Clean up timeseries code

[kyleabeauchamp]

Right now, the FFT-based code is experimental and may have issues (cc @smcantab)

1. Fix, benchmark, and test---or delete---the FFT code. I don't want to maintain two copies of code.

[smcantab]

PR #159 has the essential bug fix and an additional change that I though would be good (see first comit of PR). It seems to be stable now. I have used this method and the old detectEquilibration on 3 time series generated with testsystems.correlated_timeseries_example(N=500, tau-5.0), shifted by 1 with respect to one another and concatenated into one time series. I have done it 1000 times and histogrammed the difference in the predicted equilibration time. The histogram is sharply peaked around 0 and with small tails, as one would expect, so it seems to be working fine now.

[jchodera]

This is great --- thanks!

Would you be willing to share the test you suggested? We can include that as a nosetest to make sure these stay in sync. On Jan 14, 2015 8:38 AM, "Stefano Martiniani" [email protected] wrote:

PR #159 https://github.com/choderalab/pymbar/pull/159 has the essential bug fix and an additional change that I though would be good (see first comit of PR). It seems to be stable now. I have used this method and the old detectEquilibration on 3 time series generated with testsystems.correlated_timeseries_example(N=500, tau-5.0), shifted by 1 with respect to one another and concatenated into one time series. I have done it 1000 times and histogrammed the difference in the predicted equilibration time. The histogram is sharply peaked around 0 and with small tails, as one would expect, so it seems to be working fine now.

— Reply to this email directly or view it on GitHub https://github.com/choderalab/pymbar/issues/158#issuecomment-69916675.

[smcantab]

ok I have pushed the changes (with another fix) and added a test that checks that the mode of the distribution of residuals (difference between tau computed with the 2 methods) is 0.