CPD: Examples of spurious fitted activity on c11m1d15

kimtonyhyun commented 7 years ago

In #232, I described a particular class of deviation between raw traces and the CPD-fitted traces, where the latter predicts activity whereas the actual data is silent. Let me call these cases "spurious fitted activity." Here, I want to give additional examples of this phenomenon.

I manually compared X and Xest using eval_fit (with the "Signed diff" score), and found about ~20 example cells with spurious fitted activity (out of 487 total cells). I am going to highlight a few examples here. Generally, I think the example show:

The true trace shows mostly reliable firing from trial-to-trial, but for whatever reason, is silent on a few trials.
The fitted CPD trace fills in those missing trials.

kimtonyhyun commented 7 years ago

Cell 80 (Rsq=0.4554): c11m1d15_cell80_fit