HCI 2-4 InCarTech Deep Learning to Predict Headlamp Temperature

Human Computer Interaction Section 2

Team 2-4

21400704 Jung Haejun

21500675 Jeong Pil Won

21500737 Choi Yeabin

21800741 Choi Yun Seo

date: 2020 06 06

---

library(h2o)
library(caret)
library(tidyverse)

setwd("C:/Users/PHILIP JEONG/Documents/R/deep")
load("C:/Users/PHILIP JEONG/Documents/datasets/Fulldata.Rdata")
Sys.setenv(JAVA_HOME = "C:/Program Files/Java/jre1.8.0_251")

load("plane_lens.Rdata")

Fulldata %>% filter(bulb=="p") %>% select(-bulb) -> Fulldata_p
index <- createDataPartition(Fulldata_p$volume,list=F,p=0.8)
train_p <- Fulldata_p[index,]
test_p <- Fulldata_p[-index,]

Deeplearning

deep1 <- h2o.deeplearning(
  model_id="deep1",
  training_frame=train_p_h2o,
  x=predictors,
  y=response,
  hidden=c(256,128,128),                  ## small network, runs faster
  epochs=1000,                      ## hopefully converges earlier...
  stopping_rounds=5,
  stopping_metric="RMSE", ## could be "MSE","logloss","r2"
  stopping_tolerance=0.0001
)
summary(deep1)

prediction1 <- h2o.predict(deep1,test_p_h2o[,-10])
prediction1 <- as.data.frame(prediction1)
actual1 <- as.data.frame(test_p$temp_c)
predactual <- cbind(prediction1,actual1)
sqrt(mean((predactual$predict-predactual$`test_p$temp_c`)^2)) #0.1738127


prediction2 <- h2o.predict(deep1,test_p_virtual_h2o)
testp150150120 <- cbind(as.data.frame(prediction2),testp150150120)
testp150150120 %>% group_by(c,w,volume) %>% summarise(mean=mean(predict))

library(readxl)

각 전구별 데이터셋 만들기 딥러닝 해서 하이퍼파라미터 주면 딥러닝 돌리기 데이터셋 만든거 돌려서 개별은 저장해놓고 평균온도로 주기 w 4개 10 20 30 40 c 5개 50 70 85 90 100

grid search

h2o.init(nthreads = 2,max_mem_size = "8g", enable_assertions = FALSE)
train_p_h2o <- as.h2o(train_p,destination_frame = "train_p_h2o")
test_p_h2o <- as.h2o(test_p,destination_frame = "test_p_h2o")
#test_p_virtual_h2o <- as.h2o(testp150150120,destination_frame = "test_p_virtual_h2o")

response <- "temp_c"
predictors <- setdiff(names(train_p_h2o), response)

hyper_params <- list(
  hidden = list(c(128, 128, 64), c(256, 256, 128)),
  epochs = 500,  
  rate = c(0.05,0.01)
)
  #input_dropout_ratio=c(0,0.05),
  #l1=seq(0,1e-6,length.out = 3),
  #l2=seq(0,1e-6,length.out = 3)

search_criteria <- list(
  strategy = "RandomDiscrete", stopping_rounds = 5,
  stopping_metric = "RMSE", ## could be "MSE","logloss","r2"
  stopping_tolerance = 0.001)


p_grid_model <- h2o.grid(algorithm = "deeplearning", 
                        x = predictors, y = response,
                        grid_id = "p_grid_model",
                        training_frame = train_p_h2o,
                        nfolds = 3,
                        hyper_params = hyper_params,
                        search_criteria = search_criteria)

p_grid <- h2o.getGrid(grid_id = "p_grid_model",
                             sort_by = "rmse",
                             decreasing = TRUE)
p_grid

p_grid_model.h2o <- h2o.getModel(p_grid_model@model_ids[[1]])

setwd("C:/Users/PHILIP JEONG/Documents/R/deep")
hidden_rate_path <- h2o.saveModel(object = hidden_rate_model@model_ids[[1]], 
                                  path = getwd(), force = TRUE)

deeplearning_grid_1 <- h2o.loadModel(hidden_rate_path)

Grab the top deeplearning model

######

#best for hidden: c(256,128) and c(128,64) from list(c(256,128), c(128,64), c(256,256,128), c(128,128,64))

# 모델 rmse / test셋 rmse
# 1: 0.2667804 / 0.2673417
# 2: 0.2169014 / 






#for(i in 1:10){
#  j <- cbind(test_p$temp_c,as.data.frame(get(paste0("pred",i))))
#  print(
#    sqrt(mean((j$`test_p$temp_c`-j$predict)^2))
#  )
#}

Shutdown

h2o::h2o.shutdown(prompt = FALSE)

fulldata (hidden rate, rate)

Hyper-Parameter Search Summary: ordered by decreasing rmse epochs hidden rate model_ids 1 55.337133806655295 [128, 128, 64] 0.01 p_grid_model_model_2 2 39.74071333035025 [256, 256, 128] 0.01 p_grid_model_model_3 3 54.41185987854036 [128, 64] 0.01 p_grid_model_model_1 4 35.43564494774071 [128, 64] 0.05 p_grid_model_model_5 5 36.11255034716078 [128, 64] 0.01 p_grid_model_model_6 6 33.920045818944836 [256, 128] 0.01 p_grid_model_model_8 7 37.67070705997133 [256, 128] 0.01 p_grid_model_model_4 8 40.661579306478814 [256, 128] 0.05 p_grid_model_model_7 rmse 1 0.2708127656077341 2 0.2586228534850487 3 0.24570491827919483 4 0.24377382888693253 5 0.2027425537057507 6 0.19774907468459532 7 0.15651192077650344 8 0.15074363912858893

Merged by bulb, lens, up

h2o.init(nthreads = 2,max_mem_size = "8g")

unseen_bulb <- unseen_bulb[-c(14:15)]
unseen_lens <- unseen_lens[-length(unseen_lens)]
unseen_up <- unseen_up[-length(unseen_up)]
unseen_bulb <- unseen_bulb[-length(unseen_bulb)]

unseen_bulb_h2o <- as.h2o(unseen_bulb)
unseen_lens_h2o <- as.h2o(unseen_lens)
unseen_up_h2o <- as.h2o(unseen_up)

unseen_bulb <- cbind(unseen_bulb,as.data.frame(h2o.predict(h2o.getModel(p_grid_model@model_ids[[1]]),unseen_bulb_h2o)))
rmse(unseen_bulb$temp_c,unseen_bulb$predict)

unseen_lens <- cbind(unseen_lens,as.data.frame(h2o.predict(h2o.getModel(plane_h_lens_model@model_ids[[1]]),unseen_lens_h2o[-c(7,9,11)])))
rmse(unseen_lens$temp_c,unseen_lens$predict)

unseen_up <- cbind(unseen_up,as.data.frame(h2o.predict(h2o.getModel(plane_h_up_model@model_ids[[1]]),unseen_up_h2o[-c(7,9,11)])))
rmse(unseen_up$temp_c,unseen_up$predict)

bulb

unseen_bulb %>% filter(w==15 & c==50 & box_x==65 & box_y==65 & box_z==64) -> bulb1
unseen_bulb %>% filter(w==20 & c==60 & box_x==65 & box_y==65 & box_z==64) -> bulb2
unseen_bulb %>% filter(w==35 & c==85 & box_x==80 & box_y==80 & box_z==94) -> bulb3
unseen_bulb %>% filter(w==40 & c==95 & box_x==80 & box_y==80 & box_z==94) -> bulb4
unseen_bulb %>% filter(w==15 & c==60 & box_x==60 & box_y==60 & box_z==64) -> bulb5
unseen_bulb %>% filter(w==20 & c==50 & box_x==60 & box_y==60 & box_z==64) -> bulb6
unseen_bulb %>% filter(w==20 & c==60 & box_x==60 & box_y==60 & box_z==64) -> bulb7
unseen_bulb %>% filter(w==15 & c==60 & box_x==60 & box_y==60 & box_z==84) -> bulb8
unseen_bulb %>% filter(w==20 & c==50 & box_x==65 & box_y==65 & box_z==84) -> bulb9
unseen_bulb %>% filter(w==35 & c==95 & box_x==90 & box_y==90 & box_z==94) -> bulb10
unseen_bulb %>% filter(w==40 & c==85 & box_x==90 & box_y==90 & box_z==104) -> bulb11
unseen_bulb %>% filter(w==35 & c==95 & box_x==100 & box_y==100 & box_z==104) -> bulb12
unseen_bulb %>% filter(w==40 & c==85 & box_x==100 & box_y==100 & box_z==104) -> bulb13
unseen_bulb %>% filter(w==40 & c==95 & box_x==100 & box_y==100 & box_z==104) -> bulb14


bulb1_h2o <- as.h2o(bulb1)
bulb2_h2o <- as.h2o(bulb2)
bulb3_h2o <- as.h2o(bulb3)
bulb4_h2o <- as.h2o(bulb4)
bulb5_h2o <- as.h2o(bulb5)
bulb6_h2o <- as.h2o(bulb6)
bulb7_h2o <- as.h2o(bulb7)
bulb8_h2o <- as.h2o(bulb8)
bulb9_h2o <- as.h2o(bulb9)
bulb10_h2o <- as.h2o(bulb10)
bulb11_h2o <- as.h2o(bulb11)
bulb12_h2o <- as.h2o(bulb12)
bulb13_h2o <- as.h2o(bulb13)
bulb14_h2o <- as.h2o(bulb14)

bulb1 <- cbind(bulb1,as.data.frame(h2o.predict(h2o.getModel(p_grid_model_1@model_id[[1]]),bulb1_h2o[-c(7,9,11)])))
bulb2 <- cbind(bulb2,as.data.frame(h2o.predict(h2o.getModel(p_grid_model_1@model_id[[1]]),bulb2_h2o[-c(7,9,11)])))
bulb3 <- cbind(bulb3,as.data.frame(h2o.predict(h2o.getModel(p_grid_model_1@model_id[[1]]),bulb3_h2o[-c(7,9,11)])))
bulb4 <- cbind(bulb4,as.data.frame(h2o.predict(h2o.getModel(p_grid_model_1@model_id[[1]]),bulb4_h2o[-c(7,9,11)])))
bulb5 <- cbind(bulb5,as.data.frame(h2o.predict(h2o.getModel(p_grid_model_1@model_id[[1]]),bulb5_h2o[-c(7,9,11)])))
bulb6 <- cbind(bulb6,as.data.frame(h2o.predict(h2o.getModel(p_grid_model_1@model_id[[1]]),bulb6_h2o[-c(7,9,11)])))
bulb7 <- cbind(bulb7,as.data.frame(h2o.predict(h2o.getModel(p_grid_model_1@model_id[[1]]),bulb7_h2o[-c(7,9,11)])))
bulb8 <- cbind(bulb8,as.data.frame(h2o.predict(h2o.getModel(p_grid_model_1@model_id[[1]]),bulb8_h2o[-c(7,9,11)])))
bulb9 <- cbind(bulb9,as.data.frame(h2o.predict(h2o.getModel(p_grid_model_1@model_id[[1]]),bulb9_h2o[-c(7,9,11)])))
bulb10 <- cbind(bulb10,as.data.frame(h2o.predict(h2o.getModel(p_grid_model_1@model_id[[1]]),bulb10_h2o[-c(7,9,11)])))
bulb11 <- cbind(bulb11,as.data.frame(h2o.predict(h2o.getModel(p_grid_model_1@model_id[[1]]),bulb11_h2o[-c(7,9,11)])))
bulb12 <- cbind(bulb12,as.data.frame(h2o.predict(h2o.getModel(p_grid_model_1@model_id[[1]]),bulb12_h2o[-c(7,9,11)])))
bulb13 <- cbind(bulb13,as.data.frame(h2o.predict(h2o.getModel(p_grid_model_1@model_id[[1]]),bulb13_h2o[-c(7,9,11)])))
bulb14 <- cbind(bulb14,as.data.frame(h2o.predict(h2o.getModel(p_grid_model_1@model_id[[1]]),bulb14_h2o[-c(7,9,11)])))

rmse(bulb1$temp_c,bulb1$predict) #3.765109
rmse(bulb2$temp_c,bulb2$predict) #1.739043
rmse(bulb3$temp_c,bulb3$predict) #10.12168
rmse(bulb4$temp_c,bulb4$predict) #3.407646
rmse(bulb5$temp_c,bulb5$predict) #2.73814
rmse(bulb6$temp_c,bulb6$predict) #2.482949
rmse(bulb7$temp_c,bulb7$predict) #1.401555
rmse(bulb8$temp_c,bulb8$predict) #4.064616
rmse(bulb9$temp_c,bulb9$predict) #7.929102
rmse(bulb10$temp_c,bulb10$predict) #6.308332
rmse(bulb11$temp_c,bulb11$predict) #6.282534
rmse(bulb12$temp_c,bulb12$predict) #0.6292883
rmse(bulb13$temp_c,bulb13$predict) #0.8309286
rmse(bulb14$temp_c,bulb14$predict) #0.6835631

lens

unseen_lens %>% filter(w==15 & c==50 & box_x==65 & box_y==65 & box_z==64) -> lens1
unseen_lens %>% filter(w==20 & c==60 & box_x==65 & box_y==65 & box_z==64) -> lens2
unseen_lens %>% filter(w==35 & c==85 & box_x==80 & box_y==80 & box_z==94) -> lens3
unseen_lens %>% filter(w==40 & c==95 & box_x==80 & box_y==80 & box_z==94) -> lens4
unseen_lens %>% filter(w==15 & c==60 & box_x==60 & box_y==60 & box_z==64) -> lens5
unseen_lens %>% filter(w==20 & c==50 & box_x==60 & box_y==60 & box_z==64) -> lens6
unseen_lens %>% filter(w==20 & c==60 & box_x==60 & box_y==60 & box_z==64) -> lens7
unseen_lens %>% filter(w==15 & c==60 & box_x==60 & box_y==60 & box_z==84) -> lens8
unseen_lens %>% filter(w==20 & c==50 & box_x==65 & box_y==65 & box_z==84) -> lens9
unseen_lens %>% filter(w==35 & c==95 & box_x==90 & box_y==90 & box_z==94) -> lens10
unseen_lens %>% filter(w==40 & c==85 & box_x==90 & box_y==90 & box_z==104) -> lens11
unseen_lens %>% filter(w==35 & c==95 & box_x==100 & box_y==100 & box_z==104) -> lens12
unseen_lens %>% filter(w==40 & c==85 & box_x==100 & box_y==100 & box_z==104) -> lens13
unseen_lens %>% filter(w==40 & c==95 & box_x==100 & box_y==100 & box_z==104) -> lens14

lens1_h2o <- as.h2o(lens1)
lens2_h2o <- as.h2o(lens2)
lens3_h2o <- as.h2o(lens3)
lens4_h2o <- as.h2o(lens4)
lens5_h2o <- as.h2o(lens5)
lens6_h2o <- as.h2o(lens6)
lens7_h2o <- as.h2o(lens7)
lens8_h2o <- as.h2o(lens8)
lens9_h2o <- as.h2o(lens9)
lens10_h2o <- as.h2o(lens10)
lens11_h2o <- as.h2o(lens11)
lens12_h2o <- as.h2o(lens12)
lens13_h2o <- as.h2o(lens13)
lens14_h2o <- as.h2o(lens14)

lens1 <- cbind(lens1,as.data.frame(h2o.predict(h2o.getModel(plane_p_lens_model@model_ids[[1]]),lens1_h2o[-c(7,9,11)])))
lens2 <- cbind(lens2,as.data.frame(h2o.predict(h2o.getModel(plane_p_lens_model@model_ids[[1]]),lens2_h2o[-c(7,9,11)])))
lens3 <- cbind(lens3,as.data.frame(h2o.predict(h2o.getModel(plane_p_lens_model@model_ids[[1]]),lens3_h2o[-c(7,9,11)])))
lens4 <- cbind(lens4,as.data.frame(h2o.predict(h2o.getModel(plane_p_lens_model@model_ids[[1]]),lens4_h2o[-c(7,9,11)])))
lens5 <- cbind(lens5,as.data.frame(h2o.predict(h2o.getModel(plane_p_lens_model@model_ids[[1]]),lens5_h2o[-c(7,9,11)])))
lens6 <- cbind(lens6,as.data.frame(h2o.predict(h2o.getModel(plane_p_lens_model@model_ids[[1]]),lens6_h2o[-c(7,9,11)])))
lens7 <- cbind(lens7,as.data.frame(h2o.predict(h2o.getModel(plane_p_lens_model@model_ids[[1]]),lens7_h2o[-c(7,9,11)])))
lens8 <- cbind(lens8,as.data.frame(h2o.predict(h2o.getModel(plane_p_lens_model@model_ids[[1]]),lens8_h2o[-c(7,9,11)])))
lens9 <- cbind(lens9,as.data.frame(h2o.predict(h2o.getModel(plane_p_lens_model@model_ids[[1]]),lens9_h2o[-c(7,9,11)])))
lens10 <- cbind(lens10,as.data.frame(h2o.predict(h2o.getModel(plane_p_lens_model@model_ids[[1]]),lens10_h2o[-c(7,9,11)])))
lens11 <- cbind(lens11,as.data.frame(h2o.predict(h2o.getModel(plane_p_lens_model@model_ids[[1]]),lens11_h2o[-c(7,9,11)])))
lens12 <- cbind(lens12,as.data.frame(h2o.predict(h2o.getModel(plane_p_lens_model@model_ids[[1]]),lens12_h2o[-c(7,9,11)])))
lens13 <- cbind(lens13,as.data.frame(h2o.predict(h2o.getModel(plane_p_lens_model@model_ids[[1]]),lens13_h2o[-c(7,9,11)])))
lens14 <- cbind(lens14,as.data.frame(h2o.predict(h2o.getModel(plane_p_lens_model@model_ids[[1]]),lens14_h2o[-c(7,9,11)])))

rmse(lens1$temp_c,lens1$predict) #2.53
rmse(lens2$temp_c,lens2$predict) #2.91
rmse(lens3$temp_c,lens3$predict) #32.10
rmse(lens4$temp_c,lens4$predict) #5.50
rmse(lens5$temp_c,lens5$predict) #0.90
rmse(lens6$temp_c,lens6$predict) #2.43
rmse(lens7$temp_c,lens7$predict) #1.26
rmse(lens8$temp_c,lens8$predict) #5.69
rmse(lens9$temp_c,lens9$predict) #8.59
rmse(lens10$temp_c,lens10$predict) #1.47
rmse(lens11$temp_c,lens11$predict) #5.54
rmse(lens12$temp_c,lens12$predict) #1.93
rmse(lens13$temp_c,lens13$predict) #2.95
rmse(lens14$temp_c,lens14$predict) #2.16

up

unseen_up %>% filter(w==15 & c==50 & box_x==65 & box_y==65 & box_z==64) -> up1
unseen_up %>% filter(w==20 & c==60 & box_x==65 & box_y==65 & box_z==64) -> up2
unseen_up %>% filter(w==35 & c==85 & box_x==80 & box_y==80 & box_z==94) -> up3
unseen_up %>% filter(w==40 & c==95 & box_x==80 & box_y==80 & box_z==94) -> up4
unseen_up %>% filter(w==15 & c==60 & box_x==60 & box_y==60 & box_z==64) -> up5
unseen_up %>% filter(w==20 & c==50 & box_x==60 & box_y==60 & box_z==64) -> up6
unseen_up %>% filter(w==20 & c==60 & box_x==60 & box_y==60 & box_z==64) -> up7
unseen_up %>% filter(w==15 & c==60 & box_x==60 & box_y==60 & box_z==84) -> up8
unseen_up %>% filter(w==20 & c==50 & box_x==65 & box_y==65 & box_z==84) -> up9
unseen_up %>% filter(w==35 & c==95 & box_x==90 & box_y==90 & box_z==94) -> up10
unseen_up %>% filter(w==40 & c==85 & box_x==90 & box_y==90 & box_z==104) -> up11
unseen_up %>% filter(w==35 & c==95 & box_x==100 & box_y==100 & box_z==104) -> up12
unseen_up %>% filter(w==40 & c==85 & box_x==100 & box_y==100 & box_z==104) -> up13
unseen_up %>% filter(w==40 & c==95 & box_x==100 & box_y==100 & box_z==104) -> up14

up1_h2o <- as.h2o(up1)
up2_h2o <- as.h2o(up2)
up3_h2o <- as.h2o(up3)
up4_h2o <- as.h2o(up4)
up5_h2o <- as.h2o(up5)
up6_h2o <- as.h2o(up6)
up7_h2o <- as.h2o(up7)
up8_h2o <- as.h2o(up8)
up9_h2o <- as.h2o(up9)
up10_h2o <- as.h2o(up10)
up11_h2o <- as.h2o(up11)
up12_h2o <- as.h2o(up12)
up13_h2o <- as.h2o(up13)
up14_h2o <- as.h2o(up14)

up1 <- cbind(up1,as.data.frame(h2o.predict(h2o.getModel(plane_h_up_model@model_ids[[1]]),up1_h2o[-c(7,9,11)])))
up2 <- cbind(up2,as.data.frame(h2o.predict(h2o.getModel(plane_h_up_model@model_ids[[1]]),up2_h2o[-c(7,9,11)])))
up3 <- cbind(up3,as.data.frame(h2o.predict(h2o.getModel(plane_h_up_model@model_ids[[1]]),up3_h2o[-c(7,9,11)])))
up4 <- cbind(up4,as.data.frame(h2o.predict(h2o.getModel(plane_h_up_model@model_ids[[1]]),up4_h2o[-c(7,9,11)])))
up5 <- cbind(up5,as.data.frame(h2o.predict(h2o.getModel(plane_h_up_model@model_ids[[1]]),up5_h2o[-c(7,9,11)])))
up6 <- cbind(up6,as.data.frame(h2o.predict(h2o.getModel(plane_h_up_model@model_ids[[1]]),up6_h2o[-c(7,9,11)])))
up7 <- cbind(up7,as.data.frame(h2o.predict(h2o.getModel(plane_h_up_model@model_ids[[1]]),up7_h2o[-c(7,9,11)])))
up8 <- cbind(up8,as.data.frame(h2o.predict(h2o.getModel(plane_h_up_model@model_ids[[1]]),up8_h2o[-c(7,9,11)])))
up9 <- cbind(up9,as.data.frame(h2o.predict(h2o.getModel(plane_h_up_model@model_ids[[1]]),up9_h2o[-c(7,9,11)])))
up10 <- cbind(up10,as.data.frame(h2o.predict(h2o.getModel(plane_h_up_model@model_ids[[1]]),up10_h2o[-c(7,9,11)])))
up11 <- cbind(up11,as.data.frame(h2o.predict(h2o.getModel(plane_h_up_model@model_ids[[1]]),up11_h2o[-c(7,9,11)])))
up12 <- cbind(up12,as.data.frame(h2o.predict(h2o.getModel(plane_h_up_model@model_ids[[1]]),up12_h2o[-c(7,9,11)])))
up13 <- cbind(up13,as.data.frame(h2o.predict(h2o.getModel(plane_h_up_model@model_ids[[1]]),up13_h2o[-c(7,9,11)])))
up14 <- cbind(up14,as.data.frame(h2o.predict(h2o.getModel(plane_h_up_model@model_ids[[1]]),up14_h2o[-c(7,9,11)])))



rmse(up1$temp_c,up1$predict) #18.57
rmse(up2$temp_c,up2$predict) #11.64
rmse(up3$temp_c,up3$predict) #40.61
rmse(up4$temp_c,up4$predict) #3.88
rmse(up5$temp_c,up5$predict) #18.44
rmse(up6$temp_c,up6$predict) #11.58
rmse(up7$temp_c,up7$predict) #11.38
rmse(up8$temp_c,up8$predict) #18.87
rmse(up9$temp_c,up9$predict) #13.63
rmse(up10$temp_c,up10$predict) #3.57
rmse(up11$temp_c,up11$predict) #3.88
rmse(up12$temp_c,up12$predict) #2.59
rmse(up13$temp_c,up13$predict) #2.77
rmse(up14$temp_c,up14$predict) #2.42