HOGAR

Introduction

In recent years, practice has shown that more and more training data and larger models tend to produce better precision in a variety of applications. In this era of big data, distributed machine learning can provide more computing power, so that it becomes more important than ever.

Work

HOGAR consists of 2 parts, the deployment and management. First, the deployment framework is a library, which is written in python and supports defining and parsing graph information into configuration file to be feed in EARG system. On the other hand, management framework is based on and a web-based monitoring interface for visualizing pipelining machine learning tasks.

Development of UDGL

Based on Autonomous Recovery Graph, this project will further research on User Defined Graph Language and its interrelation with the architecture of distributed machine learning model. For UDGL design, I will do more research on distributed machine learning scheduling and directed acyclic graph analysis. After literature review, new improvement will be suggested to generalization of UDGL. For UDGL implementation, EARG system’s config file will be taken as a model for the output of UDGL.

Development of monitoring web interface

Web interface will be developed in Django, a high-level python framework and link to parameters in EARG systems. Meanwhile, more literature review about evaluating machine learning performance will be done and more metric for monitor will be suggested.

Current Progress

September

1. Meeting with the supervisor.
2. Project plan writing.
3. Project website designing.
4. Literature review.

Results

To be released.

Documentations

• Project Plan

Contact

Guoyang Cui(Team Member)

• Facebook
• Email

Heming Cui(Supervisor)

• Email

Elements

i = 0;

while (!deck.isInOrder()) {
    print 'Iteration ' + i;
    deck.shuffle();
    i++;
}

print 'It took ' + i + ' iterations to sort the deck.';