Bug in the perf_csi function

[vancities]

highlighted code won't assign the value to the y column. It will raise key_error of y later.

However, with y being NaN,
distr_prob = dat.groupby(['ae', 'bin'])\ ['y'].agg([good, bad])\ .assign(N=lambda x: x.good+x.bad, badprob=lambda x: x.bad/(x.good+x.bad) ).reset_index()
won't have distr attribute at the step of ploting.

Normally, PSI won't require label data and plot should still be okay to have without the line-plot of bad probability.