cpmpercussion / impsy

84%

push

github

Fix object-dtype crash when training on a single-performance dataset

np.array([one_2d_array], dtype=object) collapses into a 2D object
array where every cell is a Python float, so slices from it inherit
dtype=object and Keras rejects them with "Invalid dtype: object".
Explicitly passing dtype=np.float32 to the np.array() calls in
PredictiveMusicMDRNN.train() forces the correct type regardless of
how the corpus was stored.

The existing test suite only exercises multi-performance datasets
(heterogeneous array lengths), where numpy stores each performance
as an intact float32 array inside the object array, hiding the bug.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

2 of 2 new or added lines in 1 file covered. (100.0%)

1587 of 1889 relevant lines covered (84.01%)

2.52 hits per line