DIANN 1.9.2 - Single pass and double pass NN give exact same results

[rodvrees]

Hi Vadim and team,

I was doing some benchmarking runs with DIA-NN 1.9.2, using different settings on the same data.
When I ran both the single pass NN and double pass NN machine learning modes (with all other settings kept identical), I noticed that all the results look identical.
Here's the logs and reports pdf for both runs:
report_doublepass.log.txt
report_doublepass.pdf
report.logsinglepass.txt
reportsinglepass.pdf

The docs mention that Double pass NN mode should give better results in most cases at the cost of time, so is it normal/possible that results look identical, or am I doing/understanding something wrong?

Thanks!

[vdemichev]

Hi Robbe,

Please see the warnings printed, this is what causes it. However, double pass is normally not recommended. In fact, it will be removed in future versions (may comeback later in a different form, but for now there are zero use cases with a proven benefit from it).

Best,
Vadim

[rodvrees]

Ah whoops, I missed that warning, thanks!

That said if you hover over the Machine learning selection box, it does say the following:
, which may be confusing if it turns out not to be the case.

[vdemichev]

Thanks for pointing out :) The docs have been updated since then, but not the popup tip, will fix.