[remislp]

What this adds

ekdist.adjacent measures, from a record, what scalcs.sccurves and scalcs.scalcslib give in theory, so one can be superimposed on the other. It is what Figs 6D and 6E of Colquhoun, Hatton & Hawkes (2003) display, and what a fit to a single-channel record is checked against once the rate constants are in hand.

pairs	puts each opening beside a neighbouring shut time
conditional_mean_open_time	mean apparent open time for openings whose adjacent shut time falls in each of a list of ranges — the diamonds of Fig. 6D
dependency	the observed dependency of Magleby & Song (1992) on a log-log grid — Fig. 6E
log_edges	grid edges, per decade

Which neighbour, and why it is an argument

Every function takes side='next', 'previous' or 'adjacent', because the theory distinguishes them: scalcs returns the mean open time given the previous gap and given the next gap as two separate curves, and its HJC_dependency is the joint density of an open time and the shut time that follows it.

Passing the wrong one raises nothing. It quietly compares a measurement against the wrong curve, which is why it is a required-by-default parameter with a documented meaning rather than a convention buried in the code.

One bug found, with a regression test

The dependency's marginals are taken over every pair, not over the pairs that fall inside the grid.

Using the row and column sums of the joint histogram renormalises to whatever rectangle the grid happens to cover, and a shut-time grid never spans the distribution — the gaps between activations run to seconds. In a record of 200 000 intervals only 7 316 of 49 236 pairs were inside the grid, and the mistake turned a true dependency of zero into values of two and three.

test_marginals_are_taken_over_every_pair_not_only_the_grid pins it.

Tests

15, on constructions whose answer is known by hand or by symmetry rather than by comparison with another library — independent times must give zero dependency, correlated ones a positive diagonal, and the conditional means are checked against sums done by hand.

Agreement with the Q-matrix theory is checked separately, where a record can be simulated from a mechanism: on 150 000 intervals of the CHH 2003 AChR mechanism the observed dependency agrees over 21 cells with a mean difference of −0.004, against a theory spanning −0.90 to +2.58. That check needs scalcs and so lives outside this repository; nothing here imports it, and the module needs only numpy.

Also: the clean-install job was not doing what it says

Separate commit. The job is titled "Import every module with nothing else installed" and exists because setup.py has to name everything the package imports — scipy was missing from install_requires for a while and only a clean environment could reveal it.

But it imported a hand-written list, and the list had already fallen behind: ekdist.adjacent is not in it. So the job that catches an undeclared dependency would have missed one in any module added since the list was written.

pkgutil.iter_modules now finds whatever is actually installed. It reports 7 where the list had 6.

Provenance

Written on 2 September during the reproduction of CHH 2003, committed to a local master and never pushed. This is that work rebased onto origin/master, plus the CI fix.

🤖 Generated with Claude Code