bitermplus: v0.6.3

This release fixes a bug in transform method that occurred when empty documents were passed as inputs.

bitermplus: v0.6.2

This release fixes a bug in document vs topics matrix shape (reported in this issue).

bitermplus: v0.6.1

This is a minor release that fixes buffer types mismatch on creating biterms (critical bug that appeared under Windows).

bitermplus: v0.6.0

This is a major release that fixes critical bugs in arrays initialization. The previous versions of bitermplus are not recommended for use.

Changelog:

• Arrays (n_bz, n_wz) are now properly initialized. This procedure was broken in the previous versions that led to biased results.
• Data normalization (via _normalize hidden method) improved.
• New NumPy random generators are used to initially assign topics to biterms.
• Biterms (biterms_ model attribute) and topics probabilities (theta_ model attribute) are now available.
• Biterms are now serialized as well when model is saved.

bitermplus: v0.5.10

This release improves model pickling and adds seed argument to fit() method of BTM class.

bitermplus: v0.5.9

In this release public extension attributes were converted to properties with comprehensible names and docstrings.

bitermplus: v0.5.8

This release fixed numerous bugs in the code of inference methods, optimizes memory usage, and covers most part of model fitting and inferring code with tests.