Expressing complex data aggregations with Histogrammar
summary
Claims that histograms become composable aggregator monoids and that those monoids are associative and weight-homogeneous for safe parallelism. Shows concrete tooling: a Histogrammar library with JSON interchange, language ports (Python, Scala), and Spark backends for cluster-wide fills; demos include tapping intermediate values on GPUs and moving results between systems. Gives scale context — historical 100 GB per year to 25 petabytes per year today — and shows the implementation actually runs and combines in distributed environments.
AI-written summary · about
chapters (4)
related talks
38mAdd ALL the things: abstract algebra meets analytics
22mA New Kid in Histogram Town
34mVisualising Multi Dimensional Data
42mUX Techniques Supporting Varying Levels of Aggregation
53mTips and tricks for robust big data applications
57mHow to Display 100,500 Metrics of a Distributed System and Not Go Insane
30mSeeing the needle AND the haystack: single-datapoint selection for billion-poin...
34mMeaningful histogramming with Physt