The result shows that the coupon usage rate is different among different categories. So I think it maybe necessary to add single category columns. (The existing category columns are combined versions)
Single category (Already done)
For i = 1,2,...,31
nCategory1[i] = how many times was this category i seen in column 1
nCategory2[i] = how many times was this category i seen in column 2
nCategory3[i] = how many times was this category i seen in column 3
propCategory1[i] = what proportion of times was this category seen in column 1
propCategory2[i] = what proportion of times was this category seen in column 2
propCategory3[i] = what proportion of times was this category seen in column 3
But it seems that we have to add too many columns... So I didn't export it to the feature matrix yet, expecting for better ideas.
I have been studying the relationship between
categoryandcouponUsed. Here is my findings: https://github.com/imouzon/dmc2015/blob/master/weicheng/categoryExploration.mdThe result shows that the coupon usage rate is different among different categories. So I think it maybe necessary to add single category columns. (The existing category columns are combined versions)
Single category (Already done)
For i = 1,2,...,31 nCategory1[i] = how many times was this category
iseen in column 1 nCategory2[i] = how many times was this categoryiseen in column 2 nCategory3[i] = how many times was this categoryiseen in column 3 propCategory1[i] = what proportion of times was this category seen in column 1 propCategory2[i] = what proportion of times was this category seen in column 2 propCategory3[i] = what proportion of times was this category seen in column 3But it seems that we have to add too many columns... So I didn't export it to the feature matrix yet, expecting for better ideas.