sample_windows#
- hybrid_aif.train.sample_windows(buffer, key, batch_size)[source]#
A uniform batch of
seq_len-step windows: leaves(batch_size, seq_len, ...).Start offsets are uniform over valid starts, with replacement;
is_firstmarks every episode boundary in the window, and step 0’s flag is forced true. A stream shorter than one window floors the offset range at one slot, so guard withStreamBuffer.ready().