- Bugs due to use of int32s instead of int64s on Windows (thanks to Roman Yurchak).
- Goodbooks dataset.
- Raise ValueError if loss becomes NaN or 0.
- Updated to work with PyTorch 0.3.0.
spotlight.layers.BloomEmbedding: bloom embedding layers that reduce the number of parameters required by hashing embedding indices into some fixed smaller dimensionality, following Serrà, Joan, and Alexandros Karatzoglou. “Getting deep recommenders fit: Bloom embeddings for sparse binary input/output networks.”
sequence_mrr_scorenow accepts an option that excludes previously seen items from scoring.
optimizerarguments is now
optimizer_func. It accepts a function that takes a single argument (list of model parameters) and return a PyTorch optimizer (thanks to Ethan Rosenthal).
fitcalls will resume from previous model state when called repeatedly (Ethan Rosenthal).
- Updated to work with PyTorch v0.2.0.
- Factorization predict APIs now work as advertised in the documentation.