mlhub 1.1.2

Machine learning model repository manager

Under Development

Introduction

The machine learning hub is an open source project and resource for sharing pre-built machine learning models. The models are accessed and managed using the ml command from the mlhub package.

The models collected in the ML Hub archive are listed as Packages on MLHub.ai and the models themselves can be browsed from the main pool.

Quick Start

The command line interface can be installed using PyPi:

$ pip install mlhub

Once installed you will be able to run the sample rain-tomorrow model assuming that you have the free and open source R statistical software package installed. The TL;DR version is below. Note that you type the command ml ... and that everything from the # to the end of the line is ignored (it’s a comment):

$ ml install   rain-tomorrow # Install the model named rain-tomorrow.
$ ml demo      rain-tomorrow # Run the demonstration of the model
$ ml display   rain-tomorrow # Graphical display of pre-built model.

The following commands are available and below is a brief description of each command:

$ ml                         # Show a usage message.
$ ml available               # List of pre-buld models on the MLHub.
$ ml installed               # List of pre-built models installed locally
$ ml install   rain-tomorrow # Install the model named rain-tomorrow.
$ ml readme    rain-tomorrow # View background information about the model.
$ ml commands  rain-tomorrow # List of commands supported by the model.
$ ml configure rain-tomorrow # Install any required packages.
$ ml demo      rain-tomorrow # Run the demonstration of the model
$ ml print     rain-tomorrow # Textual summary of the model.
$ ml display   rain-tomorrow # Graphical display of pre-built model.
$ ml score     rain-tomorrow # Run model on your own data.

Different model packages will have different dependencies and these will be installed by the configure command.

Quick Start: Azure DSVM

A particularly attractive and simple way to get started with exploring the mlhub functionality is to fire up a Ubuntu Data Science Virtual Machine (DSVM) on Azure for as little as USD10 per month for quite a small server or USD90 for a reasonable one. You can get free credit (USD200) from Microsoft to trial the DSVM.

Using this virtual machine will save a lot of time compared with setting up your own machine with the required dependencies, which of course you can do if you wish as all the dependencies are open source.

To set up the virtual machine, with an Azure subscription log in to the portal and add a new Data Science Virtual Machine for Linux (Ubuntu). You need to provide a name (for the virtual machine), a user name and a password, and then create a new resource group and give it a name, and finally choose a location. Go with all the defaults for everything else, except choose a size to suit the budget (B1s is cheap though a D2s is a better interactive experience). Note that you are only charged whilst the machine is fired up so USD90 per month is no where near what you will spend if you only fire up the server when you need.

Once the DSVM is set up go to its Overview page and click on DNS name Configure and provide a name by which to refer to the server publicly (e.g., myml.westus2.cloudapp.azure.com).

We now have a server ready to showcase the pre-built Machine Learning models. There are several options to connect to the server but a recommended one is to use X2Go which supports Linux, Windows, and Mac. Install it and point it to your server (e.g., myml.westus2.cloudapp.azure.com) in the setup.

Connect to the DSVM. Close the Firefox window that pops up. Click on the terminal icon down the bottom, and you are ready to go:

$ pip install mlhub
$ ml
$ ml available

etc.

Pre-Built Model Archives

A model is a zip file archived as .mlm files and hosted in a repository. The public repository is https://mlhub.ai. The ml command can install the pre-built model locally, ready to run a demo, to print and display the model, and to score new data using the model. Some models provide ability to retrain the model with user provided data.

Alternative Install

A tar.gz containing the mlhub package and the command line interface is available as mlhub_1.1.2.tar.gz within the distribution folder of the MLHub.

To install from the tar.gz file:

$ wget https://mlhub.ai/dist/mlhub_1.1.2.tar.gz
$ pip install mlhub_1.1.2.tar.gz
$ ml

Contributions

The open source mlhub command line tool (ml) and sample models are being hosted on GitHub and contributions to both the command line tool and contributions of new open source pre-built machine learning models are most welcome. Feel free to submit pull requests.