PyCon Pune 2017

Machine learning is one of the hottest topics in the market right now with every company wanting a piece of it. Machine learning basically is letting a machine learn about how to do a task without explicitly programming it for that task. The applications this idea has are humungous and are if said in simple terms, basically taking over the world. There are now tons of people working in this field along with large number of openings for people with ML skills. The amount of research grants in ML research has also shot up and we can see a large number startups sprouting up offering a service based on ML. All in all this is the most exciting time to learn about this interesting field.
Machine learning for the people who are working in the field is essentially a three step process: data preprocessing, selecting a model and training it and finally making predictions using that model. One of the most tedious tasks in this pipeline is the one of model selection and the other is of tuning the hyper-parameters of that model to obtain the best fit. AutoML is a challenge oraganized with a goal of designing a perfect machine learning black-box capable of performing both model selection and hyper-parameter tuning without any human intervention.

Outline

This talk will be about how we went on to build a full stack machine learning pipeline for our project with Python. So essentially given a random dataset, our autoML black-box will figure out what to do with it and also give out the best model with optimum parameters for that model. The talk will consist of various methods to perform automatic model selection given a dataset. We will explore multiple packages that let us do automatic hyper-parameter tuning post selecting a model. We will conclude with results we obtained with this setup and other alternatives available for this task at various levels of abstraction.

Come build your own autoML pipeline with us at PyCon Pune 2017!

This talk is based upon the project “An automatic Machine Learning Pipeline” done in collaboration with Nagma Khan, Ebin Chacko and Jagadisha K.

Requirements

sklearn, Pandas and Numpy

Speaker bio

I am a Final year Masters student at the Electrical Engineering department, IIT Bombay. I am a Python Enthusiast and have worked on various Machine Learning and Computer Vision projects with Python and have developed multiple workload automation tools with it. I also presented a workshop on “Solving Allocation Problems with Pandas” at SciPy India 2015.