0

Which version is better for machine learning?

I have heard that 14.04 works better but I have ran into more that a few snags while using it (WiFi issues). I want to shift to 16.04 but am afraid it might not work as well as 14.04 when it comes to machine learning.

I'll mostly be using Keras library and/or Caffe.

Zanna
  • 70,465
Dhananjay Gahlawat
  • 3
  • 4
    There is no better in regards to machine learning and this highly depends on the quality of your system. I would create a workflow so you can reinstall quickly if needed. Generally I would always go with the latest LTS (but that is not specific to machine learning). – Rinzwind Aug 21 '17 at 14:36
  • I second @Rinzwind's comment.... – You'reAGitForNotUsingGit Aug 21 '17 at 14:37
  • There is no difference. – Pilot6 Aug 21 '17 at 14:43
  • If "Machine learning" is your goal, you can minimize the major system upgrades and ensure support availability by sticking to the Long Term Support (LTS) releases of Ubuntu. 16.04LTS is the most recent LTS release. – waltinator Aug 21 '17 at 15:32

1 Answers1

1

Use whichever release that works best for the hardware.

The release that works well with your hardware would be the most relevant choice, given that machine learning tools are capable of running on CUDA-enabled graphic processing unit (GPU). Both Keras and Caffe seems to have been tested running on some hardware models.

When your hardware support is good, then only look at how to install the packages. Apparently, which release to use does not really matter according to the documentation for Keras and Caffe.

From documentation for Keras:

Keras is compatible with: Python 2.7-3.5.

You can install Keras from PyPI (easiest method):

sudo pip install keras

From documentation for Caffe:

Caffe has several dependencies:

  • CUDA is required for GPU mode.

    • library version 7+ and the latest driver version are recommended, but 6.* is fine too
    • 5.5, and 5.0 are compatible but considered legacy
  • BLAS via ATLAS, MKL, or OpenBLAS.
    • Boost >= 1.55
    • protobuf, glog, gflags, hdf5

The instruction to install Caffe varies by Ubuntu releases.

For Ubuntu (>= 17.04)

CPU-only version:

sudo apt install caffe-cpu

CUDA-version:

sudo apt install caffe-cuda

Installation seems to be easy on newer releases of Ubuntu. On the other hand, packages must be installed separately on releases older than 17.04 (ouch).

Note that I have merely quoted relevant information from the documentation for Keras and Caffe; I have no experience for using these tools.

One thing for sure: Most users wouldn't want to deal with trivial issues i.e. unable to use Wi-Fi connection, screen tearing, etc. even before attempting to install the desired packages. Hence my suggestion to use whichever release that works best for the hardware. That comes first.