Working Version

Constitutes a fully operational replication of the ICLR paper's models, updated with the new data format as (v1.1). All models from the paper have been benchmarked in comparable parameter spaces (except baseline RNNs, for which a standard configuration was used) and the results strongly agree with those published.

Benchmark ready

Fixes several bugs and improves documentation and usability compared to (v0.1) alpha. Early testing on public data has been completed with all models reported in the paper and results appear to align so far.

First working version with released data

This release contains a fully operational public implementation of the models, training approach, and data (mainly linked externally) of the original paper. This is an alpha release: I am still benchmarking the original models and expect to make small bug fixes and adjustments. The, externally hosted, data format is likely to change slightly too.

Placeholder

v0.0

Create README.md