Building Enterprise grade ML Apps : Tools and Architectures
Submitted by Soma Dhavala (@soma) on Saturday, 13 April 2019
Session type: Full talk of 40 mins
ML Products are unfinished by design. ML Centric quality attributes such as MSE and F1-score etcc are necessary but not sufficient. How do we address this fundamentally unsettling characteristic? And the existing Data Science practices are not scalable beyond the confines. In the first part of the talk, an axiomatic framework is provided to address these issues.
In the secon part of the talk, few architecture patterns, along with their corresponding complementary open source implementations will be discussed, each focusing on a specific aspect.
1) DAGGIT: We will look at a refernce architecture of an ML-As-A-Serice platofrm, wherein much of responsibilty is delegated to the tools, and the developer is only accountable. For example, mundane things such as versioning code, data and runtime are delegated to the tool. Every App is provided as a configuration so that running multiple experiments is akin to changing few options in the configuraiton file. In addition, every App comes with an API. A developer can trigger it on-demand or it can be triggered via an API with a call back – so going from development to deployment or running multuple experiments and monitering them is a breeze.
2) GraphRover: It is not uncommon to see both business use cases and the core telemery structures changing rapidly. It is a double whammy for ML Apps – they have to deal with changing data and evolving use cases. In a reference architecture, we treat every intent as a query, powered by graph engines.
3) IMLY: Data Scientist writing in code in Python and reading csv files and Production Engineer re-writing it in Scala and reading data from Cassandra is not uncommon either. How do we deal with two language problem. We need interporability at two levels: at Data and at Models. We will look at IMLY, an source project aims at making models written in the Pythone ecosytem to many run time environments, including a browser.
4) PAPA: On the data side, we will look at isomorphism between Graph and Relational Data and demonstrate the interporability between them via common interfaces and look at Spark both as an OLAP engine as well an Orchestrator
- Anotomy of an ML App and its life cycle
- Scaling Data Science: issues and challanges
- An axiomatic framework
- daggit: ML-As-A-Serice reference platform. Delegate repsonsibility to tools
- GraphRover: Every ML intent is a query. Outsource coding to graph and db engines
- imly: exploit the power of deep learning to achieve ML model interoparability
- papa: make database hopping a reality. abtract out the database details – only focus on the intent.
Soma S Dhavala is freelance consultant operating at the interface of Statistics, Machine Learning, Computing, and Internet of Things. His interests are in Graphs, Meta Machine Learning and their application in representing, and reasoning with information. Since last few years, he has been working on ML-as-Infrastructure.
Soma currently consults for Framewirk, is a member of the Design Council at EkStep. Since 2014 and is leading the efforts in designing Machine Learning Infrastructure and developing various DataProducts concerned with Learning such as Recommendation Engines, Auto-Tagging Content. He designed and ran a Deep Learning and Natural Language Processing immersive bootcamp for NIIT. He is also working on Deep Variational Inference with his collaborators in the academia. He is also a co-founder of VitalTicks Pvt Ltd, a start-up in the digital health care space.
In the past, Soma worked with Dow AgroSciences in the area of ML applications in Systems Biology and with General Electric Global Research Center in the area of Information Theory and Parallel Computing.
Soma obtained his Ph.D (TAMU, 2010) in Statistics where he worked on applying Bayesian Nonparametrics to systems biology (2010), has a Masters (IIT-M, 2000) in EE and a B.E (SRKREC-Andhra University, 1997) in ECE. He also worked as a post-doc close to a year in the area of Dynamical Systems at TAMU. He has over 10+ publications and multiple patents.