Cornac 1.0.0

New models

• Neural Matrix Factorization (NeuMF) / Neural Collaborative Filtering (NCF) (#215)
• Generalized Matrix Factorization (GMF) (#216)
• Multi-Layer Perceptron (MLP) (#217)

New features and improvements

• Update VAECF model (#213)
• Add num_zeros argument into Dataset.uir_iter() to support negative sampling (#214)
• Add val_set into fit() function for model selection (#219)
• Support early stopping in Recommender (#220)
• Fix bug in read_text() function of reader (#221)
• Unify TrainSet and TestSet into single object Dataset (#222)
• Optimize evaluate() function in BaseMethod (#223)
• Dataset checks duplicate observations independently (#224)

Cornac 0.3.5

New models

• Baseline Only (#208)
• Singular Value Decomposition (SVD) (#207)

New features and improvements

• Remove shuffle option in RatioSplit (#209)
• Fix bug of binarizing rating values in Reader (#211)
• Add UITup data format to Reader (#212)

Cornac 0.3.4

New models

• Non-negative Matrix Factorization (NMF) (#204)

New features and improvements

• Tutorial on cross-modality from Text to Graph (#199)
• Tutorial on cross-modality from Text to Image (#195)
• Input data to Modality left unchanged (#205)

Cornac 0.3.3

New features and improvements

• Add random seed to BaseMethod used for all evaluation methods (#198)
• Add random seed to MatrixTrainSet for reproducing data sampling (#202)
• Update BPR, SBPR, and MF models to use single-thread when the random seed is specified (#200, #201)

Cornac 0.3.2

New models

• Hidden Factors and Hidden Topics (HFT) (#196)

New features and improvements

• Amazon Clothing dataset (#197)
• CountVectorizer updated with vocab conditions (#193)

Cornac 0.3.1

New features and improvements

• Rename Module to Modality
• Add a tutorial for adding a new model into Cornac
• Reorder the list of models by year then alphabetically
• Update description, tagline
• Improve documentation

Cornac 0.3.0

New models

• Variational Autoencoder for Collaborative Filtering (VAECF)
• Collaborative Topic Modelling (CTR)

New features and improvements

• Update CVAE to use mini-batch gradient descent
• Remove stopwords from TextModule and CountVectorizer, only input stopwords directly to BaseTokenizer
• GraphModule can build KNN from input feature matrix
• Fix bug of partitioning in CrossValidation
• Swap both ids according to data in FeatureModule and TextModule

Cornac 0.2.1

New models

• Collaborative Variational Autoencoder (CVAE)

New features and improvements

• Update from_splits() function of BaseMethod to support multimodal data modules
• Data modules build() functions return self

Beta Release

New models

• Convolutional Matrix Factorization (ConvMF)
• Collaborative Deep Ranking (CDR)
• Visual Matrix Factorization (VMF)
• Matrix Co-Factorization (MCF)
• Social Bayesian Personalized Ranking (SBPR)
• Social Recommendation (SoRec)

New built-in datasets

• Netflix
• Tradesy
• Amazon Office
• CiteULike
• Epinions

New features and improvements

• Support multimodal recommenders with new data module including TextModule, ImageModule, and GraphModule in data.
• Reader support data filtering based on set of users/items and user/item threshold.
• Models can access to user_text, user_image, user_graph, item_text, item_image, and item_graph through MultimodalTrainSet which is input of the fit() function.

Credits

Thanks to our 4 contributors (alphabetical) whose commits are featured in this release:

@binhtrantt @jyguo @saghiles @tqtg

Alpha Release

New models

• Bayesian Personalized Ranking (BPR)
• Collaborative Context Poisson Factorization (C2PF)
• Collaborative Deep Learning (CDL)
• Collaborative Ordinal Embedding (COE)
• Hierarchical Poisson Factorization (HPF)
• Indexable Bayesian Personalized Ranking (IBPR)
• Online Indexable Bayesian Personalized Ranking (Online IBPR)
• Probabilistic Collaborative Representation Learning (PCRL)
• Probabilistic Matrix Factorization (PMF)
• Spherical K-means (SKM)
• Visual Bayesian Personalized Ranking (VBPR)

Credits

Thanks to our 5 contributors (alphabetical) whose commits are featured in this release:

@Andrew-DungLe @jyguo @llkute @saghiles @tqtg