Resolves import problems due to use of implicit namespace. Subsumes v.0.7.1.

A long overdue release, that actually subsumes several release steps (hence, it's not a direct successor of version 0.6.4). Below is a non-exhaustive list of the major changes and improvements for this release.

Models

• HybridSVD is finally in the master branch.
• New wrapper for BPR model from implicit library.

Scenarios

• Item cold start support for all SVD-based models.
• Item cold start support for LCE model.
• Item cold start support for LightFM model.
• Computationally-efficient support for negative sampling in evaluation. Uses smart "on-the-fly" sampling and calculation of scores.

Metrics

• MAP and ARHR are now a part of ranking evaluation output.

Tutorials

• Comprehensive tutorial on tuning and evaluation with polara.
• Usecase: comparing LightFM and HybridSVD in cold start.

Features

• It is now possible to provide a dictionary with desired config directly into data model constructor instead of setting each attribute individually after construction.
• Many new convenience functions for data preprocessing/sampling in both dataframe and sparse matrix formats. In particular, there's a new time-aware sampling that ensures no "recommendations from future" are generated due to sampling of last-consumed items.

Improvements

• Better handling of item cold start scenarios.
• Better handling of LCE and LightFM models.
• Better handling of scaled svd for different scenarios.

This release also includes a number of fixes that improve stability and reliability in certain scenarios.

This release introduces a massive update to the framework with new internal design and additional functionality. With this release the long broken support for Python 2 is abandoned and all future releases will be aimed at Python 3 only starting from 3.6 version.

New models and additional functionality

• New Kernelized Probabilistic MF model.
• Built-in support for scaled version of PureSVD (see Reproducing EIGENREC results tutorial for details).
• Simple hybrid model that uses feature-similarity scores aggregation.
• Baseline models for item cold start regime: popularity-based, random, similarity-aggregation model, PureSVD.
• New classes to support item post-filtering.
• Unified handling of side feature-based relations.
• Support for several learning-rate schedules in SGD: adagrad, adam, rmsprop + my own 3 heuristic schedules adanorm, gnprop and gnpropz.

Hyper-parameter tuning

• Generic find_optimal_config function to perform random grid search over user-defined hyper-parameter space.
• New find_optimal_svd_rank routine to quickly and efficiently tune SVD.
• New find_optimal_tucker_ranks routine to quickly and efficiently tune tensor-based models.
• User can now define, which configurations to skip from random grid search.

Evaluation

• New versatile run_cv_experiment routine to automate cross-validation experiments. Supports both the default and the user-defined evaluation protocols.
• More ways to evaluate against the specific set of metrics supported by Polara.

Performance improvements

• Efficient handling of indices in LightFM model (allows to reduce memory load by orders of magnitude comparing to native LightFM implementation).
• Rating prediction with tensor-based model is now more efficient.
• Computation of Tucker core in tensor-based models is now optional.

Other improvements

• Revived Turi Create (ex Graphlab Create) support with its factorization models including Factorization Machines.
• Refactored evaluation code.
• Refactored and improved code for SGD-based matrix factorization. Now supports both naive and probabilistic implementations.
• Improved handling of sparse operations.
• Better handling of side features.
• Improved timing functionality.
• Internal naming is now more consistent.
• Support for Amazon and Epinions datasets
• Allow unpacking the probe part of the Netflix dataset.
• Some other minor improvements and fixes.

Fixes the setup.py file to add LightFM functionality.

The release introduces some performance improvements, extends evaluation metrics and adds new functionality:

New features and convenience functions

• new metric named coverage;
• support for LightFM model;
• new set_config method, which allows to more easily set desired configuration to a model during grid search experiments;

Performance improvements

• tensor-times-matrix products in the Tucker Decomposition can now be computed in parallel;

Other improvements

• more convenient routine to build one-hot encoding matrix from feature data;
• additional tuning parameter for iALS model;
• more flexible handling of input arguments in plotting functions;
• improved MyMediaLite wrapper + removed some inconsistencies.

This is mostly a bugfix release with several improvements and additions, including:

• the functionality to select data based on some feedback threshold value, controlled by the feedback_threshold attribute
• improved handling of custom testing scenarios
• new tutorial with example on how to reproduce the EigenRec paper results
• improved messaging on current status

This release provides a number of new features as well as performance improvements:

New features and convenience functions

• feedback parameter can now be omitted in RecommenderData instances, which simplifies work with purely implicit positive-only data;
• separate routine to unfold tensor along a specified mode;
• new random grid search routine random_grid in polara/evaluation/pipelines
• evaluation now allows for parallel execution on test data chunks; it helps to reduce evaluation time in certain cases;

Performance improvements

• tensor rounding is now a part of tensor model, allowing for efficient rank truncation (similarly to SVD) without the need to recompute the whole model;
• computing recommendation scores in the tensor model is now more efficient in terms of both memory and CPU load;
• better handling of iALS algorithm from implicit library; now in standard scenario instead of relying on inefficient recommend function, the evaluation is performed fully on polara side;

Other improvements

• get_movielens_data now allows to load tags and timestamp data;
• HR and MRR metrics con now be calculated independently of the number of holdout items;
• user defined memory usage limit is now a computed value, allowing for dynamic changes;
• many improvements on code readability and naming consistency;
• several bugfixes and a number of other improvements, mostly related to computational efficiency and general workflow control.

The release contains usability improvements and minor issues fixes. The most notable additions are:

1. Simplified import of the main models directly from the root of the package.
2. New mixins for handling features data in cold-start regime
3. Simple content-based model for item cold-start regime.

There are also a few code improvements and fixes.

Changes

Mostly codestyle corrections and several bug fixes

Interoperability between Python 2 and Python 3 is added.