prcbench 1.1.16

• Cap the numpy thread pool to two threads while the bundled Python module is imported. The pool was sized to the number of cores, which made the CPU time of the import several times its elapsed time

prcbench 1.1.15

• Fix the help page of create_testset, which described the naming convention of benchmark test sets the wrong way round. The prefix ‘b’ is a balanced set and ‘i’ an imbalanced one. The ‘r’ prefix, which uses a random ratio of positives, is now documented too

• Show sklearn in the curve evaluation plot of the readme

• Show recorded running times in the readme, as the mean of each tool on four balanced test sets from 100 to 100,000 data points, instead of timing the tools every time the readme is built. data-raw/run_readme_benchmark.R records them

prcbench 1.1.14

• Add the def7, auc7, and crv7 tool sets that contain all seven tools

• Change the members of the def6, auc6, crv6, def5, auc5, crv5, def4, auc4, and crv4 tool sets. Every predefined set now contains sklearn, and the smaller sets drop PerfMeas, then AUCCalculator, then PRROC

• Return a flat dummy curve from the sklearn tool when reticulate, Python, or numpy is unavailable, in the same way as the AUCCalculator tool does without rJava

• Document that run_benchmark timings of the sklearn tool include the cost of crossing the R/Python boundary and are therefore not comparable with the timings of the tools written in R

prcbench 1.1.13

• Add sklearn as a wrapped tool, calculated by a standalone Python module bundled in inst/python and derived from the scikit-learn source, so scikit-learn itself is not required

• Require reticulate, Python, and numpy to run the sklearn tool, which is requested by name with create_toolset("sklearn") and is not a member of any predefined tool set

prcbench 1.1.12

• Add yardstick as a wrapped tool

• Add the def6, auc6, and crv6 tool sets that include yardstick

• Regenerate the help pages with roxygen2 8.1.0

prcbench 1.1.11

• Improve the readme and the introduction vignette

• Regenerate the help pages with roxygen2 8.0.0

• Build the pkgdown site from the main branch only

prcbench 1.1.10

• Stop using unsigned int as index in Rcpp

prcbench 1.1.8

• Use new R setup configurations of GitHub actions

prcbench 1.1.3

• Use patchwork to combine multiple ggplots

• Use styler and lintr to reformat R source files

• Improve readme and introduction

prcbench 1.1.2

• Update package sticker and favicons

prcbench 1.1.1

• Use const unsigned int for index variable in PerfMeas Rcpp code

prcbench 1.1.0

• Port precision-recall calculation from obsolete PerfMeas package to internal RCpp

• HTML5 documents

prcbench 1.0.1

• Update citation

prcbench 1.0.0

• Transfer the GitHub repository to evalclass/prcbench

prcbench 0.9.1

• Update Travis configurations for rJava

prcbench 0.9

• Improve code quality using the results from lintr and CodeFactor.io

prcbench 0.8.2

• Avoid data.frame() factor conversion in unit tests

prcbench 0.8

• Fix hard-coded JAR file path issue

prcbench 0.7.3

• Update curve evaluation for PRROC version 1.2

prcbench 0.6.2

• Create github pages with pkgdown

prcbench 0.5.2

• Update README

• Update wrapper functions so that prcbench works when PerfMeas is not available

prcbench 0.5

• Change predefined C3 data

• Update AppVeyor config for rJava

prcbench 0.4

• Enhance create_usrtool

  • x and y values can be specified as precalculated precision and recall

• Add a new test set

  • C4

• Add test categories to curve evaluation test result

• Improve graph options

prcbench 0.3

• Improve the testing environment
  • unit tests
  • codecov
• Change Java version
  • 1.7 -> 1.6

prcbench 0.2

• Fix microbenchmark
  • Change from ‘Imports’ to ‘Suggests’
  • Use system time when microbenchmark is unavailable
• Improve several documents
  • help files (.Rd)
  • package vignette
  • README

prcbench 0.1

• The first release version of prcbench

• The package offers four main functions

  • Common tool interface for multiple tools
  • Common test data interface for benchmarking and curve evaluation
  • Benchmarking of tools that generate Precision-Recall curves
  • Evaluation of Precision-Recall curves

• The package contains predefined interfaces of the following five tool

  • ROCR
  • AUCCalculator
  • PerfMeas
  • PRROC
  • precrec

mirror server hosted at Truenetwork, Russian Federation.