I have created a model using glmer(). The battle_seq_num is the number of sessions, each player_idhas 100 sessions (observations). My goal is to make a prediction for the test sample whether they are going to buy a membership or not. xp is a continuous quantitative variable.
m2 <- glmer(is_tier_ten ~ xp + battle_seq_num + (battle_seq_num || player_id), family = binomial(link="logit"), control = glmerControl(optimize = "nloptwrap"), data = dfs)
summary(m2) returns:
Generalized linear mixed model fit by maximum likelihood (Laplace Approximation) ['glmerMod']
Family: binomial ( logit )
Formula: is_tier_ten ~ xp + battle_seq_num + (battle_seq_num || player_id)
Data: dfs
Control: glmerControl(optimize = "nloptwrap")
AIC BIC logLik deviance df.resid
10911.5 10969.4 -5450.7 10901.5 799995
Scaled residuals:
Min 1Q Median 3Q Max
-0.01098 0.00000 0.00000 0.00000 0.10277
Random effects:
Groups Name Variance Std.Dev.
player_id (Intercept) 1.476e-05 0.003841
player_id.1 battle_seq_num 1.379e+03 37.130895
Number of obs: 800000, groups: player_id, 8000
Fixed effects:
Estimate Std. Error z value Pr(>|z|)
(Intercept) -3.01428 0.34902 -8.636 <2e-16 ***
xp 0.07115 0.06879 1.034 0.301
battle_seq_num -6.79224 0.40190 -16.900 <2e-16 ***
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
Correlation of Fixed Effects:
(Intr) xp
xp -0.047
battl_sq_nm -0.921 0.025
Prediction:
str(p2 <- predict(m2,bs2, allow.new.levels = TRUE, type = "response"))
Result:
39046 32744 36225 40022 37105 32809 39818 41095 38585 36820 40429
1.000000e+00 1.000000e+00 1.000000e+00 3.631456e-12 9.999998e-01 1.513329e-11 2.048973e-13 2.220446e-16 2.220446e-16 2.220446e-16 1.000000e+00
36854 42161 3625 40777 50639 37231 36263 36170 36179 159788 186746
2.220446e-16 2.220446e-16 2.220446e-16 2.220446e-16 2.220446e-16 2.220446e-16 2.220446e-16 2.220446e-16 2.220446e-16 2.220446e-16 2.220446e-16
While my identifiers for bs2 are:
[1] 7 10 16 27 29 32 37 51 58 60 64 75 77 80 87 90 98 102 103 108 113 121 122 130 136 141 150 166
[29] 174 184 189 193 194 198 204 212 214 230 232 233 235 241 258 267 271 279 280 282 287 293 294 295 296 299 303 313
[57] 315 330 336 346 347 355 358 361 367 387 399 404 405 414 420 424 432 446 454 456 459 469 472 484 488 495 496 506
[85] 524 526 528 530 531 537 541 542 550 551 552 553 561 582 587 588 591 595 605 606 622 624 634 656 664 665 671 674
[113] 676 680 681 691 694 702 705 711 714 717 720 722 729 734 737 744 759 760 765 766 772 777 786 789 793 804 819 823
[141] 825 827 840 849 855 868 869 870 871 874 878 890 898 907 909 916 919 921 924 927 933 938 939 942 944 947 956 958
[169] 961 974 980 984 987 989 990 991 997 1006 1007 1019 1022 1025 1028 1029 1031 1032 1034 1038 1043 1053 1055 1059 1060 1066 1071 1073
[197] 1076 1077 1080 1092 1098 1100 1101 1104 1110 1114 1120 1126 1129 1133 1136 1137 1145 1150 1161 1163 1167 1169 1180 1185 1189 1193 1201 1203
Is there any way of making a prediction with the initials ids? Just a prediction of whether a person will become is_tier_ten generally for each id and not for every repeated obs (battle_seq_num). As a result, I have to make a .csv document for every id from bs2 with individual prediction.
