blind test

Our function can be expressed as:


xact(t)mshare(td) = yref(t)../f/fpeople(t) + yout(t)

where xact is the number of activities, mshare is the market share, yref reference data - cachier receipts, ../f/fpeople people per receipt, yout people outside the shop, t time, td day

reference vs reference

To check the capability of the prediction we take as reference the customer data and as activities the customer data multiplied by a random noise

reference reference vs reference + gaussian noise, a single location might have low correlation

reference reference vs reference + gaussian noise, overall sum neutralizes the gaussian noise

scor1 score gauss noise 50%

scor1 score gauss noise 90% no smoothing, final score on 30 days

scor_ref score gauss noise 30%, final score on june

scor_ref score gauss noise 20%, final score on june

day correlation mapping

We take the first 20 cilacs close to a poi and we calculate activites on daily basis.

Activities are processed with a 20km previous distance filter and we match activity chirality with the poi chirality.

6% of the total cilacs correlates over 0.6 with reference data.

The sum of the activities over all country is

cilac sum no day filtering

We have to filter out bad days

cilac sum bad day filtering

We perform a weekday correction

cilac sum weekday correction

cilac_cor 2d correlation between cilac patterns

country adjustment country adjustment

etl - lowcount - wday - filter - reg 00 - 04 - 22 - 30 - 41

scor new mapping iteration of the new mapping scoring

blind test on real data

play equal_learn test on performance on learn & play on same days

learn_play smoothing correlates neighboring events and improves the score, june in blind test

curve blind curve blind, june

curve blind_single2 curve blind_single2

curve blind_single curve blind_single

learn_play scoring on the different learning steps until blind test

learn play_randomDays learn play_randomDays

correlation over location correlation over locations