[TheNetherWatcher]

rr2ann has two problems:

import numpy as np, wfdb
from wfdb.processing.hr import rr2ann

rr2ann(np.array([0.29, 0.57, 1.13]), "t", "atr", fs=100, as_time=True)
wfdb.rdann("t", "atr").sample
# array([ 28,  56, 112])   (expected [29, 57, 113])

rr2ann(np.array([[4, 17], ["V", "N"]]).T, "t", "atr")
# TypeError: the resolved dtypes are not compatible with add.accumulate. Resolved
# (dtype('<U21'), dtype('<U21'), dtype('<U42'))

1. With as_time=True, times are converted to samples with .astype(np.int64), which truncates. 0.29 * 100 is 28.999999999999996, so the annotation lands one sample early. This now rounds to the nearest sample.
2. The docstring says a second column can give the annotation symbols. That makes the array hold strings, and without as_time the intervals go straight into np.cumsum, which fails on strings. They are now converted to integers first (the as_time path already converted to float).

Tests: times that truncate wrongly, intervals with a symbol column (both fail on main), and the docstring example with times and symbols. pytest -m "not network" passes (99 tests), and the docs build passes.

[tompollard]

Thanks, this looks good! Rounding the times fixes the truncation, and the symbol column now works.

One small thing: converting to integers before cumsum changes the result for non-integer intervals, because each interval is truncated and the error builds up:

rr2ann(np.array([4.5, 4.5, 4.5]), "t", "atr")
# main: [4, 9, 13]   this PR: [4, 8, 12]

Intervals like "4.5" with a symbol column also raise ValueError: invalid literal for int(). Summing as floats and then rounding, as the as_time branch does, handles both:

ann_sample = np.round(np.cumsum(ann_sample.astype(np.float64))).astype(np.int64)

That gives [4, 9, 14] for the example above, and your three tests still pass. ann2rr returns integers, so the usual case isn't affected either way.

[TheNetherWatcher]

Thanks, good point! I've switched to summing as floats and rounding once, as you suggested. That also fixes "4.5" with a symbol column. I added a test for [4.5, 4.5, 4.5] with and without symbols ([4, 9, 14]), which fails with the integer-first version, and mentioned it in the changelog. I also rebased onto main. Tests pass (103).

[tompollard]

lgtm