Skip to content

HTTPS clone URL

Subversion checkout URL

You can clone with HTTPS or Subversion.

Download ZIP

classifier-range

Investigate BDT classifier distribution b/w signal & background
Initially, the BDT classifier distribution was fluctuating and
asymmetric b/w signal and background; the background distribution went
up to ~ -1, however the signal distribution went only up to ~ 0.6.
Training was done on 2.7M background and 67k signal events.

Reducing the background events in the training sample to the same order
as signal (2.7M → 55k) made the classifier distribution very smooth and
symmetric b/w signal and background.  My hypothesis is: the number of
events in each sample when training should be of similar order.

This commit reduces the number of background events further (55k → 1.8k)
in order to investigate the hypothesis.  The classifier distribution for
the background after this change extends only up to -0.8, whereas for
signal it is 1.  This is a bit difficult to investigate since now we are
in a regime where there might be too few events in the training sample.
However, I do see the opposite behaviour to what was observed initially.

Maybe my hypothesis is correct.

simfit

Simultaneous Dsπ and DsK acceptance fit works!
1. Because of Python scoping rules in loops, DsK dataset was empty
2. Add dataset print outs

Reminder to self: always print every thing

simfit-w-bug

Rewrite acceptance fit as a simultaneous fit
Doesn't converge.  Do a 2 step fit, then follow up by the simultaneous
fit.  Hopefully this will have the parameter initial values at the
right place helping it converge.

2step

Acceptance ratio plot with error band

segfault

Factorise acceptance function into a module

oanglePID

Finished study of opening angle as kinematic PID variable
The opening angle between the bachelor particle momentum in the lab
frame with the Bs momentum in the rest frame is sensitive to the
boost. The boost in turn is sensitive to the mass hypothesis used for
the bachelor particle. This sensitivity of the variable can be
exploited as a kinematic PID variable. It should be noted this is
unsuitable for a cut based selection as it is strongly correlated with
the invariant mass of Bs meson. However, this should help improve the
mass fit by better constraining the DsK and the Dsπ distributions.

Acknowledgements: Marcel, Manuel and Rose
Something went wrong with that request. Please try again.