MacroBase

Abstract

As data volumes continue to rise, manual inspection is becoming increasingly untenable. In response, we present MacroBase, a data analytics engine that prioritizes end-user attention in high-volume fast data streams. MacroBase enables efficient, accurate, and modular analyses that highlight and aggregate important and unusual behavior, acting as a search engine for fast data. MacroBase is able to deliver order-of-magnitude speedups over alternatives by optimizing the combination of explanation and classification tasks and by leveraging a new reservoir sampler and heavy-hitters sketch specialized for fast data streams. As a result, MacroBase delivers accurate results at speeds of up to 2M events per second per query on a single core. The system has delivered meaningful results in production, including at a telematics company monitoring hundreds of thousands of vehicles.

Debugging Machine Learning Pipelines

“…Sometimes, when Data X-Ray uses the instances generated by MLDebugger, it does better, at least in recall. 2 https://github.com/raonilourenco/MLDebugger This is expected for the case where the root causes are conjunctions of property-comparator-value triples since Data X-Ray was designed to find relevant conjunctions. That is, Data X-Ray produces a conjunction of property-value combinations that lead to bad scenarios.…”

“…For purposes of reproducibility and community use, we will make our code and experiments available. 2…”

“…Recently, the problem of explaining query results and interesting features in data has received substantial attention in the literature [2,10,13,21,24]. Some works have focused on explaining where and how errors occur in the data generation process [24] and which data items are more likely to be causes of relational query outputs [21,25].…”

“…Some works have focused on explaining where and how errors occur in the data generation process [24] and which data items are more likely to be causes of relational query outputs [21,25]. Others have attempted to use data to explain salient features in data (e.g., outliers) by discovering relationships between attribute values [2,10,13]. These approaches have either focused on using data, including provenance, to explain data or considered pipelines consisting of relational algebra operations.…”

An Asymptotic Ensemble Learning Framework for Big Data Analysis

“…Consequently, analyzing a big data set all at once may require more than the available resources in order to meet specific application requirements [8], [9]. Random sampling is a common strategy to alleviate these challenges [10], e.g., in approximate and incremental computing [8], [11]- [15]. However, drawing random samples from big data is itself an expensive operation [16] especially with the shared-nothing architectures in the mainstream distributed computing frameworks for big data analysis.…”