Stratified Sampling Meets Machine Learning

Edo Liberty, Kevin Lang, Konstantin Shmakov
Proceedings of The 33rd International Conference on Machine Learning, PMLR 48:2320-2329, 2016.

Abstract

This paper solves a specialized regression problem to obtain sampling probabilities for records in databases. The goal is to sample a small set of records over which evaluating aggregate queries can be done both efficiently and accurately. We provide a principled and provable solution for this problem; it is parameterless and requires no data insights. Unlike standard regression problems, the loss is inversely proportional to the regressed-to values. Moreover, a cost zero solution always exists and can only be excluded by hard budget constraints. A unique form of regularization is also needed. We provide an efficient and simple regularized Empirical Risk Minimization (ERM) algorithm along with a theoretical generalization result. Our extensive experimental results significantly improve over both uniform sampling and standard stratified sampling which are de-facto the industry standards.

Cite this Paper

BibTeX
@InProceedings{pmlr-v48-liberty16, title = {Stratified Sampling Meets Machine Learning}, author = {Liberty, Edo and Lang, Kevin and Shmakov, Konstantin}, booktitle = {Proceedings of The 33rd International Conference on Machine Learning}, pages = {2320--2329}, year = {2016}, editor = {Balcan, Maria Florina and Weinberger, Kilian Q.}, volume = {48}, series = {Proceedings of Machine Learning Research}, address = {New York, New York, USA}, month = {20--22 Jun}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v48/liberty16.pdf}, url = {https://proceedings.mlr.press/v48/liberty16.html}, abstract = {This paper solves a specialized regression problem to obtain sampling probabilities for records in databases. The goal is to sample a small set of records over which evaluating aggregate queries can be done both efficiently and accurately. We provide a principled and provable solution for this problem; it is parameterless and requires no data insights. Unlike standard regression problems, the loss is inversely proportional to the regressed-to values. Moreover, a cost zero solution always exists and can only be excluded by hard budget constraints. A unique form of regularization is also needed. We provide an efficient and simple regularized Empirical Risk Minimization (ERM) algorithm along with a theoretical generalization result. Our extensive experimental results significantly improve over both uniform sampling and standard stratified sampling which are de-facto the industry standards.} }
Endnote
%0 Conference Paper %T Stratified Sampling Meets Machine Learning %A Edo Liberty %A Kevin Lang %A Konstantin Shmakov %B Proceedings of The 33rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2016 %E Maria Florina Balcan %E Kilian Q. Weinberger %F pmlr-v48-liberty16 %I PMLR %P 2320--2329 %U https://proceedings.mlr.press/v48/liberty16.html %V 48 %X This paper solves a specialized regression problem to obtain sampling probabilities for records in databases. The goal is to sample a small set of records over which evaluating aggregate queries can be done both efficiently and accurately. We provide a principled and provable solution for this problem; it is parameterless and requires no data insights. Unlike standard regression problems, the loss is inversely proportional to the regressed-to values. Moreover, a cost zero solution always exists and can only be excluded by hard budget constraints. A unique form of regularization is also needed. We provide an efficient and simple regularized Empirical Risk Minimization (ERM) algorithm along with a theoretical generalization result. Our extensive experimental results significantly improve over both uniform sampling and standard stratified sampling which are de-facto the industry standards.
RIS
TY - CPAPER TI - Stratified Sampling Meets Machine Learning AU - Edo Liberty AU - Kevin Lang AU - Konstantin Shmakov BT - Proceedings of The 33rd International Conference on Machine Learning DA - 2016/06/11 ED - Maria Florina Balcan ED - Kilian Q. Weinberger ID - pmlr-v48-liberty16 PB - PMLR DP - Proceedings of Machine Learning Research VL - 48 SP - 2320 EP - 2329 L1 - http://proceedings.mlr.press/v48/liberty16.pdf UR - https://proceedings.mlr.press/v48/liberty16.html AB - This paper solves a specialized regression problem to obtain sampling probabilities for records in databases. The goal is to sample a small set of records over which evaluating aggregate queries can be done both efficiently and accurately. We provide a principled and provable solution for this problem; it is parameterless and requires no data insights. Unlike standard regression problems, the loss is inversely proportional to the regressed-to values. Moreover, a cost zero solution always exists and can only be excluded by hard budget constraints. A unique form of regularization is also needed. We provide an efficient and simple regularized Empirical Risk Minimization (ERM) algorithm along with a theoretical generalization result. Our extensive experimental results significantly improve over both uniform sampling and standard stratified sampling which are de-facto the industry standards. ER -
APA
Liberty, E., Lang, K. & Shmakov, K.. (2016). Stratified Sampling Meets Machine Learning. Proceedings of The 33rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 48:2320-2329 Available from https://proceedings.mlr.press/v48/liberty16.html.

Related Material