Análisis del crimen en L.A. con H2O
Primero, las librerías
# Para cada librería preguntamos primero si está instalada
if(!is.element("dplyr", installed.packages()[, 1]))
install.packages("dplyr", repos = 'http://cran.us.r-project.org')
library(dplyr)
if(!is.element("ggplot2", installed.packages()[, 1]))
install.packages("ggplot2", repos = 'http://cran.us.r-project.org')
library(ggplot2)
if(!is.element("lubridate", installed.packages()[, 1]))
install.packages("lubridate", repos = 'http://cran.us.r-project.org')
library(lubridate)
if(!is.element("chron", installed.packages()[, 1]))
install.packages("chron", repos = 'http://cran.us.r-project.org')
library(chron)
if(!is.element("pROC", installed.packages()[, 1]))
install.packages("pROC", repos = 'http://cran.us.r-project.org')
library(pROC)La de H2O la tratamos aparte. Si hay que instalarla o actualizarla, es mejor descargarla desde su cuenta de AWS
# http://h2o-release.s3.amazonaws.com/h2o/rel-wolpert/2/index.html
# The following two commands remove any previously installed H2O packages for R.
if ("package:h2o" %in% search()) { detach("package:h2o", unload=TRUE) }
if ("h2o" %in% rownames(installed.packages())) { remove.packages("h2o") }
# Next, we download packages that H2O depends on.
pkgs <- c("RCurl","jsonlite")
for (pkg in pkgs) {
if (! (pkg %in% rownames(installed.packages()))) { install.packages(pkg) }
}
# Now we download, install and initialize the H2O package for R.
install.packages("h2o", type="source", repos="http://h2o-release.s3.amazonaws.com/h2o/rel-wolpert/2/R")Ahora ya podemos cargar el paquete
library(h2o)##
## ----------------------------------------------------------------------
##
## Your next step is to start H2O:
## > h2o.init()
##
## For H2O package documentation, ask for help:
## > ??h2o
##
## After starting H2O, you can use the Web UI at http://localhost:54321
## For more information visit http://docs.h2o.ai
##
## ----------------------------------------------------------------------##
## Attaching package: 'h2o'## The following object is masked from 'package:pROC':
##
## var## The following objects are masked from 'package:lubridate':
##
## day, hour, month, week, year## The following objects are masked from 'package:stats':
##
## cor, sd, var## The following objects are masked from 'package:base':
##
## ||, &&, %*%, apply, as.factor, as.numeric, colnames,
## colnames<-, ifelse, %in%, is.character, is.factor, is.numeric,
## log, log10, log1p, log2, round, signif, truncObtenemos los datos de la web: https://data.lacity.org/A-Safe-City/Crime-Data-from-2010-to-Present/y8tr-7khq/data en la pestaña export y en download as elegimos el formato csv
Cargamos los datos (acordándonos de modificar la ruta)
# Datos descargados el 15-02-2018
crime_data <- read.csv('/media/enero/Disco1/Varios/R/Archivos/clases_r/h2o/data/Crime_Data_from_2010_to_Present-15-02-2018.csv')Un summary()
summary(crime_data)## DR.Number Date.Reported Date.Occurred
## Min. : 210 08/07/2017: 805 01/01/2010: 2155
## 1st Qu.:120315238 12/18/2017: 794 01/01/2011: 2069
## Median :140620602 07/24/2017: 783 01/01/2012: 1664
## Mean :138030952 10/23/2017: 771 01/01/2013: 1489
## 3rd Qu.:160624731 11/28/2017: 762 01/01/2014: 1353
## Max. :910220366 01/29/2018: 761 01/01/2015: 1301
## (Other) :1678633 (Other) :1673278
## Time.Occurred Area.ID Area.Name Reporting.District
## Min. : 1 Min. : 1.00 77th Street: 117154 Min. : 100
## 1st Qu.: 930 1st Qu.: 6.00 Southwest : 108517 1st Qu.: 645
## Median :1430 Median :12.00 N Hollywood: 91554 Median :1204
## Mean :1364 Mean :11.14 Pacific : 88964 Mean :1161
## 3rd Qu.:1900 3rd Qu.:16.00 Southeast : 88578 3rd Qu.:1675
## Max. :2359 Max. :21.00 Mission : 84802 Max. :2198
## (Other) :1103740
## Crime.Code Crime.Code.Description
## Min. :110.0 BATTERY - SIMPLE ASSAULT :154114
## 1st Qu.:330.0 BURGLARY FROM VEHICLE :129308
## Median :440.0 VEHICLE - STOLEN :129271
## Mean :506.8 BURGLARY :121439
## 3rd Qu.:626.0 THEFT PLAIN - PETTY ($950 & UNDER):120106
## Max. :956.0 THEFT OF IDENTITY :105852
## (Other) :923219
## MO.Codes Victim.Age Victim.Sex Victim.Descent
## : 182688 Min. :10.00 :155068 H :582840
## 0344 : 182670 1st Qu.:23.00 -: 1 W :414424
## 0329 : 72917 Median :34.00 F:715645 B :269782
## 1501 : 36281 Mean :35.97 H: 56 O :162664
## 0416 : 25262 3rd Qu.:48.00 M:784098 :155104
## 0325 : 22698 Max. :99.00 X: 28441 X : 47013
## (Other):1160793 NA's :135446 (Other): 51482
## Premise.Code Premise.Description
## Min. :101 STREET :375205
## 1st Qu.:102 SINGLE FAMILY DWELLING :346791
## Median :210 MULTI-UNIT DWELLING (APARTMENT, DUPLEX, ETC):216227
## Mean :312 PARKING LOT :119893
## 3rd Qu.:501 SIDEWALK : 84044
## Max. :971 OTHER BUSINESS : 75323
## NA's :97 (Other) :465826
## Weapon.Used.Code
## Min. :101.0
## 1st Qu.:400.0
## Median :400.0
## Mean :370.6
## 3rd Qu.:400.0
## Max. :516.0
## NA's :1125843
## Weapon.Description
## :1125844
## STRONG-ARM (HANDS, FIST, FEET OR BODILY FORCE): 339404
## VERBAL THREAT : 46246
## UNKNOWN WEAPON/OTHER WEAPON : 43517
## HAND GUN : 26849
## SEMI-AUTOMATIC PISTOL : 10620
## (Other) : 90829
## Status.Code Status.Description Crime.Code.1
## IC :1302599 Adult Arrest: 172666 Min. :110.0
## AO : 190613 Adult Other : 190613 1st Qu.:330.0
## AA : 172666 Invest Cont :1302599 Median :440.0
## JA : 13246 Juv Arrest : 13246 Mean :506.7
## JO : 4154 Juv Other : 4154 3rd Qu.:626.0
## CC : 26 UNK : 31 Max. :999.0
## (Other): 5 NA's :6
## Crime.Code.2 Crime.Code.3 Crime.Code.4
## Min. :210.0 Min. : 93.0 Min. :421.0
## 1st Qu.:998.0 1st Qu.:998.0 1st Qu.:998.0
## Median :998.0 Median :998.0 Median :998.0
## Mean :953.3 Mean :970.3 Mean :969.9
## 3rd Qu.:998.0 3rd Qu.:998.0 3rd Qu.:998.0
## Max. :999.0 Max. :999.0 Max. :999.0
## NA's :1576655 NA's :1680910 NA's :1683231
## Address
## 6TH ST : 3897
## 7TH ST : 3014
## 9300 TAMPA AV: 2873
## 5TH ST : 2415
## 6TH : 2409
## 6600 TOPANGA CANYON BL: 2340
## (Other) :1666361
## Cross.Street Location
## :1403492 (0, 0) : 6416
## BROADWAY : 4923 (34.1016, -118.3387): 3192
## FIGUEROA : 2969 (34.1905, -118.6059): 2316
## VERMONT AV: 2953 (33.9892, -118.3089): 2001
## SAN PEDRO : 2821 (34.1576, -118.438) : 1797
## WESTERN AV: 2821 (34.2216, -118.4488): 1684
## (Other) : 263330 (Other) :1665903Y un str()
str(crime_data)## 'data.frame': 1683309 obs. of 26 variables:
## $ DR.Number : int 1208575 102005556 418 101822289 42104479 120125367 101105609 101620051 101910498 120908292 ...
## $ Date.Reported : Factor w/ 2963 levels "01/01/2010","01/01/2011",..: 623 217 663 2556 95 67 244 2556 812 743 ...
## $ Date.Occurred : Factor w/ 2963 levels "01/01/2010","01/01/2011",..: 599 190 655 2548 32 67 235 2524 812 130 ...
## $ Time.Occurred : int 1800 2300 2030 1800 2300 1400 2230 1600 1600 800 ...
## $ Area.ID : int 12 20 18 18 21 1 11 16 19 9 ...
## $ Area.Name : Factor w/ 21 levels "77th Street",..: 1 12 15 15 17 2 11 4 8 18 ...
## $ Reporting.District : int 1241 2071 1823 1803 2133 111 1125 1641 1902 904 ...
## $ Crime.Code : int 626 510 510 510 745 110 510 510 510 668 ...
## $ Crime.Code.Description: Factor w/ 143 levels "","ABORTION/ILLEGAL",..: 76 139 139 139 137 40 139 139 139 59 ...
## $ MO.Codes : Factor w/ 370891 levels "","0100","0100 0104",..: 168286 1 1 1 48103 264208 1 1 1 107446 ...
## $ Victim.Age : int 30 NA 12 NA 84 49 NA NA NA 27 ...
## $ Victim.Sex : Factor w/ 6 levels "","-","F","H",..: 3 1 1 1 5 3 1 1 1 3 ...
## $ Victim.Descent : Factor w/ 21 levels "","-","A","B",..: 19 1 1 1 19 19 1 1 1 14 ...
## $ Premise.Code : int 502 101 101 101 501 501 108 101 101 203 ...
## $ Premise.Description : Factor w/ 304 levels "","7TH AND METRO CENTER (NOT LINE SPECIFIC)",..: 193 261 261 261 250 250 214 261 261 206 ...
## $ Weapon.Used.Code : int 400 NA NA NA NA 400 NA NA NA NA ...
## $ Weapon.Description : Factor w/ 80 levels "","AIR PISTOL/REVOLVER/RIFLE/BB GUN",..: 67 1 1 1 1 67 1 1 1 1 ...
## $ Status.Code : Factor w/ 10 levels "","13","19","AA",..: 5 7 7 7 7 4 7 7 7 7 ...
## $ Status.Description : Factor w/ 6 levels "Adult Arrest",..: 2 3 3 3 3 1 3 3 3 3 ...
## $ Crime.Code.1 : int 626 510 510 510 745 110 510 510 510 668 ...
## $ Crime.Code.2 : int NA NA NA NA NA NA NA NA NA NA ...
## $ Crime.Code.3 : int NA NA NA NA NA NA NA NA NA NA ...
## $ Crime.Code.4 : int NA NA NA NA NA NA NA NA NA NA ...
## $ Address : Factor w/ 71796 levels "00","00000 20TH AV",..: 49292 70835 25051 58794 53019 48048 71755 64955 65708 53176 ...
## $ Cross.Street : Factor w/ 11487 levels "","100TH ST",..: 1 139 1 10697 1 1 1121 10158 3362 1 ...
## $ Location : Factor w/ 61121 levels "","(0, 0)","(33.3427, -118.3258)",..: 9181 20839 4882 6203 45655 24053 34484 52498 60520 45810 ...Es un dataset grande, de mas de 1,5 millones de observaciones
Cambio de nombre de las variables
Vemos los nombres
names(crime_data)## [1] "DR.Number" "Date.Reported"
## [3] "Date.Occurred" "Time.Occurred"
## [5] "Area.ID" "Area.Name"
## [7] "Reporting.District" "Crime.Code"
## [9] "Crime.Code.Description" "MO.Codes"
## [11] "Victim.Age" "Victim.Sex"
## [13] "Victim.Descent" "Premise.Code"
## [15] "Premise.Description" "Weapon.Used.Code"
## [17] "Weapon.Description" "Status.Code"
## [19] "Status.Description" "Crime.Code.1"
## [21] "Crime.Code.2" "Crime.Code.3"
## [23] "Crime.Code.4" "Address"
## [25] "Cross.Street" "Location"Vamos a cambiarlas
# Minusculas
names(crime_data) <- tolower(names(crime_data))
# cambiar . por _
names(crime_data) <- gsub('.', '_', names(crime_data), fixed = TRUE)
# Las vemos
names(crime_data)## [1] "dr_number" "date_reported"
## [3] "date_occurred" "time_occurred"
## [5] "area_id" "area_name"
## [7] "reporting_district" "crime_code"
## [9] "crime_code_description" "mo_codes"
## [11] "victim_age" "victim_sex"
## [13] "victim_descent" "premise_code"
## [15] "premise_description" "weapon_used_code"
## [17] "weapon_description" "status_code"
## [19] "status_description" "crime_code_1"
## [21] "crime_code_2" "crime_code_3"
## [23] "crime_code_4" "address"
## [25] "cross_street" "location"Análisis de NAs
Algunos cambios en valores para pasarlos a NA
Mo Codes
# Los pasamos a caracter
crime_data$mo_codes <- as.character(crime_data$mo_codes)
# Cambiamos los valores en blanco por NA
crime_data$mo_codes[crime_data$mo_codes==''] <- NA
# table(crime_data$mo_code, useNA = 'ifany') # Ocupa muchoCreamos un objeto con los valores únicos
# Los valores únicos de crime_data$mo_codes
mo_codes <- unique(as.character(crime_data$mo_codes))
# Quitamos los NAs
mo_codes <- mo_codes[!is.na(mo_codes)]# Sacamos los valores únicos
mo_codes_unique <- unique(paste(mo_codes, collapse = ' '))
# Separamos los valores con ' '
mo_codes <- unique(strsplit(mo_codes_unique, ' '))
# Pasamos a character
mo_codes <- unlist(mo_codes)
# Nos quedamos con los valores únicos
mo_codes <- unique(mo_codes)
# Los ordenamos
mo_codes <- mo_codes[order(mo_codes)]
# Cuantos hay
length(mo_codes)## [1] 674Aquí podríamos obtener un one-hot-encoding de los mo_codes. No lo hacemos porque estamos evaluando la eficacia de autoML, pero es uno de los temas a tratar para mejorar el modelado
Status codes
# Los pasamos a caracter
crime_data$status_code <- as.character(crime_data$status_code)
# Cambiamos los valores en blanco por NA
crime_data$status_code[crime_data$status_code==''] <- NA
# Volvemos a pasar a factor
crime_data$status_code <- as.factor(crime_data$status_code)
# Sacamos una tabla
table(crime_data$status_code, useNA = 'ifany')##
## 13 19 AA AO CC IC JA JO TH
## 1 1 172666 190613 26 1302599 13246 4154 1
## <NA>
## 2Cross Street
# Pasamos a caracter
crime_data$cross_street <- as.character(crime_data$cross_street)
# Cambiamos los valores en blanco por NA
crime_data$cross_street[crime_data$cross_street %in% c('')] <- NA
# table(crime_data$cross_street, useNA = 'ifany') # Muy largo
# Volvemos a pasar a factor
crime_data$cross_street <- as.factor(crime_data$cross_street)Weapon description
# Pasamos a caracter
crime_data$weapon_description <- as.character(crime_data$weapon_description)
# Cambiamos los valores en blanco por NA
crime_data$weapon_description[crime_data$weapon_description %in% c('')] <- NA
# table(crime_data$weapon_description, useNA = 'ifany') # Muy largo
# Volvemos a pasar a factor
crime_data$weapon_description <- as.factor(crime_data$weapon_description)Crime code description
crime_data$crime_code_description <- as.character(crime_data$crime_code_description)
crime_data$crime_code_description[crime_data$crime_code_description %in% c('')] <- NA
# table(crime_data$crime_code_description, useNA = 'ifany')
# Volvemos a pasar a factor
crime_data$crime_code_description <- as.factor(crime_data$cross_street)Premise description
# Pasamos a caracter
crime_data$premise_description <- as.character(crime_data$premise_description)
# Cambiamos los valores en blanco por NA
crime_data$premise_description[crime_data$premise_description %in% c('')] <- NA
# table(crime_data$premise_description, useNA = 'ifany')
# Volvemos a pasar a factor
crime_data$premise_description <- as.factor(crime_data$premise_description)Address
# Pasamos a caracter
crime_data$address <- as.character(crime_data$address)
# Cambiamos los valores '00' por NA
crime_data$address[crime_data$address=='00'] <- NA
# table(crime_data$address, useNA = 'ifany')
# Pasamos a factor
crime_data$address <- as.factor(crime_data$address)Sex
# Pasamos a caracter
crime_data$victim_sex <- as.character(crime_data$victim_sex)
# Cambiamos los valores en blanco y los '-' por NA
crime_data$victim_sex[crime_data$victim_sex %in% c('', '-')] <- NA
# Una tabla
table(crime_data$victim_sex, useNA = 'ifany')##
## F H M X <NA>
## 715645 56 784098 28441 155069# Volvemos a pasar a factor
crime_data$victim_sex <- as.factor(crime_data$victim_sex)Descent Code
A - Other Asian B - Black C - Chinese D - Cambodian F - Filipino G - Guamanian H - Hispanic/Latin/Mexican I - American Indian/Alaskan Native J - Japanese K - Korean L - Laotian O - Other P - Pacific Islander S - Samoan U - Hawaiian V - Vietnamese W - White X - Unknown Z - Asian Indian
# Renombramos los niveles
levels(crime_data$victim_descent) <- c("", "-", " Other_Asian", "Black",
"Chinese", "Cambodian", "Filipino",
"Guamanian", "Hispanic-Latin-Mexican",
"American_Indian-Alaskan_Native",
"Japanese", "Korean", "Laotian", "Other",
"Pacific_Islander", "Samoan",
"Hawaiian", "Vietnamese", "White",
"Unknown", "Asian_Indian")
# Pasamos a caracter
crime_data$victim_descent <- as.character(crime_data$victim_descent)
# Cambiamos los valores en blanco y los '-' por 'Unknown'
crime_data$victim_descent[crime_data$victim_descent %in% c('', '-')] <- 'Unknown'
# Una tabla
table(crime_data$victim_descent, useNA = 'ifany')##
## American_Indian-Alaskan_Native Asian_Indian
## 711 58
## Black Cambodian
## 269782 15
## Chinese Filipino
## 645 1981
## Guamanian Hawaiian
## 69 156
## Hispanic-Latin-Mexican Japanese
## 582840 248
## Korean Laotian
## 7531 10
## Other Other_Asian
## 162664 39648
## Pacific_Islander Samoan
## 292 26
## Unknown Vietnamese
## 202118 91
## White
## 414424# Y volvemos a factor
crime_data$victim_descent <- as.factor(crime_data$victim_descent)Location
# pasamos a caracter
crime_data$location <- as.character(crime_data$location)
# Los que tienen location '' los pasamos a '(0, 0)'
crime_data$location[crime_data$location==""] <- "(0, 0)"
# table(crime_data$location, useNA = 'ifany')Creamos un objeto (crime_data_location) y limpiamos crime_data_location, quitando ‘’, ‘(’ y ‘)’
# Copiamos de crime_data$location
crime_data_location <- crime_data$location
# Quitamos los '('
crime_data_location <- gsub('(', '', crime_data_location, fixed = TRUE)
# Quitamos los ')'
crime_data_location <- gsub(')', '', crime_data_location, fixed = TRUE)
# Quitamos los ' '
crime_data_location <- gsub(' ', '', crime_data_location, fixed = TRUE)temp2 <- strsplit(crime_data_location, ',')
# temp3 <- as.data.frame(temp2)
temp4 <- unlist(temp2)
temp5 <- stringr::str_split(crime_data_location, ',')
# temp6 <- as.data.frame(temp5, optional = TRUE)
# https://stackoverflow.com/questions/4227223/r-list-to-data-frame
temp7 <- do.call(rbind.data.frame, temp2)
# temp8 <- do.call(rbind, lapply(temp2, data.frame, stringsAsFactors=FALSE))
# temp9 <- do.call(rbind, lapply(temp2, matrix, stringsAsFactors=FALSE))
temp10 <- data.frame(matrix(unlist(temp2), nrow=1674208, byrow=T),
stringsAsFactors=FALSE)Vamos a incluir en crime_data las variables location_x y location_y
# https://stackoverflow.com/questions/4227223/r-list-to-data-frame
crime_data_location_df <- data.frame(matrix(unlist(strsplit(crime_data_location, ',')),
nrow=length(crime_data_location),
byrow=T), stringsAsFactors=FALSE)
# Cambiamos los nombres
names(crime_data_location_df) <- c('location_x', 'location_y')
# Pasamos a numérico
crime_data$location_x <- as.numeric(crime_data_location_df$location_x)
crime_data$location_y <- as.numeric(crime_data_location_df$location_y)
# Cambiamos los 0 por NAs
crime_data$location_x[crime_data$location_x==0] <- NA
crime_data$location_y[crime_data$location_y==0] <- NA
# Cambiamos también los (0, 0) por NAs
crime_data$location[crime_data$location=="(0, 0)"] <- NA
# Y pasamos location a factor (ya tenemos un location_x y location_y numericos)
crime_data$location <- as.factor(crime_data$location)Limpiamos el entorno
# Borramos dos datasets
rm(crime_data_location_df, crime_data_location)
# Limpiamos y vemos la memoria
gc()## used (Mb) gc trigger (Mb) max used (Mb)
## Ncells 1314514 70.3 3886542 207.6 3409184 182.1
## Vcells 31284553 238.7 93917411 716.6 93746544 715.3Creamos un contador de NAs para cada columna
# Creamos la nueva variable con un count
na_count <- do.call(rbind, lapply(colnames(crime_data), function(x) {return(data.frame(x, sum(is.na(crime_data[[x]]))))}))
# Cambiamos el nombre de las variables en el dataset
colnames(na_count) <- c('attribute_name', 'na_count')
# na_count$attribute_name <- as.character(na_count$attribute_name)
# Nos quedamos con los distintos de cero
na_count <- na_count[na_count$na_count != 0,]
# Lo vemos en gráfico
plot(na_count)
Nuevas variables
# Creamos una variable con el mes del crimen
crime_data$crime_month <- as.factor(substr(as.character(crime_data$date_occurred), 1, 2))
# Vemos en tabla
table(crime_data$crime_month, useNA = 'ifany')##
## 01 02 03 04 05 06 07 08 09 10
## 160757 127346 138530 135394 141384 139076 143685 144028 138439 143885
## 11 12
## 132921 137864Cambiamos los valores tipo fecha a formato as.Date
crime_data$date_reported <- as.Date(as.character(crime_data$date_reported), '%m/%d/%Y')
crime_data$date_occurred <- as.Date(as.character(crime_data$date_occurred), '%m/%d/%Y')Ordenamos por fecha
crime_data <- crime_data[order(crime_data$date_occurred), ]Generamos algunas columnas
Año
# Con lubridate
# crime_data$crime_year <- year(crime_data$date_occurred)
crime_data$crime_year <- as.factor(substr(as.character(crime_data$date_occurred), 1, 4))
table(crime_data$crime_year, useNA = 'ifany')##
## 2010 2011 2012 2013 2014 2015 2016 2017 2018
## 208631 200251 200888 191887 194632 214212 223690 227931 21187Mes
# Mes
# Con lubridate
# crime_data$crime_month <- month(crime_data$date_occurred, label = TRUE)
crime_data$crime_month <- as.factor(substr(as.character(crime_data$date_occurred), 6, 7))
table(crime_data$crime_month, useNA = 'ifany')##
## 01 02 03 04 05 06 07 08 09 10
## 160757 127346 138530 135394 141384 139076 143685 144028 138439 143885
## 11 12
## 132921 137864Semana
crime_data$crime_week <- week(crime_data$date_occurred)
table(crime_data$crime_week, useNA = 'ifany')Convertimos el tiempo con el paquete chron
crime_data$crime_time <- times(substr(as.POSIXct(sprintf("%04.0f", crime_data$time_occurred), origin = '1970-01-01', "Asia/Calcutta", format='%H%M'), 12, 19))Creamos una variable minuto
crime_data$crime_minute <- (crime_data$time_occurred %/% 100)*60 + (crime_data$time_occurred %% 100)*100
table(crime_data$crime_minute, useNA = 'ifany')##
## 60 100 120 160 180 200 220 240 260 280 300 320
## 20680 29008 17806 16 13081 77 16 9022 29 22 7879 19
## 340 360 380 400 420 440 460 480 500 520 540 560
## 9 12866 26 52 17138 28 34 41277 8995 45 31455 800
## 580 600 620 640 660 680 700 720 740 760 780 800
## 37 34706 789 40 27988 592 86 93519 487 70 30837 432
## 820 840 860 880 900 920 940 960 980 1000 1020 1040
## 155 34841 421 68 39481 550 74 39456 749 3365 44112 795
## 1060 1080 1100 1120 1140 1160 1180 1200 1220 1240 1260 1280
## 1836 50136 983 1577 44675 1100 1205 46920 1355 961 43079 1211
## 1300 1320 1340 1360 1380 1400 1420 1440 1460 1480 1500 1520
## 873 44105 1343 812 34900 1369 1286 122 1346 1639 3755 1305
## 1540 1560 1580 1600 1620 1640 1660 1680 1700 1720 1740 1760
## 1633 2590 1359 1706 2436 1302 2049 1828 1456 2450 1379 1345
## 1780 1800 1820 1840 1860 1880 1900 1920 1940 1960 1980 2000
## 2557 1319 1252 2675 1536 1014 2802 2572 139 2810 3074 2394
## 2020 2040 2060 2080 2100 2120 2140 2160 2180 2200 2220 2240
## 2657 3271 2358 2655 3799 2076 2670 3779 1602 3099 4302 1318
## 2260 2280 2300 2320 2340 2360 2380 2400 2420 2440 2460 2480
## 2626 4034 1194 2564 4341 1179 2173 4842 1900 170 4283 2247
## 2500 2520 2540 2560 2580 2600 2620 2640 2660 2680 2700 2720
## 1285 4671 2232 1100 4623 2868 925 4698 3009 746 4769 3112
## 2740 2760 2780 2800 2820 2840 2860 2880 2900 2920 2940 2960
## 616 4409 3405 573 4113 3826 574 3560 3761 757 166 3570
## 2980 3000 3020 3040 3060 3080 3100 3120 3140 3160 3180 3200
## 909 11729 3597 1000 9782 3738 1238 7778 3844 1308 5242 3628
## 3220 3240 3260 3280 3300 3320 3340 3360 3380 3400 3420 3440
## 1505 4431 3558 1517 4581 3071 1614 6409 2771 1604 10260 163
## 3460 3480 3500 3520 3540 3560 3580 3600 3620 3640 3660 3680
## 1513 13182 1069 1707 12650 976 1554 14451 863 1607 14138 694
## 3700 3720 3740 3760 3780 3800 3820 3840 3860 3880 3900 3920
## 1659 14330 522 1497 14500 446 1374 16570 483 1161 18353 777
## 3940 3960 3980 4000 4020 4040 4060 4080 4100 4120 4140 4160
## 112 18389 837 1950 19603 992 2084 20945 1155 1739 20827 1251
## 4180 4200 4220 4240 4260 4280 4300 4320 4340 4360 4380 4400
## 1377 20649 1364 938 20509 1444 1066 19558 1425 1259 16228 1608
## 4420 4440 4460 4480 4500 4520 4540 4560 4580 4600 4620 4640
## 1871 130 1470 1926 2909 1539 2191 2805 1443 2534 2317 1569
## 4660 4680 4700 4720 4740 4760 4780 4800 4820 4840 4860 4880
## 2924 1815 1602 2933 1517 1405 3117 1598 1275 3258 2171 1233
## 4900 4920 4940 4960 4980 5000 5020 5040 5060 5080 5100 5120
## 3278 3354 138 3338 3283 1952 3303 3678 2230 3333 3983 1723
## 5140 5160 5180 5200 5220 5240 5260 5280 5300 5320 5340 5360
## 3391 4639 1347 3381 4672 1107 3419 4556 1070 2951 4699 1396
## 5380 5400 5420 5440 5460 5480 5500 5520 5540 5560 5580 5600
## 2795 5067 2073 125 5271 2053 932 5346 2369 931 5408 2518
## 5620 5640 5660 5680 5700 5720 5740 5760 5780 5800 5820 5840
## 832 5426 2912 625 5162 3014 500 4983 3350 517 4047 3311
## 5860 5880 5900 5920 5940 5960 5980 6000 6020 6040 6060 6080
## 592 4708 3290 824 130 3167 869 133 3470 970 135 3369
## 6100 6120 6140 6160 6180 6200 6220 6240 6260 6280 6300 6320
## 1110 147 3356 1426 149 3129 1297 103 3079 1301 129 2927
## 6340 6360 6380 6400 6420 6440 6460 6480 6500 6520 6540 6560
## 1397 104 3609 1280 110 116 1257 105 115 1480 84 151
## 6580 6600 6620 6640 6660 6680 6700 6720 6740 6760 6780 6800
## 1390 87 99 1361 78 118 1361 83 72 1214 69 79
## 6820 6840 6860 6880 6900 6920 6940 6960 6980 7000 7020 7040
## 1082 43 76 2666 28 69 51 26 62 38 29 26
## 7060 7080 7100 7120 7160 7180 7220 7280
## 40 32 34 51 27 41 29 1121Establecemos algunas franjas horarias
crime_data$crime_time_interval <- cut(crime_data$crime_time, breaks = times(c('00:00:00', '05:00:00', '09:00:00', '13:00:00', '17:00:00', '21:00:00', '23:59:00')), labels = c('Late_night', 'Early_morning', 'Late_morning', 'Early_afternoon', 'Late_afternoon', 'Evening'), include.lowest = TRUE)
table(crime_data$crime_time_interval, useNA = 'ifany')##
## Late_night Early_morning Late_morning Early_afternoon
## 218525 186980 336565 340978
## Late_afternoon Evening
## 385880 214381Eliminamos la variable crime_time, ya que es equivalente a crime_minute
# crime_data$crime_time <- as.factor(crime_data$crime_time)
crime_data$crime_time <- NULLDía de la semana
crime_data$occurred_week_day <- factor(weekdays(crime_data$date_occurred),
levels = c("lunes", "martes",
"miércoles", "jueves",
"viernes", "sábado",
"domingo"))
table(crime_data$occurred_week_day, useNA = 'ifany')##
## lunes martes miércoles jueves viernes sábado domingo
## 239881 235817 238315 236632 259729 242043 230892Tiempo de retraso a la hora de denunciar
crime_data$delay_reported <- crime_data$date_reported - crime_data$date_occurred
crime_data$delay_reported <- as.numeric(crime_data$delay_reported)# En esta parte se crean nuevas variables a partir de la suma y de la media
# No se ejecuta al tener eval=FALSE
week_day_occurred_sum <- aggregate(date_occurred ~ occurred_week_day,
data = crime_data, "sum")
week_day_occurred_mean <- aggregate(date_occurred ~ occured_week_day,
data = crime_data, "mean")
all_sum <- aggregate(target ~ date_occurred + occurred_week_day,
data = train, "sum")
all_mean <- aggregate(target ~ date_occurred + occured_week_day,
data = train, "mean")
plot(all_mean$date_occurred, all_mean$target)Otro summary()
summary(crime_data)## dr_number date_reported date_occurred
## Min. : 210 Min. :2010-01-01 Min. :2010-01-01
## 1st Qu.:120315238 1st Qu.:2012-02-11 1st Qu.:2012-01-21
## Median :140620602 Median :2014-04-13 Median :2014-03-21
## Mean :138030952 Mean :2014-03-08 Mean :2014-02-19
## 3rd Qu.:160624731 3rd Qu.:2016-04-13 3rd Qu.:2016-03-28
## Max. :910220366 Max. :2018-02-10 Max. :2018-02-10
##
## time_occurred area_id area_name reporting_district
## Min. : 1 Min. : 1.00 77th Street: 117154 Min. : 100
## 1st Qu.: 930 1st Qu.: 6.00 Southwest : 108517 1st Qu.: 645
## Median :1430 Median :12.00 N Hollywood: 91554 Median :1204
## Mean :1364 Mean :11.14 Pacific : 88964 Mean :1161
## 3rd Qu.:1900 3rd Qu.:16.00 Southeast : 88578 3rd Qu.:1675
## Max. :2359 Max. :21.00 Mission : 84802 Max. :2198
## (Other) :1103740
## crime_code crime_code_description
## Min. :110.0 BROADWAY : 4923
## 1st Qu.:330.0 FIGUEROA : 2969
## Median :440.0 VERMONT AV: 2953
## Mean :506.8 SAN PEDRO : 2821
## 3rd Qu.:626.0 WESTERN AV: 2821
## Max. :956.0 (Other) : 263330
## NA's :1403492
## mo_codes victim_age victim_sex
## Length:1683309 Min. :10.00 F :715645
## Class :character 1st Qu.:23.00 H : 56
## Mode :character Median :34.00 M :784098
## Mean :35.97 X : 28441
## 3rd Qu.:48.00 NA's:155069
## Max. :99.00
## NA's :135446
## victim_descent premise_code
## Hispanic-Latin-Mexican:582840 Min. :101
## White :414424 1st Qu.:102
## Black :269782 Median :210
## Unknown :202118 Mean :312
## Other :162664 3rd Qu.:501
## Other_Asian : 39648 Max. :971
## (Other) : 11833 NA's :97
## premise_description weapon_used_code
## STREET :375205 Min. :101.0
## SINGLE FAMILY DWELLING :346791 1st Qu.:400.0
## MULTI-UNIT DWELLING (APARTMENT, DUPLEX, ETC):216227 Median :400.0
## PARKING LOT :119893 Mean :370.6
## SIDEWALK : 84044 3rd Qu.:400.0
## (Other) :537862 Max. :516.0
## NA's : 3287 NA's :1125843
## weapon_description
## STRONG-ARM (HANDS, FIST, FEET OR BODILY FORCE): 339404
## VERBAL THREAT : 46246
## UNKNOWN WEAPON/OTHER WEAPON : 43517
## HAND GUN : 26849
## SEMI-AUTOMATIC PISTOL : 10620
## (Other) : 90829
## NA's :1125844
## status_code status_description crime_code_1
## IC :1302599 Adult Arrest: 172666 Min. :110.0
## AO : 190613 Adult Other : 190613 1st Qu.:330.0
## AA : 172666 Invest Cont :1302599 Median :440.0
## JA : 13246 Juv Arrest : 13246 Mean :506.7
## JO : 4154 Juv Other : 4154 3rd Qu.:626.0
## (Other): 29 UNK : 31 Max. :999.0
## NA's : 2 NA's :6
## crime_code_2 crime_code_3 crime_code_4
## Min. :210.0 Min. : 93.0 Min. :421.0
## 1st Qu.:998.0 1st Qu.:998.0 1st Qu.:998.0
## Median :998.0 Median :998.0 Median :998.0
## Mean :953.3 Mean :970.3 Mean :969.9
## 3rd Qu.:998.0 3rd Qu.:998.0 3rd Qu.:998.0
## Max. :999.0 Max. :999.0 Max. :999.0
## NA's :1576655 NA's :1680910 NA's :1683231
## address
## 6TH ST : 3897
## 7TH ST : 3014
## 9300 TAMPA AV: 2873
## 5TH ST : 2415
## 6TH : 2409
## (Other) :1668668
## NA's : 33
## cross_street location
## BROADWAY : 4923 (34.1016, -118.3387): 3192
## FIGUEROA : 2969 (34.1905, -118.6059): 2316
## VERMONT AV: 2953 (33.9892, -118.3089): 2001
## SAN PEDRO : 2821 (34.1576, -118.438) : 1797
## WESTERN AV: 2821 (34.2216, -118.4488): 1684
## (Other) : 263330 (Other) :1665894
## NA's :1403492 NA's : 6425
## location_x location_y crime_month crime_year
## Min. :33.34 Min. :-118.8 01 :160757 2017 :227931
## 1st Qu.:34.01 1st Qu.:-118.4 08 :144028 2016 :223690
## Median :34.06 Median :-118.3 10 :143885 2015 :214212
## Mean :34.08 Mean :-118.4 07 :143685 2010 :208631
## 3rd Qu.:34.18 3rd Qu.:-118.3 05 :141384 2012 :200888
## Max. :34.79 Max. :-117.7 06 :139076 2011 :200251
## NA's :6425 NA's :6425 (Other):810494 (Other):407706
## crime_minute crime_time_interval occurred_week_day
## Min. : 60 Late_night :218525 lunes :239881
## 1st Qu.: 840 Early_morning :186980 martes :235817
## Median :1400 Late_morning :336565 miércoles:238315
## Mean :2382 Early_afternoon:340978 jueves :236632
## 3rd Qu.:3960 Late_afternoon :385880 viernes :259729
## Max. :7280 Evening :214381 sábado :242043
## domingo :230892
## delay_reported
## Min. : 0.00
## 1st Qu.: 0.00
## Median : 1.00
## Mean : 16.76
## 3rd Qu.: 2.00
## Max. :2941.00
## Modelado
Estado del informe del crimen. Vemos una tabla con las opciones
table(crime_data$status_description, useNA = 'ifany')##
## Adult Arrest Adult Other Invest Cont Juv Arrest Juv Other
## 172666 190613 1302599 13246 4154
## UNK
## 31Creamos una variable para predecir, target
# Empezamos con todos los valores como 'Closed'
crime_data$target <- 'Closed'
# Y ahora los que crime_data$status_description=='Invest Cont' pasan a 'Open'
crime_data$target[crime_data$status_description=='Invest Cont'] <- 'Open'
# Una tabla
table(crime_data$target, useNA = 'ifany')##
## Closed Open
## 380710 1302599# Y pasamos a factor
crime_data$target <- as.factor(crime_data$target)Crear training y validating
# training hasta el año 2016
training <- crime_data[crime_data$date_occurred<as.Date('2017-01-01'), ]
# Validating el año 2017
validating <- crime_data[(crime_data$date_occurred>as.Date('2016-12-31') & crime_data$date_occurred<as.Date('2018-01-01')), ]
# testing el año 2018
testing <- crime_data[crime_data$date_occurred>as.Date('2017-12-31'), ]Iniciar H2O en local
# Iniciamos con todos los cores posibles y la memoria de 32G (si hay menos se
# autoregula)
h2o.init(nthreads=-1, max_mem_size="32G")## Connection successful!
##
## R is connected to the H2O cluster:
## H2O cluster uptime: 45 minutes 43 seconds
## H2O cluster version: 3.16.0.2
## H2O cluster version age: 3 months and 12 days !!!
## H2O cluster name: H2O_started_from_R_enero_ahz489
## H2O cluster total nodes: 1
## H2O cluster total memory: 6.35 GB
## H2O cluster total cores: 4
## H2O cluster allowed cores: 4
## H2O cluster healthy: TRUE
## H2O Connection ip: localhost
## H2O Connection port: 54321
## H2O Connection proxy: NA
## H2O Internal Security: FALSE
## H2O API Extensions: XGBoost, Algos, AutoML, Core V3, Core V4
## R Version: R version 3.4.3 (2017-11-30)## Warning in h2o.clusterInfo():
## Your H2O cluster version is too old (3 months and 12 days)!
## Please download and install the latest version from http://h2o.ai/download/# Si queremos conectar a un cluster que ya está funcionando
# h2o.init(ip="mycluster", port=55555) Pasamos los objetos a H2O y les damos nombre para poder identificarlos
training_h2o <- as.h2o(training,
destination_frame = "training")##
|
| | 0%
|
|=================================================================| 100%validating_h2o <- as.h2o(validating,
destination_frame = "validating")##
|
| | 0%
|
|=================================================================| 100%testing_h2o <- as.h2o(testing,
destination_frame = "testing")##
|
| | 0%
|
|=================================================================| 100%AutoML
Tras estos pasos en los que someramente hemos hecho ETL, tratamiento de NAs y algunas variables nuevas, pasamos a la parte de autoML
# La columna target
y <- "target"
# Las otras variables a usar, primero todas menos el target
x <- setdiff(names(training), y)
# Ahora quitamos también: 'dr_number', 'status_code', 'status_description'
x <- x[!(x %in% c('dr_number', 'status_code', 'status_description'))]Llamamos a automl. Solo le decimos los datos, la columna target, que hay un dataset de train y uno de validación y cuantos minutos de ejecución
tiempo_inicio <- Sys.time()
automl_models_h2o <- h2o.automl(
x = x,
y = y,
training_frame = training_h2o,
validation_frame = validating_h2o,
# leaderboard_frame = test_h2o,
max_runtime_secs = 3600, # 180
stopping_metric = "AUTO")##
|
| | 0%
|
|== | 3%
|
|=== | 5%
|
|==== | 5%
|
|==== | 6%
|
|==== | 7%
|
|===== | 7%
|
|===== | 8%
|
|====== | 8%
|
|====== | 9%
|
|====== | 10%
|
|======= | 10%
|
|======= | 11%
|
|======= | 12%
|
|======== | 12%
|
|======== | 13%
|
|========= | 13%
|
|========= | 14%
|
|========= | 15%
|
|========== | 15%
|
|============= | 20%
|
|============= | 21%
|
|============== | 21%
|
|=================================================================| 100%
##
|
| | 0%
|
|=================================================================| 100%print(Sys.time()-tiempo_inicio)## Time difference of 1.099437 hoursPodemos ver lo que hace H2O en: http://localhost:54321/flow/index.html
Creamos un objeto con el mejor modelo
automl_leader <- automl_models_h2o@leaderPredicción
# Pasamos el modelo y los datos nuevos
pred_testing <- as.data.frame(h2o.predict(automl_leader, newdata = testing_h2o))##
|
| | 0%
|
|=================================================================| 100%## Warning in doTryCatch(return(expr), name, parentenv, handler): Test/
## Validation dataset column 'crime_code_description' has levels not trained
## on: [238TH ST, AARON ST, ALSACE DR, ALVA, APPLETON WY, ARMACOST BL, ARTESIA
## BL, AVE OF STARS, BALI WY, BARBARA AV, BELOIT BL, BRIGHT, BROADLEAF AV,
## BROOKE, BUTTE ST, CALDUS AV, CAMINO DEL ORO ST, CARPENTER PL, CIMMARON BL,
## CLOVERDALE RD, CORINGA DR, CRESTKNOLL DR, DOBBS, DOHENY AV, DON LORENZO
## ST, DOREEN PL, E CESAR E CHAVEZ, E DENKER AV, E HILL ST, E KOHLER ST,
## E NEW HAMPSHIRE AV, E OXFORD AV, E SPRING ST, E SPROULE AV, EDGECLIFFE
## DR, ESCARPA DR, EUNICE, FAIRVIEW BL, FALKIRK LN, FORNEY, GENESSEE AV,
## GLADBECK, GRAND PARK, GROSVENOR BL, JACON WY, JARVIS, LUCIA PL, MARKLEIN
## AV, MATEO AV, MERCHANT ST, MERIT AV, METRO, MILKY WY, N 30TH ST, N 40TH
## PL, N BESSEMER ST, N POLK, PARMALEE AV, PASADENA ST, PIER, RAPHAEL, REAZA
## PL, REDMONT AV, REFORMA, RINCONIA DR, S 28TH ST, S CLINTON ST, S MULLEN AV,
## S PACIFIC AV, S SATICOY ST, S SAVANNAH AV, SAMPSON AV, SHERWOOD, SMILEY,
## STONE OAK DR, THE GROVE WY, TUPACHI ST, URBANA, VALIJEAN AV, VAN NUYS PL,
## VAN NUYS ST, VESTAL AV, W 74TH ST, W ANAHEIM ST, W GAGE ST, W ORANGE DR,
## WEDGEWOOD PL, WEYBURN PL, ZPHYR]## Warning in doTryCatch(return(expr), name, parentenv, handler): Test/
## Validation dataset column 'premise_description' has levels not trained on:
## [7TH AND METRO CENTER (NOT LINE SPECIFIC), BUS DEPOT/TERMINAL, OTHER THAN
## MTA, ELECTRONICS STORE (IE:RADIO SHACK, ETC.), FURNITURE STORE, LA UNION
## STATION (NOT LINE SPECIFIC), MTA - BLUE LINE - 103RD/WATTS TOWERS, MTA -
## BLUE LINE - 7TH AND METRO CENTER, MTA - BLUE LINE - GRAND/LATTC, MTA - BLUE
## LINE - PICO, MTA - BLUE LINE - SAN PEDRO, MTA - BLUE LINE - WASHINGTON,
## MTA - EXPO LINE - 7TH AND METRO CENTER, MTA - EXPO LINE - EXPO PARK/USC,
## MTA - EXPO LINE - EXPO/BUNDY, MTA - EXPO LINE - EXPO/CRENSHAW, MTA - EXPO
## LINE - EXPO/LA BREA, MTA - EXPO LINE - EXPO/SEPULVEDA, MTA - EXPO LINE -
## EXPO/VERMONT, MTA - EXPO LINE - EXPO/WESTERN, MTA - EXPO LINE - FARMDALE,
## MTA - EXPO LINE - JEFFERSON/USC, MTA - EXPO LINE - LA CIENEGA/JEFFERSON,
## MTA - EXPO LINE - LATTC/ORTHO INSTITUTE, MTA - EXPO LINE - PALMS, MTA -
## GOLD LINE - HIGHLAND PARK, MTA - GOLD LINE - MARIACHI PLAZA, MTA - GOLD
## LINE - SOUTHWEST MUSEUM, MTA - GOLD LINE - UNION STATION, MTA - GREEN LINE
## - AVALON, MTA - GREEN LINE - AVIATION/LAX, MTA - GREEN LINE - HARBOR FWY,
## MTA - PURPLE LINE - CIVIC CENTER/GRAND PARK, MTA - PURPLE LINE - PERSHING
## SQUARE, MTA - PURPLE LINE - WILSHIRE/VERMONT, MTA - RED LINE - 7TH AND
## METRO CENTER, MTA - RED LINE - CIVIC CENTER/GRAND PARK, MTA - RED LINE
## - HOLLYWOOD/HIGHLAND, MTA - RED LINE - HOLLYWOOD/VINE, MTA - RED LINE -
## HOLLYWOOD/WESTERN, MTA - RED LINE - NORTH HOLLYWOOD, MTA - RED LINE - UNION
## STATION, MTA - RED LINE - UNIVERSAL CITY/STUDIO CITY, MTA - RED LINE -
## VERMONT/BEVERLY, MTA - RED LINE - VERMONT/SANTA MONICA, MTA - RED LINE -
## VERMONT/SUNSET, MTA - RED LINE - WESTLAKE/MACARTHUR PARK, MTA - RED LINE
## - WILSHIRE/VERMONT, MTA - SILVER LINE - HARBOR FWY, MTA - SILVER LINE -
## HARBOR GATEWAY TRANSIT CTR, MTA - SILVER LINE - ROSECRANS, MUSCLE BEACH,
## SEX ORIENTED/BOOK STORE/STRIP CLUB/GENTLEMAN'S CLUB, SHORT-TERM VACATION
## RENTAL, TERMINAL, OTHER THAN MTA, TRAIN DEPOT/TERMINAL, OTHER THAN MTA,
## TRANSITIONAL HOUSING/HALFWAY HOUSE, VETERINARIAN/ANIMAL HOSPITAL]## Warning in doTryCatch(return(expr), name, parentenv, handler): Test/
## Validation dataset column 'address' has levels not trained on: [00 PORTOLA
## RD, 00000 LOYOLA MARYM DR, 04600 HUNTINGTON DR, 100 COPA DE ORO RD, 100
## HILGARD AV, 100 MESA ST, 1000 SPENCER ST, 1000 S BUDLONG, 1000 W 2ND ST,
## 10000 SANTA MONICA BL, 10100 GRANDEE AV, 10100 MOORPARK ST, 10100 S ALAMEDA
## ST, 10300 STATE ROUTE, 10500 CALIFAST AV, 10500 SELKIRK LN, 10600 JORDAN
## AV, 10600 LANARK ST, 10800 OSTEGO ST, 11000 COLUMBUS AV, 112TH PL, 11300
## BABBITT AV, 11300 S OLIVE ST, 11TH DR, 1200 LOS LOMAS AV, 1200 PRODUCE,
## 1200 TOWER RD, 12000 MARIBEL LN, 12000 SHERMAN RD, 12100 ANETA ST, 12200
## MILLENNIUM, 12300 5TH HELENA DR, 12300 CENTINELA AV, 12300 VICTORIA AV,
## 12400 HART ST, 12600 BLUFF CREEK DR, 12600 HUSTON ST, 12900 HOFFMAN ST,
## 13500 RIVERSIDE DR, 13500 SILVER PALMS LN, 13800 KAGEL CANYON ST, 1400
## BECKWITH AV, 1400 CALLE DEL JONELA, 1400 N BERENDO ST, 14200 CANTLAY ST,
## 14800 ASTORIA, 1500 EXPOSITION BL, 1500 W 249TH ST, 15700 HIGH KNOLL RD,
## 1600 E 3RD ST, 1600 S NEW ENGLAND ST, 16300 N SAN FERNANDO MISSION BL,
## 16400 MIRILLA ST, 16600 OLDHAM PL, 169TH ST, 1700 N NEW HAMPSHIRE, 1700 S
## BOYLE AV, 18200 CHATHAM LN, 18600 LEMAY ST, 189TH ST, 1900 PACIFIC COAST
## HY, 1900 RAPALLO PL, 19000 WINNETKA AV, 19300 LAVY CT, 19300 OLYMPIA
## ST, 19500 EMPIRE LN, 19800 CASTLEBAY LN, 19800 ITASCA ST, 200 ENERO CT,
## 200 LAKESHORE TR, 200 S INDIANA ST, 200 S SALTAIR AV, 200 W WORLD WY,
## 2000 TESLA AV, 20000 DAMPA LN, 20600 DUMONT ST, 2100 S BUDLONG AV, 2200
## BRENTFORD RD, 2200 MALAGA RD, 2200 RIPPLE ST, 22000 LINFIELD LN, 22TH,
## 2300 SKIRBALL CENTER DR, 23600 MARTHA ST, 2400 FERN DELL, 24000 SCHOENBORN
## ST, 2700 PATRICIA AV, 27700 HOLLYRIDGE DR, 300 E 38TH ST, 300 S AVENUE
## 16, 3100 BUDAN AV, 3200 MCCARTHY DR, 3200 WONDER VIEW DR, 3400 EXPOSITION
## BL, 3400 W MARTIN LUTHER KING BL, 400 S GRAND VIEW ST, 4000 FOLSOM ST,
## 4000 VIA MIRASOL, 4000 S CENTINELA AV, 4100 S GRAND AV, 4100 W 42ND PL,
## 4600 HUNTINGTON DR SOUTH, 4800 MT HELENA AV, 4800 SANDFORD DR, 4800 THOMAS
## ST, 500 S AVENUE 17, 5100 GALVEZ ST, 5100 LATHROP ST, 5300 DANNY AV, 5300
## DURATE ST, 5500 DUBOIS AV, 5700 ROLLING RD, 5800 DREXEL AV, 5900 W 83RD
## ST, 600 DUCOMMUN ST, 600 E L ST, 6700 BERQUIST AV, 6900 VANASCOY AV, 700
## ARMACOST AV, 700 SHORE, 700 VENICE WY, 7000 RIDGE AV, 7200 N MAPLE RD,
## 7400 CLIFFSIDE CT, 7700 WOODHALL AV, 800 APPLEBY, 800 MILWOOD CT, 800
## SOUTHERLAND AV, 800 TACUBA ST, 8000 PARK HILL DR, 8100 SHADY GLADE AV,
## 8300 BARHAM BL, 8300 BIG CANYON DR, 900 COOPER AV, 900 E 18TH ST, 900 E
## MARTIN LUTHER KING BL, 900 N STALEY LN, 900 S UNION AV, 9100 LARKE ELLEN
## CI, 9200 VENA AV, 93200 GRAHAM AV, 9400 GLORY AV, 9500 REGENT ST, ALTIVO
## WY, BARMAN AV, BIRD, BROADMOORE ST, BUENA VISTA DR, CAHUENGA BL, CALDUS
## AV, CASS, CENTURY PARK DR, CIVIC CENTER, COPPER, CRAFT AV, CRAVEN ST, DENNI
## AV, DOLPHIN AV, E 104TH ST, E AVENUE 28, E AVENUE 33, EARL ST, ELKLAND PL,
## FACTORY BL, GARFIELD ST, GAZETTE ST, GRAHAM PL, HAYES AV, HOSTETER FIRE
## RD, JAMES M WOODS BL, LA FAYETTE P, LONG BEACH FY, LONGFELLOW, MC CLURE
## AV, MIDWOOD DR, MONTECITO ST, MOUNT SHASTA, N 97TH ST, PEMBROKE, PERRIS ST,
## PETROLEUM, RAJAH, RIDGELEY ST, RINCONIA DR, ROXBURY DR, S MAPLE AV, S SAINT
## LOUIS ST, S WESTLAKE, SCARFF ST, STONEHAVEN, SUMMIT, SWEETZER DR, TRUMAN
## ST, VALLEY MEADOW, VANOWNEN ST, W 106TH ST, WEST WAY, WHITESIDE, WHITLEY
## PL]## Warning in doTryCatch(return(expr), name, parentenv, handler): Test/
## Validation dataset column 'cross_street' has levels not trained on: [238TH
## ST, AARON ST, ALSACE DR, ALVA, APPLETON WY, ARMACOST BL, ARTESIA BL, AVE OF
## STARS, BALI WY, BARBARA AV, BELOIT BL, BRIGHT, BROADLEAF AV, BROOKE, BUTTE
## ST, CALDUS AV, CAMINO DEL ORO ST, CARPENTER PL, CIMMARON BL, CLOVERDALE
## RD, CORINGA DR, CRESTKNOLL DR, DOBBS, DOHENY AV, DON LORENZO ST, DOREEN
## PL, E CESAR E CHAVEZ, E DENKER AV, E HILL ST, E KOHLER ST, E NEW HAMPSHIRE
## AV, E OXFORD AV, E SPRING ST, E SPROULE AV, EDGECLIFFE DR, ESCARPA DR,
## EUNICE, FAIRVIEW BL, FALKIRK LN, FORNEY, GENESSEE AV, GLADBECK, GRAND PARK,
## GROSVENOR BL, JACON WY, JARVIS, LUCIA PL, MARKLEIN AV, MATEO AV, MERCHANT
## ST, MERIT AV, METRO, MILKY WY, N 30TH ST, N 40TH PL, N BESSEMER ST, N POLK,
## PARMALEE AV, PASADENA ST, PIER, RAPHAEL, REAZA PL, REDMONT AV, REFORMA,
## RINCONIA DR, S 28TH ST, S CLINTON ST, S MULLEN AV, S PACIFIC AV, S SATICOY
## ST, S SAVANNAH AV, SAMPSON AV, SHERWOOD, SMILEY, STONE OAK DR, THE GROVE
## WY, TUPACHI ST, URBANA, VALIJEAN AV, VAN NUYS PL, VAN NUYS ST, VESTAL AV,
## W 74TH ST, W ANAHEIM ST, W GAGE ST, W ORANGE DR, WEDGEWOOD PL, WEYBURN PL,
## ZPHYR]## Warning in doTryCatch(return(expr), name, parentenv, handler): Test/
## Validation dataset column 'location' has levels not trained on:
## [(33.7315, -118.2971), (33.7372, -118.317), (33.7379, -118.3065),
## (33.7529, -118.2971), (33.7851, -118.2808), (33.7875, -118.2507),
## (33.7921, -118.2563), (33.7987, -118.3053), (33.8614, -118.285),
## (33.8698, -118.2849), (33.878, -118.2886), (33.933, -118.2402), (33.933,
## -118.2794), (33.937, -118.2792), (33.9389, -118.2328), (33.9454,
## -118.2283), (33.947, -118.243), (33.9555, -118.2749), (33.9751, -118.4211),
## (33.9766, -118.3946), (33.9768, -118.4634), (33.9773, -118.2916),
## (33.9795, -118.4082), (33.9799, -118.3703), (33.9818, -118.4683),
## (33.982, -118.4685), (33.984, -118.4017), (33.9853, -118.4092), (33.9869,
## -118.4728), (33.9881, -118.4146), (33.9893, -118.2438), (33.9943,
## -118.4578), (33.9945, -118.4615), (33.9996, -118.459), (34.0006,
## -118.4312), (34.0055, -118.4429), (34.0073, -118.4347), (34.0092,
## -118.259), (34.0209, -118.4373), (34.0222, -118.2394), (34.0249,
## -118.3551), (34.026, -118.3508), (34.0271, -118.2537), (34.0274,
## -118.3973), (34.0294, -118.4367), (34.0343, -118.2445), (34.0367,
## -118.2964), (34.0369, -118.473), (34.0371, -118.2963), (34.0384,
## -118.4572), (34.0392, -118.4141), (34.0419, -118.2877), (34.0421,
## -118.2112), (34.0434, -118.1929), (34.0438, -118.3342), (34.0477,
## -118.4631), (34.048, -118.3878), (34.0495, -118.2237), (34.0505,
## -118.2599), (34.0513, -118.2222), (34.0522, -118.2342), (34.0532,
## -118.4786), (34.0533, -118.4958), (34.0543, -118.2031), (34.0606,
## -118.3496), (34.0637, -118.2864), (34.064, -118.4141), (34.0675,
## -118.2214), (34.0717, -118.4628), (34.072, -118.2103), (34.0723,
## -118.3658), (34.078, -118.568), (34.0783, -118.2057), (34.0789, -118.4397),
## (34.079, -118.2969), (34.0898, -118.3244), (34.0915, -118.2458), (34.0917,
## -118.213), (34.0948, -118.2495), (34.1009, -118.283), (34.1042, -118.2647),
## (34.1053, -118.2504), (34.1109, -118.3079), (34.1112, -118.1868),
## (34.1112, -118.4483), (34.1113, -118.3451), (34.1121, -118.1887),
## (34.1167, -118.3256), (34.1198, -118.3179), (34.1211, -118.3193),
## (34.1218, -118.1744), (34.1218, -118.2239), (34.1226, -118.1738), (34.123,
## -118.4792), (34.1302, -118.21), (34.142, -118.2699), (34.1432, -118.4947),
## (34.1464, -118.1957), (34.1566, -118.3734), (34.1576, -118.4295),
## (34.1594, -118.4076), (34.1616, -118.5822), (34.1635, -118.6071),
## (34.1701, -118.6398), (34.1707, -118.5203), (34.1747, -118.6097),
## (34.1877, -118.5183), (34.1892, -118.4774), (34.191, -118.5386), (34.1912,
## -118.5244), (34.1922, -118.6346), (34.193, -118.39), (34.1975, -118.4705),
## (34.1975, -118.4726), (34.1992, -118.5873), (34.2012, -118.4043), (34.2018,
## -118.394), (34.2024, -118.4424), (34.2056, -118.65), (34.206, -118.6344),
## (34.2104, -118.6368), (34.2205, -118.6351), (34.228, -118.3693),
## (34.2294, -118.4454), (34.2322, -118.5815), (34.2364, -118.5361),
## (34.2393, -118.2798), (34.2418, -118.4338), (34.2429, -118.5675),
## (34.2447, -118.5657), (34.2468, -118.4512), (34.255, -118.5393),
## (34.2566, -118.4332), (34.2622, -118.6039), (34.2629, -118.2923),
## (34.2636, -118.6067), (34.2656, -118.553), (34.2674, -118.4664),
## (34.2697, -118.4027), (34.2722, -118.463), (34.2735, -118.5697), (34.2761,
## -118.4259), (34.2818, -118.5309), (34.2846, -118.3923), (34.2931,
## -118.5538), (34.3117, -118.4186), (34.3218, -118.4468), (34.3265,
## -118.4341)]## Warning in doTryCatch(return(expr), name, parentenv, handler): Test/
## Validation dataset column 'crime_year' has levels not trained on: [2018]Vemos el objeto de la predicción
str(pred_testing)## 'data.frame': 21187 obs. of 3 variables:
## $ predict: Factor w/ 2 levels "Closed","Open": 2 2 2 2 2 2 2 2 2 2 ...
## $ Closed : num 0.0531 0.0603 0.1482 0.0616 0.1985 ...
## $ Open : num 0.947 0.94 0.852 0.938 0.801 ...Un resumen de la performance en el objeto de test
h2o.performance(automl_leader, newdata = testing_h2o)## H2OBinomialMetrics: stackedensemble
##
## MSE: 0.1189653
## RMSE: 0.3449135
## LogLoss: 0.3745473
## Mean Per-Class Error: 0.4982028
## AUC: 0.7885211
## Gini: 0.5770422
##
## Confusion Matrix (vertical: actual; across: predicted) for F1-optimal threshold:
## Closed Open Error Rate
## Closed 8 2095 0.996196 =2095/2103
## Open 4 19080 0.000210 =4/19084
## Totals 12 21175 0.099070 =2099/21187
##
## Maximum Metrics: Maximum metrics at their respective thresholds
## metric threshold value idx
## 1 max f1 0.089551 0.947863 397
## 2 max f2 0.079476 0.978436 399
## 3 max f0point5 0.393782 0.923999 290
## 4 max accuracy 0.089551 0.900930 397
## 5 max precision 0.950663 1.000000 0
## 6 max recall 0.079476 1.000000 399
## 7 max specificity 0.950663 1.000000 0
## 8 max absolute_mcc 0.708349 0.297129 162
## 9 max min_per_class_accuracy 0.731609 0.727532 151
## 10 max mean_per_class_accuracy 0.758218 0.731465 137
##
## Gains/Lift Table: Extract with `h2o.gainsLift(<model>, <data>)` or `h2o.gainsLift(<model>, valid=<T/F>, xval=<T/F>)`as.data.frame(automl_models_h2o@leaderboard)## model_id auc logloss
## 1 StackedEnsemble_AllModels_0_AutoML_20180313_122540 NA NA
## 2 StackedEnsemble_BestOfFamily_0_AutoML_20180313_122540 NA NA
## 3 DRF_0_AutoML_20180313_122540 NA NA
## 4 XRT_0_AutoML_20180313_122540 NA NA
## 5 GBM_grid_0_AutoML_20180313_122540_model_0 NA NAY vemos el modelo
automl_leader## Model Details:
## ==============
##
## H2OBinomialModel: stackedensemble
## Model ID: StackedEnsemble_AllModels_0_AutoML_20180313_122540
## NULL
##
##
## H2OBinomialMetrics: stackedensemble
## ** Reported on training data. **
##
## MSE: 0.07489115
## RMSE: 0.2736625
## LogLoss: 0.2651635
## Mean Per-Class Error: 0.1527908
## AUC: 0.9428592
## Gini: 0.8857184
##
## Confusion Matrix (vertical: actual; across: predicted) for F1-optimal threshold:
## Closed Open Error Rate
## Closed 245185 86336 0.260424 =86336/331521
## Open 49794 1052876 0.045158 =49794/1102670
## Totals 294979 1139212 0.094918 =136130/1434191
##
## Maximum Metrics: Maximum metrics at their respective thresholds
## metric threshold value idx
## 1 max f1 0.556156 0.939279 210
## 2 max f2 0.301893 0.962042 296
## 3 max f0point5 0.756018 0.939519 137
## 4 max accuracy 0.583495 0.905195 201
## 5 max precision 0.952286 1.000000 0
## 6 max recall 0.065308 1.000000 394
## 7 max specificity 0.952286 1.000000 0
## 8 max absolute_mcc 0.626318 0.727934 188
## 9 max min_per_class_accuracy 0.805609 0.869694 115
## 10 max mean_per_class_accuracy 0.766066 0.871189 133
##
## Gains/Lift Table: Extract with `h2o.gainsLift(<model>, <data>)` or `h2o.gainsLift(<model>, valid=<T/F>, xval=<T/F>)`
## H2OBinomialMetrics: stackedensemble
## ** Reported on validation data. **
##
## MSE: 0.1183872
## RMSE: 0.3440744
## LogLoss: 0.3783586
## Mean Per-Class Error: 0.3445659
## AUC: 0.8457715
## Gini: 0.691543
##
## Confusion Matrix (vertical: actual; across: predicted) for F1-optimal threshold:
## Closed Open Error Rate
## Closed 16495 30591 0.649684 =30591/47086
## Open 7134 173711 0.039448 =7134/180845
## Totals 23629 204302 0.165511 =37725/227931
##
## Maximum Metrics: Maximum metrics at their respective thresholds
## metric threshold value idx
## 1 max f1 0.331162 0.902050 304
## 2 max f2 0.129093 0.951283 377
## 3 max f0point5 0.703089 0.894299 160
## 4 max accuracy 0.391017 0.835933 282
## 5 max precision 0.950596 0.998028 0
## 6 max recall 0.063875 1.000000 399
## 7 max specificity 0.950596 0.999958 0
## 8 max absolute_mcc 0.627734 0.473369 192
## 9 max min_per_class_accuracy 0.781789 0.767426 123
## 10 max mean_per_class_accuracy 0.802383 0.768738 113
##
## Gains/Lift Table: Extract with `h2o.gainsLift(<model>, <data>)` or `h2o.gainsLift(<model>, valid=<T/F>, xval=<T/F>)`
## H2OBinomialMetrics: stackedensemble
## ** Reported on cross-validation data. **
## ** 5-fold cross-validation on training data (Metrics computed for combined holdout predictions) **
##
## MSE: 0.1158442
## RMSE: 0.340359
## LogLoss: 0.3723338
## Mean Per-Class Error: 0.2803854
## AUC: 0.868436
## Gini: 0.7368721
##
## Confusion Matrix (vertical: actual; across: predicted) for F1-optimal threshold:
## Closed Open Error Rate
## Closed 165562 165959 0.500599 =165959/331521
## Open 66350 1036320 0.060172 =66350/1102670
## Totals 231912 1202279 0.161979 =232309/1434191
##
## Maximum Metrics: Maximum metrics at their respective thresholds
## metric threshold value idx
## 1 max f1 0.462545 0.899213 248
## 2 max f2 0.141606 0.945331 367
## 3 max f0point5 0.770436 0.897762 130
## 4 max accuracy 0.525549 0.839002 226
## 5 max precision 0.951479 0.995525 0
## 6 max recall 0.053853 1.000000 399
## 7 max specificity 0.951479 0.999937 0
## 8 max absolute_mcc 0.710315 0.536213 155
## 9 max min_per_class_accuracy 0.825580 0.791672 102
## 10 max mean_per_class_accuracy 0.830853 0.791855 99
##
## Gains/Lift Table: Extract with `h2o.gainsLift(<model>, <data>)` or `h2o.gainsLift(<model>, valid=<T/F>, xval=<T/F>)`Matriz de confusión sobre datos de testing con el paquete caret
confusion_matrix <- caret::confusionMatrix(pred_testing$predict, testing$target)
confusion_matrix## Confusion Matrix and Statistics
##
## Reference
## Prediction Closed Open
## Closed 592 1356
## Open 1511 17728
##
## Accuracy : 0.8647
## 95% CI : (0.86, 0.8693)
## No Information Rate : 0.9007
## P-Value [Acc > NIR] : 1.000000
##
## Kappa : 0.2176
## Mcnemar's Test P-Value : 0.004026
##
## Sensitivity : 0.28150
## Specificity : 0.92895
## Pos Pred Value : 0.30390
## Neg Pred Value : 0.92146
## Prevalence : 0.09926
## Detection Rate : 0.02794
## Detection Prevalence : 0.09194
## Balanced Accuracy : 0.60522
##
## 'Positive' Class : Closed
## Y por último, la información de la sesión
sessionInfo()## R version 3.4.3 (2017-11-30)
## Platform: x86_64-pc-linux-gnu (64-bit)
## Running under: Ubuntu 16.04.4 LTS
##
## Matrix products: default
## BLAS: /usr/lib/openblas-base/libblas.so.3
## LAPACK: /usr/lib/libopenblasp-r0.2.18.so
##
## locale:
## [1] LC_CTYPE=es_ES.UTF-8 LC_NUMERIC=C
## [3] LC_TIME=es_ES.UTF-8 LC_COLLATE=es_ES.UTF-8
## [5] LC_MONETARY=es_ES.UTF-8 LC_MESSAGES=es_ES.UTF-8
## [7] LC_PAPER=es_ES.UTF-8 LC_NAME=C
## [9] LC_ADDRESS=C LC_TELEPHONE=C
## [11] LC_MEASUREMENT=es_ES.UTF-8 LC_IDENTIFICATION=C
##
## attached base packages:
## [1] stats graphics grDevices utils datasets methods base
##
## other attached packages:
## [1] h2o_3.16.0.2 pROC_1.10.0 chron_2.3-52
## [4] lubridate_1.7.3 ggplot2_2.2.1.9000 dplyr_0.7.4
##
## loaded via a namespace (and not attached):
## [1] Rcpp_0.12.15 lattice_0.20-35 tidyr_0.8.0
## [4] class_7.3-14 assertthat_0.2.0 rprojroot_1.3-2
## [7] digest_0.6.15 ipred_0.9-6 psych_1.7.8
## [10] foreach_1.4.4 R6_2.2.2 plyr_1.8.4
## [13] backports_1.1.2 stats4_3.4.3 e1071_1.6-8
## [16] evaluate_0.10.1 pillar_1.2.1 rlang_0.2.0.9000
## [19] lazyeval_0.2.1 caret_6.0-78 kernlab_0.9-25
## [22] rpart_4.1-13 Matrix_1.2-12 rmarkdown_1.9
## [25] splines_3.4.3 CVST_0.2-1 ddalpha_1.3.1.1
## [28] gower_0.1.2 stringr_1.3.0 foreign_0.8-69
## [31] RCurl_1.95-4.10 munsell_0.4.3 broom_0.4.3
## [34] compiler_3.4.3 pkgconfig_2.0.1 mnormt_1.5-5
## [37] dimRed_0.1.0 htmltools_0.3.6 tidyselect_0.2.4
## [40] nnet_7.3-12 prodlim_1.6.1 tibble_1.4.2
## [43] DRR_0.0.3 codetools_0.2-15 RcppRoll_0.2.2
## [46] withr_2.1.1.9000 MASS_7.3-49 bitops_1.0-6
## [49] recipes_0.1.2 ModelMetrics_1.1.0 grid_3.4.3
## [52] nlme_3.1-131 jsonlite_1.5 gtable_0.2.0
## [55] magrittr_1.5 scales_0.5.0.9000 stringi_1.1.7
## [58] reshape2_1.4.3 bindrcpp_0.2 timeDate_3043.102
## [61] robustbase_0.92-8 lava_1.6 iterators_1.0.9
## [64] tools_3.4.3 glue_1.2.0 DEoptimR_1.0-8
## [67] purrr_0.2.4 sfsmisc_1.1-2 survival_2.41-3
## [70] parallel_3.4.3 yaml_2.1.18 colorspace_1.3-2
## [73] knitr_1.20 bindr_0.1