- Twtter: @0_u0
- JTMRCのアナリスト(2年目)
- Japanese Traditional Marketing Research Company
- 主な業務内容: エクセルのソルバーでクライアントの気持ちに寄り添う
- 地震コンペに負けました
- TokyoRチームで参加しました
- Thanks kumakichi, tsuyupon, kur0cky
- 悔しくて焼き肉になった
- TokyoRチームで参加しました
- 性格: Negative
2019/6/9
このLTで出てくる事例は所属組織と無関係です。マジで。
summary(Data_Table)
## Age Sex Brand purchase_count ## Min. :20 Min. :0.0 Length:20000 Min. : 0.000 ## 1st Qu.:20 1st Qu.:0.0 Class :character 1st Qu.: 1.000 ## Median :20 Median :1.0 Mode :character Median : 3.000 ## Mean :25 Mean :0.7 Mean : 7.623 ## 3rd Qu.:30 3rd Qu.:1.0 3rd Qu.: 9.000 ## Max. :40 Max. :1.0 Max. :132.000
unique(Data_Table$Brand)
## [1] "Lotte" "Meiji" "Grico" "Morin"
stats::glm(y ~ ., data = YourData, family = 'poisson')
mean(Data_Table$purchase_count[Data_Table$Brand == 'Lotte'])
## [1] 4.545625
var(Data_Table$purchase_count[Data_Table$Brand == 'Lotte'])
## [1] 30.44547
model_pois <- glm(purchase_count ~ Sex + Age, data = Data_Table[Data_Table$Brand == 'Lotte',], family = poisson(link = 'log')) coefficients(model_pois)
## (Intercept) Sex Age ## 0.36594109 2.19362213 -0.02811808
AIC(model_pois)
## [1] 50602.42
MASS::glm.nb
で可能model_nbin <- glm.nb(purchase_count ~ Sex + Age, data = Data_Table[Data_Table$Brand == 'Lotte',], link = log) coefficients(model_nbin)
## (Intercept) Sex Age ## 0.36099311 2.19201781 -0.02787068
model_nbin$theta
## [1] 1.468564
AIC(model_nbin)
## [1] 37974.4
AIC(model_pois)
## [1] 50602.42
AIC(model_nbin)
## [1] 37974.4
rmse(predict(model_pois, newdata = Data_Table), Data_Table$purchase_count)
## [1] 12.74778
rmse(predict(model_nbin, newdata = Data_Table), Data_Table$purchase_count)
## [1] 12.74768
mae(predict(model_pois, newdata = Data_Table), Data_Table$purchase_count)
## [1] 6.728847
mae(predict(model_nbin, newdata = Data_Table), Data_Table$purchase_count)
## [1] 6.728866
mu_Brand = (2.2 * Sex - 0.03 * Age + Lotte) purchase_count = rnegbin(Morin_N, mu = mu_Brand, theta = (1-Lotte_Share)/Lotte_Share)
Overdispersionの気持ち悪さに耐えられるかどうか
lme4::glmer.nb
でRandom-effectモデルを組もう