Recommendation.jl Versions Save

Recommendation v1.0.0

Diff since v0.5.0

Closed issues:

• Initial features for implicit feedback recommendation (#1)
• Support BPR loss for matrix factorization (#30)
• Support loading libsvm data format (#32)
• Reorganize unit tests with @testsets (#44)
• Add synthetic user-item interaction generator for experiments (#46)
• Add coverage, diversity, and serendipity metrics for recommendation lists (#53)

Merged pull requests:

• Port takuti/SyntheticImplicitFeedback.jl to synthetic.jl (#47) (@takuti)
• Add load_libsvm_file function to parse libsvm data (#48) (@takuti)
• Refactor testing modules with @testset (#49) (@takuti)
• Benchmark recommenders with fit! optimization and refactoring (#50) (@takuti)
• Unify predict and ranking for simplicity (#51) (@takuti)
• Add simple coverage metric between two lists (#54) (@takuti)
• Add aggregated non-accuracy metrics for diversity and novelty (#55) (@takuti)
• Implement intra-list diversity and serendipity metrics (#57) (@takuti)
• Implement BPR Matrix Factorization recommender (#59) (@takuti)
• Prepare for cross validation-based benchmarking (#60) (@takuti)
• Update/optimize evaluation modules with a benchmark script for testing multiple data-recommender-model pairs (#61) (@takuti)
• Optimize recommend() with bulk prediction (#64) (@takuti)
• Update cross validation interfaces per recent updates on evaluate() (#65) (@takuti)

Recommendation v0.5.0

Diff since v0.4.0

Closed issues:

• Flexible categorical data representation with one-hot encoding (#23)
• Introduce fit! method as a synonym of build! (#27)
• Implement load_xxx module to ease leading common datasets (#29)
• Bump lower bound of Julia version to 1.6 (#33)

Merged pull requests:

• Rename build! to fit! to align with the ML standard (#28) (@takuti)
• Implement MovieLens100k data file parser (#31) (@takuti)
• Download full ml-100k.zip with ZipFile.jl (#35) (@takuti)
• Support onehot encoding for categorical features (#36) (@takuti)
• Load user/item categorical attributes from MovieLens 100k data (#37) (@takuti)
• Bump lower bound of Julia version to 1.6 (#38) (@takuti)
• Fix ML100k data to be loadeed even with path=nothing (#40) (@takuti)
• Add Amazon Reviews dataset loader (#41) (@takuti)
• Add ML-latest-small loader with binarize_multi_label (#42) (@takuti)
• Add load_lastfm data loader to get the user-artist listening frequency data (#43) (@takuti)
• Add dataset loader docstrings to /getting_started page (#45) (@takuti)

Recommendation v0.4.0

Diff since v0.3.0

Closed issues:

• may you share why you switched from python to julia? (#9)
• Incorrect API implementation (#10)
• When using MAE, cross_validation function generate NaN (#13)
• MostPopular baseline model is not properly counting the scores from events input (#16)
• Redesign how to represent/handle missing values in user-item matrices (#20)

Merged pull requests:

• Add Factorization Machine (#8) (@takuti)
• Fix evaluate API (#11) (@DhairyaLGandhi)
• Install TagBot as a GitHub Action (#12) (@JuliaTagBot)
• Fix issues #13 and #10 (#14) (@ghost)
• Better count true positive (#15) (@ghost)
• Add GitHub Actions Workflow for CI (#17) (@takuti)
• Enable Documenter via GitHub Actions (#18) (@takuti)
• Fix most popular count (#19) (@takuti)
• Make MF alias of MatrixFactorization fo readability (#21) (@takuti)
• Enable to initialize params with random numbers when training (#22) (@takuti)
• Standardize missing value representation/handling logic (#24) (@takuti)

v0.3.0 (2019-07-21)

Diff since v0.2.0

Closed issues:

• same recommendations to all the users (#6)
• Example in readme not working with 1.0.2 (#5)
• Julia version v1.0.3 (#4)
• Issue with Personalized recommenders (#3)

• Drop Julia 0.6 support
• Support Julia 1.0 per #4
• Drop Compat module dependency