Mini-Batches

Split examples into batches without dropping the final short batch.

Start with the concrete trace below. It shows the state the algorithm must carry as it runs.

Problem trace

Mini-Batches: Advance one start cursor by batch size and yield the half-open slice; sequence slicing automatically clips the final end.

Input and goalSplit examples into batches without dropping the final short batch.

Validate the batch sizeFor items [0,1,2,3,4,5,6] and batch_size=3, the positive-size guard passes.

start=00011last included=22233445566

batchSize3halfOpenSlice[0:3]guard3 > 0nextStart0

Recognize it

Use it when every item must appear once in consecutive batches and the final batch may be shorter than the requested size.

Keep true

Before each yield, indexes below start have appeared exactly once; items[start:start+batch_size] contributes the next non-overlapping range, clipped only by sequence length.

Reuse it

Use half-open slices and a fixed stride to partition ordered data without overlap; let bounded slicing handle a short tail explicitly.

Pattern: Step through a sequence by batch size.

Simple idea: Slice from each start position to start + batch_size. Python stops the last slice at the sequence end.

from collections.abc import Iterator, Sequence

def batches(items: Sequence, batch_size: int) -> Iterator[Sequence]:
   if batch_size <= 0:
      raise ValueError("batch_size must be positive")
   for start in range(0, len(items), batch_size):
      yield items[start : start + batch_size]

Cost: $O(n)$ total iteration time and $O(batchsize)$ output per step.