AutoML makes Machine Learning accessible to all

Introduction

Automated Machine Learning(AutoML) is a process of automating Algorithm selection and Hyperparameter tuning, iterative modelling and model assessment steps of the Machine learning pipeline. The Field of AutoML’s Goal is to allow people with little to no knowledge of machine learning to use various machine learning algorithms and get results. Though the majority of the time spent by a Data Scientist is on data-preprocessing, feature engineering,feature extraction and feature selection. Selecting the Best algorithm for the application and tuning hyperparameters are also iterative processes, which can also consume a significant amount of time to select a good performing Model. There are tools available to automate the Data Preparation tasks and many probably being created. With the data preparation tasks automated, the AutoML will truly become an end to end Automated ML solution.

Tools of AutoML

Auto Sklearn:

Auto-sklearn frees a machine learning user from algorithm selection and hyper-parameter tuning. It takes advantage of Bayesian optimization, meta-learning and ensemble construction. Implementing auto-sklearn is as easy as implementing any other algorithm. See code below.

Tree Based pipeline optimization tools(TPOT):

The TPOT aims to automate the ML pipelines by using genetic programming algorithms. TPOT uses scikit-learn library to choose the best algorithm. TPOT is open source software and programmed in python. Implementing TPOT is also straightforward to implement similar to the implementation of sklearn library. See code below.

ML Box:

This tool can be considered as an end to end ML solution tool. This tool offers pre-processing, cleaning and formatting the data with the option to choose specific features. ML Box also offers model selection and hyper parameter search. It can handle classification and regression prediction problems and model interpreting. This library currently is more suitable for Linux operating systems.

H2O AutoML:

H2O AutoML framework is best suited for automating deep learning networks. Requires JRE(Java Run time Environment) to implement the library.

Auto Keras:

This AutoML framework can search for hyperparameters and architecture required for the deep learning networks. This AutoML framework follows scikit-learn API, and uses a neural network to search for model parameters.

Google Cloud AutoML:

Google launched a GUI based AutoML framework which integrates the power of neural network architecture. As it is GUI based it is easy to use and people with limited knowledge of Machine learning can use to train the models. This however is a paid platform.

Conclusion

With so many AutoML frameworks being released as time progresses, the run of the mill ML tasks can be easily automated. This will enable only specialized ML tasks to humans.