Hivemall: Hive scalable machine learning library

Hivemall is a scalable machine learning library that runs on Apache Hive. Hivemall is designed to be scalable to the number of training instances as well as the number of training features.

Supported Algorithms

Hivemall provides machine learning functionality as well as feature engineering functions through UDFs/UDAFs/UDTFs of Hive.

Binary Classfication

Perceptron
Passive Aggressive (PA, PA1, PA2)
Confidence Weighted (CW)
Adaptive Regularization of Weight Vectors (AROW)
Soft Confidence Weighted (SCW1, SCW2)
AdaGradRDA (w/ hinge loss)
Factorization Machine (w/ logistic loss)

My recommendation is AROW, SCW1, AdaGradRDA, and Factorization Machine while it depends.

Multi-class Classfication

Perceptron
Passive Aggressive (PA, PA1, PA2)
Confidence Weighted (CW)
Adaptive Regularization of Weight Vectors (AROW)
Soft Confidence Weighted (SCW1, SCW2)
Gradient Tree Boosting (Experimental)

My recommendation is AROW and SCW while it depends.

Regression

Logistic Regression using Stochastic Gradient Descent
AdaGrad, AdaDelta (w/ logistic Loss)
Passive Aggressive Regression (PA1, PA2)
AROW regression
Factorization Machine (w/ squared loss)
Polynomial Regression

My recommendation for is AROW regression, AdaDelta, and Factorization Machine while it depends.

Recommendation

Minhash (LSH with jaccard index)
Matrix Factorization (sgd, adagrad)
Factorization Machine (squared loss for rating prediction)

k-Nearest Neighbor

Minhash (LSH with jaccard index)
b-Bit minhash
Brute-force search using Cosine similarity

Anomaly Detection

Local Outlier Factor (LOF)

Feature engineering

Feature Hashing (MurmurHash, SHA1)
Feature scaling (Min-Max Normalization, Z-Score)
Polynomial Features
Feature instances amplifier that reduces iterations on training
TF-IDF vectorizer
Bias clause
Data generator for one-vs-the-rest classifiers

System requirements

Hive 0.11 or later

Basic Usage

Find more examples on our wiki page and find a brief introduction to Hivemall in this slide.

Copyright

Copyright (C) 2015 Makoto YUI
Copyright (C) 2013-2015 National Institute of Advanced Industrial Science and Technology (AIST)

Put the above copyrights for the services/softwares that use Hivemall.

Support

Support is through the issue list, not by a direct e-mail.

References

Please refer the following paper for research uses:

Makoto Yui. ``Hivemall: Scalable Machine Learning Library for Apache Hive'', 2014 Hadoop Summit, June 2014. [slide]
Makoto Yui and Isao Kojima. ``Hivemall: Hive scalable machine learning library'' (demo), NIPS 2013 Workshop on Machine Learning Open Source Software: Towards Open Workflows, Dec 2013.
Makoto Yui and Isao Kojima. ``A Database-Hadoop Hybrid Approach to Scalable Machine Learning'', Proc. IEEE 2nd International Congress on Big Data, July 2013 [paper] [slide]

Awards

InfoWorld Bossie Awards 2014: The best open source big data tools

Acknowledgment

This work was supported in part by a JSPS grant-in-aid for young scientists (B) #24700111 and a JSPS grant-in-aid for scientific research (A) #24240015.

dongpf/hivemall