All Versions
16
Latest Version
Avg Release Cycle
79 days
Latest Release
-
Changelog History
Page 2
Changelog History
Page 2
-
v0.3.6 Changes
- β Adds evaluation functionality with functions for computing P@k and MAP@K and generating a train/test split
- BPR model now verifies negative samples havenβt been actually liked now, leading to more accurate recommendations
- Faster KNN recommendations (up to 10x faster recommend calls)
- π Various fixes for models when fitting on the GPU
- π Fix CUDA install on Windows
- Display progress bars when fitting models using tqdm
- More datasets: added million song dataset, sketchfab, movielens 100k, 1m and 10m
-
v0.3.5 Changes
- π Use HDF5 files for distributing datasets
- β Add rank_items method to recommender
-
v0.3.3 Changes
- π Fix issue with last user having no ratings in BPR model
-
v0.3.2 Changes
- π Support more than 231 training examples in ALS and BPR models
- π Allow 64 bit factors for BPR
-
v0.3.0 Changes
- β Add a Bayesian Personalized Ranking model, with an option for fitting on the GPU
-
v0.2.7 Changes
- β Add Support for ANN libraries likes Faiss, NMSLIB and Annoy for making recommendations