about statistics than about software engineering and therefore often write code that is not very elegant nor easy to understand.[1] [1] Don't take it from me. Quoting Hadley Wickham: "Much of the R code you’ll see in the wild is written in haste to solve a pressing problem. As a result, code is not very elegant, fast, or easy to understand. Most users do not revise their code to address these shortcomings. Compared to other programming languages, the R community tends to be more focussed on results instead of processes. Knowledge of software engineering best practices is patchy." http://adv-r.had.co.nz/Introduction.html http://adv-r.had.co.nz/Introduction.html
- RA_Fisher 10y ago
  Surely we're better off if statisticians and software engineers work together. That we're worse off when the two groups come together doesn't seem probable ...
  - cs702 10y ago
    Of course. Overall this is good news.
- azuajef 10y ago
  This is a key point, well said. However, we need more people who are proficient in both "worlds" or at least, as another user points out below, more people willing to collaborate.
- wodenokoto 10y ago
  As long as the model is well described I'm all for more statistician experimenting with deep learning. I don't mind if initial code isn't great from a software perspective. If it is needed in production someone else would have to reimplement it.
  - huac 10y ago
RockyMcNuts 10y ago
Where's TensorFlow for Excel?
- max_ 10y ago
  http://www.deepexcel.net http://www.deepexcel.net
  - Houshalter 10y ago
    That's hilarious, but it doesn't use tensorflow. They seem to actually implement a neural net in spreadsheet cells with formulas for every neuron. That is insanely impractical and will be a million times slower and harder to use than tensorflow.
    - kgwgk 10y ago
      Probably the spreadsheet they used for the demo when this software was presented on April 1st, 2016 is still calculating...
  - RockyMcNuts 10y ago
    looks cool, I was thinking more along the lines of Excel functions to run Keras or sklearn off-the-shelf ML algos in TensorFlow on e.g. an AWS back end, using Excel like TensorBoard to visualize the progress of an algorithm, mapping an Excel computation graph to an TensorFlow computation graph...
dandermotj 10y ago
This is awesome - been waiting for it for ages! But bindings to the python API and not C++?
- sja 10y ago
  I imagine that they did this in order to be able to use more of (or all of) the TensorFlow functionality available in the Python API without having to rewrite a huge amount of code in R. There are Operations in TensorFlow that don't invoke the C++ layer directly. Instead, they are composed of other Operations in Python, meaning that binding directly to the C++ layer would require finding and rewriting that code. It still gets sent down to the C++ layer at the end of it, though the stack traces might end up looking pretty ridiculous.

115 points by martgnz 10y ago

This is a big leap for Machine Learning in R. R keeps moving forward in so many ways. We now have native dplyr into Spark and now a TensorFlow library from RStudio.

blahi 10y ago
"Leap" is quite the exaggeration IMO, especially talking about ML. This is improvement wrt to neural networks which had some support in R already (mxnet). In ML, in general, R is quite a bit ahead of the competition. You have been able to use dplyr on spark dataframes for a while now (at least a year).
- thatcat 10y ago
  >In ML, in general, R is quite a bit ahead of the competition. Could you elaborate on this? I'm just starting out in R.
- baldfat 10y ago
  > You have been able to use dplyr on spark dataframes for a while now (at least a year) How? I didn't get it to work till sparklyr???? https://blog.rstudio.org/2016/09/27/sparklyr-r-interface-for-apache-spark/ https://blog.rstudio.org/2016/09/27/sparklyr-r-interface-for...
  - blahi 10y ago
    https://github.com/RevolutionAnalytics/dplyr-spark https://github.com/RevolutionAnalytics/dplyr-spark
    - baldfat 10y ago
      > not yet endowed with a through test suite. Nonetheless we expect it to inherit much of its correctness, scalability and robustness from its main dependencies, dplyr and spark. > we don't recommend production use yet I was a bit disappointed by that package, though it is great to see it in the ecco-system of R. The RStudio sparklyr is 100% native and Production ready. You should check it out I was happily surprised.

cs702 10y ago

Great news for R users who largely have been left behind by the deep learning community. The main downside to this is that we will now start seeing all sorts of experimental models from R users who know a lot more about statistics than about software engineering and therefore often write code that is not very elegant nor easy to understand.[1] [1] Don't take it from me. Quoting Hadley Wickham: "Much of the R code you’ll see in the wild is written in haste to solve a pressing problem. As a result, code is not very elegant, fast, or easy to understand. Most users do not revise their code to address these shortcomings. Compared to other programming languages, the R community tends to be more focussed on results instead of processes. Knowledge of software engineering best practices is patchy." http://adv-r.had.co.nz/Introduction.html http://adv-r.had.co.nz/Introduction.html

RA_Fisher 10y ago
Surely we're better off if statisticians and software engineers work together. That we're worse off when the two groups come together doesn't seem probable ...
- cs702 10y ago
  Of course. Overall this is good news.
azuajef 10y ago
This is a key point, well said. However, we need more people who are proficient in both "worlds" or at least, as another user points out below, more people willing to collaborate.
wodenokoto 10y ago
As long as the model is well described I'm all for more statistician experimenting with deep learning. I don't mind if initial code isn't great from a software perspective. If it is needed in production someone else would have to reimplement it.
- huac 10y ago

RockyMcNuts 10y ago

Where's TensorFlow for Excel?

max_ 10y ago
http://www.deepexcel.net http://www.deepexcel.net
- Houshalter 10y ago
  That's hilarious, but it doesn't use tensorflow. They seem to actually implement a neural net in spreadsheet cells with formulas for every neuron. That is insanely impractical and will be a million times slower and harder to use than tensorflow.
  - kgwgk 10y ago
    Probably the spreadsheet they used for the demo when this software was presented on April 1st, 2016 is still calculating...
- RockyMcNuts 10y ago
  looks cool, I was thinking more along the lines of Excel functions to run Keras or sklearn off-the-shelf ML algos in TensorFlow on e.g. an AWS back end, using Excel like TensorBoard to visualize the progress of an algorithm, mapping an Excel computation graph to an TensorFlow computation graph...

dandermotj 10y ago

This is awesome - been waiting for it for ages! But bindings to the python API and not C++?

sja 10y ago
I imagine that they did this in order to be able to use more of (or all of) the TensorFlow functionality available in the Python API without having to rewrite a huge amount of code in R. There are Operations in TensorFlow that don't invoke the C++ layer directly. Instead, they are composed of other Operations in Python, meaning that binding directly to the C++ layer would require finding and rewriting that code. It still gets sent down to the C++ layer at the end of it, though the stack traces might end up looking pretty ridiculous.