Machine Learning using Storlets
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Failed to load latest commit information.
mlstorlets
test
.gitignore
LICENSE
README.md
access.cfg
install.sh
setup.cfg
setup.py
test-requirements.txt
tox.ini

README.md

mlstorlets

Machine Learning using Storlets

mlstorlets is an initial approach to leverage Storlets for machine learning.

TL;DR

On a fresh new 16.04 VM with a passwordless sudoer simply:

git clone https://github.com/eranr/mlstorlets.git
cd mlstorlets
./install.sh
tox -e functional

This will install Swift and Storlets on the VM together with a docker container that has the scikit-learn package.

Supported Algorithms

Currently, mlstorlets support Stochastic Grandiant Descent. Specifically it implements SGDRegressorProxy, SGDClassifierProxy which expose python's scikit-learn SGDRegressor, SGDClassifier API [1] together with a remote_fit and remote_score methods. Specifically, the remote_fit method allow to do mini-batch SGD where each batch comes either from local data and/or one or more objects.

For more information see test/functional/test_sgdproxies.py

[1] http://scikit-learn.org/stable/modules/sgd.html