[lilytong0919]

Thank you for opening this PR. Filling out this template will help the maintainers of pynapple to review your contribution.

This PR template has 2 sections. Please fill the first section yourself. Optionally, description of the PR by a LLM can be included in the second section.

Contributor statement

AI Assistance Disclosure

• No AI tools were used in preparing this PR.
• If AI tools were used, I have disclosed which ones. I have fully reviewed, understood and verified their output.

Uses ChatGPT6 for test case generation (and check that the desired behavior is achived by the current modification

Motivation

In issue #671 I mentioned that compute_perievent give unstable number of bins on continuous data given the same window and timeseries with persumability the same bin_size. I suspected that is due to floating point percision issue when accessing the edge cases.

PR description

This PR address the issue by changing the time_idx and windowsize computation to go through checks with tolerance, and snap time_idx on the edges when the variation in time_idx is close enought to defined window. The former change allow intial creation of new_data_array that are consistent in number of bins. The latter prevent the object intializer to crop the data array on either end due to time_idx edge bins falling slightly over the defined new_time_support.

Outstanding questions / particular feedback

I moved the definition of tolerance into compute_perievent and use it as input to both _align_regular and _is_regularly_sampled. I think this ensures the assertion to regular sampling to fail before the program start giving inconsistent behavior on binsize. I am not very certain this does what I intended it to do. But for this I check with a test with random jittering (which is not a test case because this relys on random generation of data which is not so stable). The setting here should be jittering around the boundary cases of regularity check. So we would expect the print statement in the except block to be triggered multiple times, but the final comparison of bin_count only give 1 consistent value.

import pynapple as nap

rng = np.random.default_rng(0)

nominal_dt = 0.05
window = (-4, 4)
event_time = 10.0

bin_count = []

for i in range(100):
    # print("="*80)
    # print(f"Iteration {i+1}:")
    # One tiny perturbation to dt for the whole regularly sampled series
    noise_level = 1.6e-8
    t = np.arange(0, 20, nominal_dt)
    dt_error = rng.uniform(-noise_level, noise_level, size = (len(t),))

    t_error = t + dt_error
    dt = np.median(np.diff(t_error))

    tsd = nap.Tsd(
        t=t_error,
        d=np.sin(t_error),
    )

    events = nap.Ts(t=[event_time])
    try:
        peth = nap.compute_perievent(
            data=tsd,
            events=events,
            window=window,
        )
    except Exception as e:
        print(f"Error in iteration {i+1}: {e}")
        continue
    peth_binsize = peth.t[1] - peth.t[0]
    peth_times = peth.times()
    bin_count.append(len(peth.t))

    # print(f"Peth size = {peth.shape}, peth binsize = {peth_binsize:.10f}")
    # print(f"Peth start = {peth_times[0]}, peth end = {peth_times[-1]}")
print(f"All bin counts equal: {len(set(bin_count)) == 1}, Bin count: {bin_count[0]}")

Checklist

Affirm that you have done the following:

• I have described the changes in this PR, following the template above.
• I have added any necessary tests.
• I have added any necessary documentation. This includes docstrings, updates to existing files found in doc/, or (for large changes) adding new files to the doc/ folder.
  I only added a small nested function _full_bin_count and several if statement. They are properly commented but no docstring is added. Since this PR only meant to make compute_perievent behave more stably, I think no need to change it's documentation.

AI description of the PR (optional)

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.