.Rapp.history

0.1*0.5+0.9*0.5
2*0.5*0.1
0.1*0.6+0.9*0.4
(100/58)*0.4*0.1
(100/58)*0.6*0.9
require('EvCombR')
mass_1 = c(rep(0.8, 9), 0.2, 0.2)
mass_1
mass_0 = c(rep(0.2, 9), 0.8, 0.8)
stateSpace <- c("0", "1")#
    masses = vector("list", length(test[1,])-1)#
    #Getting mass values for each feature of the example#
    for(j in 1:(length(test[1,])-1)){#
      masses[[j]] <- mass(list("0"=mass_0[1,j], "1"=mass_1[1,j], stateSpace) #
    }#
    #Using the modified Dempster combination to combine the prior distribution (the logistic probs)#
    #with the mass values from the features #
    mass_combo = dComb(masses)
stateSpace <- c("0", "1")#
    masses = vector("list", 11)#
    #Getting mass values for each feature of the example#
    for(j in 1:(length(test[1,])-1)){#
      masses[[j]] <- mass(list("0"=mass_0[1,j], "1"=mass_1[1,j], stateSpace) #
    }#
    #Using the modified Dempster combination to combine the prior distribution (the logistic probs)#
    #with the mass values from the features #
    mass_combo = dComb(masses)
mass_0
stateSpace <- c("0", "1")#
    masses = vector("list", 11)#
    #Getting mass values for each feature of the example#
    for(j in 1:(length(test[1,])-1)) {#
      masses[[j]] <- mass(list("0"=mass_0[1,j], "1"=mass_1[1,j], stateSpace)) #
    }#
    #Using the modified Dempster combination to combine the prior distribution (the logistic probs)#
    #with the mass values from the features #
    mass_combo = dComb(masses)
stateSpace <- c("0", "1")#
    masses = vector("list", 11)#
    #Getting mass values for each feature of the example#
    for(j in 1:11) {#
      masses[[j]] <- mass(list("0"=mass_0[1,j], "1"=mass_1[1,j], stateSpace)) #
    }#
    #Using the modified Dempster combination to combine the prior distribution (the logistic probs)#
    #with the mass values from the features #
    mass_combo = dComb(masses)
stateSpace <- c("0", "1")#
    masses = vector("list", 11)#
    #Getting mass values for each feature of the example#
    for(j in 1:11) {#
      masses[[j]] <- mass(list("0"=mass_0[j], "1"=mass_1[j], stateSpace)) #
    }#
    #Using the modified Dempster combination to combine the prior distribution (the logistic probs)#
    #with the mass values from the features #
    mass_combo = dComb(masses)
length(mass_0)
length(mass_1)
masses
mass_0[1]
stateSpace <- c("0", "1")#
    masses = vector("list", 11)#
    #Getting mass values for each feature of the example#
    for(j in 1:11) {#
      masses[[j]] <- mass(list("0"=mass_0[j], "1"=mass_1[j]), stateSpace) #
    }#
    #Using the modified Dempster combination to combine the prior distribution (the logistic probs)#
    #with the mass values from the features #
    mass_combo = dComb(masses)
mass_combo
prior_dist = mass(list("0"=0.2, "1"=0.8), stateSpace)#
	stateSpace <- c("0", "1")#
    masses = vector("list", 11)#
    #Getting mass values for each feature of the example#
    for(j in 1:11) {#
      masses[[j]] <- mass(list("0"=mass_0[j], "1"=mass_1[j]), stateSpace) #
    }#
    #Using the modified Dempster combination to combine the prior distribution (the logistic probs)#
    #with the mass values from the features #
    mass_combo = mComb(masses, prior_dist)
prior_dist
prior_dist@focal$0
prior_dist@focal$'0'
mass_combo = mComb(masses, list("0"=prior_dist@focal$'0', "1"=rior_dist@focal$'1'))
prior_dist = mass(list("0"=0.2, "1"=0.8), stateSpace)#
	stateSpace <- c("0", "1")#
    masses = vector("list", 11)#
    #Getting mass values for each feature of the example#
    for(j in 1:11) {#
      masses[[j]] <- mass(list("0"=mass_0[j], "1"=mass_1[j]), stateSpace) #
    }#
    #Using the modified Dempster combination to combine the prior distribution (the logistic probs)#
    #with the mass values from the features #
    mass_combo = mComb(masses, list("0"=prior_dist@focal$'0', "1"=prior_dist@focal$'1'))
mass_combo
mass_0
prior_dist = mass(list("0"=0.2, "1"=0.8), stateSpace)#
	stateSpace <- c("0", "1")#
    masses = vector("list", 11)#
    #Getting mass values for each feature of the example#
    for(j in 1:11) {#
      masses[[j]] <- mass(list("0"=mass_0[j], "1"=mass_1[j]), stateSpace) #
    }#
    #Using the modified Dempster combination to combine the prior distribution (the logistic probs)#
    #with the mass values from the features #
    mass_combo = mComb(masses, list("0"=prior_dist@focal$'1', "1"=prior_dist@focal$'0'))
mass_combo
prior_dist@focal$'1'
prior_dist@focal$'0'
masses
mass_combo_0 = mComb(masses, list("0"=0.2, "1"=0.8)) #
    mass_combo_1 = dComb(masses)
mass_combo_0
mass_combo_1
mass_combo_0 = mComb(masses, list("0"=0.8, "1"=0.2)) #
    mass_combo_1 = dComb(masses)
mass_combo_0
# state space#
stateSpace <- c("a", "b", "c")#
# mass functions#
m1 <- mass(list("a"=0.1, "a/b/c"=0.9), stateSpace)#
m2 <- mass(list("a"=0.2, "a/b/c"=0.8), stateSpace)
mComb(m1, m2, list("a"=0.1, "b"=0.1, "c"=0.8))
stateSpace <- c("0", "1")#
    masses = vector("list", 11)#
    #Getting mass values for each feature of the example#
    for(j in 1:11) {#
      masses[[j]] <- mass(list("0"=mass_0[j], "1"=mass_1[j]), stateSpace) #
    }#
    #Using the modified Dempster combination to combine the prior distribution (the logistic probs)#
    #with the mass values from the features #
    #mass_combo_0 = mComb(masses, list("0"=0.2, "1"=0.8)) #
    mass_combo_1 = dComb(masses)
mass_combo_1
mass_0 = c(0.9, 0.1, 0.9, 0.1)
mass_1 = c(0.1, 0.9, 0.1, 0.9)
stateSpace <- c("0", "1")#
    masses = vector("list", 11)#
    #Getting mass values for each feature of the example#
    for(j in 1:11) {#
      masses[[j]] <- mass(list("0"=mass_0[j], "1"=mass_1[j]), stateSpace) #
    }#
    #Using the modified Dempster combination to combine the prior distribution (the logistic probs)#
    #with the mass values from the features #
    #mass_combo_0 = mComb(masses, list("0"=0.2, "1"=0.8)) #
    mass_combo_1 = dComb(masses)
#prior_dist = mass(list("0"=0.2, "1"=0.8), stateSpace)#
	stateSpace <- c("0", "1")#
    masses = vector("list", 4)#
    #Getting mass values for each feature of the example#
    for(j in 1:4) {#
      masses[[j]] <- mass(list("0"=mass_0[j], "1"=mass_1[j]), stateSpace) #
    }#
    #Using the modified Dempster combination to combine the prior distribution (the logistic probs)#
    #with the mass values from the features #
    #mass_combo_0 = mComb(masses, list("0"=0.2, "1"=0.8)) #
    mass_combo_1 = dComb(masses)
mass_combo_1
install.packages('rmarkdown')
#time domain
#
library(fpp)
#
library(astsa)
#
library(forecast)
#
str(ausbeer)
#
plot(ausbeer)
#
#the mean is not constant, the first diff would be useful
#
plot(diff(ausbeer))
#
#the autovariance of the series looks like it depends on time, a log might help
#
plot(diff(log(ausbeer)))
#
#the plot of that looks better, I'll try some modeling now
#
acf2(diff(log(ausbeer)))
#
#definitely a seasonal component, at least one every two quarters
#
#Since the seasonal compenent tails off, it probably involves an AR(P)
#
#the speed (slow) of this decay suggests a seasonal difference
#
auto.arima(log(ausbeer))
#
#arima(1,1,2)(0,1,1)[4] interesting
#
#I'll start with trying a arima(0,1,1)(0,1,1) since the inbetween seasonal parts
#
#of the PACF are tailing off, a MA(q) must be involved
#
sarima(log(ausbeer), 0,1,1, 0,1,1, 4)
#
#ACF still looks like more than white noise, lets take a closer look
#
#The ACF is cutting off after the first lag so let's try MA(2)
#
sarima(log(ausbeer), 0,1,2, 0,1,1, 4)
#
#the model looks great...normaility, good Ljung-Box stat, ACF looks good too
#
#I'll take a closer look at the ACF and PACF to make sure it's a good fit
#
good_model = sarima(log(ausbeer), 0,1,2, 0,1,1, 4)
#
acf2(good_model$fit$residuals)
#
#looks like all the information was captured by the model, just white noise left
#
#Time to do a little forecasting...say, for the next year (2008 Q4 to 2009 Q3)
#
par(mfrow=c(1,1))
#
#doing the forecast and the graph of the time series w/ forecasted values
#
fore_cast = sarima.for(log(ausbeer), 4, 0,1,2, 0,1,1, 4)
#
#transforming the predicted values from logged values
#
exp(fore_cast$pred)
#
#prediction intervals
#
U = exp(fore_cast$pred + fore_cast$se) 
#
L = exp(fore_cast$pred - fore_cast$se)
#
for(i in 1:4){
#
  print(c(L[i],U[i])) 
#
}#
#frequency domain
#
par(mfrow=c(1,1))
#
library(itsmr)
#
#have to make the series stationary to do frequency analysis
#
ausbeer_sta = diff(log(ausbeer))
#
#Creating the periodogram, which is a estimator of spectral density
#
ausbeer.per = spec.pgram(ausbeer_sta, taper=0, log='no')
#
which(ausbeer.per$spec == max(ausbeer.per$spec))
#
#length that spec.pgram used
#
nextn(length(ausbeer))
#
#frequency is multiples of 1/4, remember only goes up to folding freq so double
#
#multiply the max obs by two since the spec.pgram only goes up 
#
#to the folding freq
#
#peaks at 1*(1/4) and 2*(1/4)
#
#one cycle every four months and another every two months
#
abline(v=1, lty="dotted")
#
which(ausbeer.per$spec == max(ausbeer.per$spec[-(53:55)]))
#
abline(v=2, lty="dotted")
#
U = qchisq(0.025, 2)
#
L = qchisq(0.975, 2)
#
c(2*ausbeer.per$spec[54]/L, 2*ausbeer.per$spec[54]/U)
#
c(2*ausbeer.per$spec[108]/L, 2*ausbeer.per$spec[108]/U)
#
abline(h=2*ausbeer.per$spec[54]/L, lty="dotted", col='blue')#higher line
#
abline(h=2*ausbeer.per$spec[108]/L, lty="dotted", col='red')#lower line
#
#ausbeer pgram at freq 1/4 = 54/216 and 2/4 = 108/216
#
#looks like both dominant frequencies are significant
install.packages('fpp')
install.packages('astsa')
install.packages('forecast')
install.packages('itsmr')
#time domain
#
library(fpp)
#
library(astsa)
#
library(forecast)
#
str(ausbeer)
#
plot(ausbeer)
#
#the mean is not constant, the first diff would be useful
#
plot(diff(ausbeer))
#
#the autovariance of the series looks like it depends on time, a log might help
#
plot(diff(log(ausbeer)))
#
#the plot of that looks better, I'll try some modeling now
#
acf2(diff(log(ausbeer)))
#
#definitely a seasonal component, at least one every two quarters
#
#Since the seasonal compenent tails off, it probably involves an AR(P)
#
#the speed (slow) of this decay suggests a seasonal difference
#
auto.arima(log(ausbeer))
#
#arima(1,1,2)(0,1,1)[4] interesting
#
#I'll start with trying a arima(0,1,1)(0,1,1) since the inbetween seasonal parts
#
#of the PACF are tailing off, a MA(q) must be involved
#
sarima(log(ausbeer), 0,1,1, 0,1,1, 4)
#
#ACF still looks like more than white noise, lets take a closer look
#
#The ACF is cutting off after the first lag so let's try MA(2)
#
sarima(log(ausbeer), 0,1,2, 0,1,1, 4)
#
#the model looks great...normaility, good Ljung-Box stat, ACF looks good too
#
#I'll take a closer look at the ACF and PACF to make sure it's a good fit
#
good_model = sarima(log(ausbeer), 0,1,2, 0,1,1, 4)
#
acf2(good_model$fit$residuals)
#
#looks like all the information was captured by the model, just white noise left
#
#Time to do a little forecasting...say, for the next year (2008 Q4 to 2009 Q3)
#
par(mfrow=c(1,1))
#
#doing the forecast and the graph of the time series w/ forecasted values
#
fore_cast = sarima.for(log(ausbeer), 4, 0,1,2, 0,1,1, 4)
#
#transforming the predicted values from logged values
#
exp(fore_cast$pred)
#
#prediction intervals
#
U = exp(fore_cast$pred + fore_cast$se) 
#
L = exp(fore_cast$pred - fore_cast$se)
#
for(i in 1:4){
#
  print(c(L[i],U[i])) 
#
}#
#frequency domain
#
par(mfrow=c(1,1))
#
library(itsmr)
#
#have to make the series stationary to do frequency analysis
#
ausbeer_sta = diff(log(ausbeer))
#
#Creating the periodogram, which is a estimator of spectral density
#
ausbeer.per = spec.pgram(ausbeer_sta, taper=0, log='no')
#
which(ausbeer.per$spec == max(ausbeer.per$spec))
#
#length that spec.pgram used
#
nextn(length(ausbeer))
#
#frequency is multiples of 1/4, remember only goes up to folding freq so double
#
#multiply the max obs by two since the spec.pgram only goes up 
#
#to the folding freq
#
#peaks at 1*(1/4) and 2*(1/4)
#
#one cycle every four months and another every two months
#
abline(v=1, lty="dotted")
#
which(ausbeer.per$spec == max(ausbeer.per$spec[-(53:55)]))
#
abline(v=2, lty="dotted")
#
U = qchisq(0.025, 2)
#
L = qchisq(0.975, 2)
#
c(2*ausbeer.per$spec[54]/L, 2*ausbeer.per$spec[54]/U)
#
c(2*ausbeer.per$spec[108]/L, 2*ausbeer.per$spec[108]/U)
#
abline(h=2*ausbeer.per$spec[54]/L, lty="dotted", col='blue')#higher line
#
abline(h=2*ausbeer.per$spec[108]/L, lty="dotted", col='red')#lower line
#
#ausbeer pgram at freq 1/4 = 54/216 and 2/4 = 108/216
#
#looks like both dominant frequencies are significant
#time domain
#
library(fpp)
#
library(astsa)
#
library(forecast)
#
str(ausbeer)
#
plot(ausbeer)
#
#the mean is not constant, the first diff would be useful
#
plot(diff(ausbeer))
#
#the autovariance of the series looks like it depends on time, a log might help
#
plot(diff(log(ausbeer)))
#
#the plot of that looks better, I'll try some modeling now
#
acf2(diff(log(ausbeer)))
#
#definitely a seasonal component, at least one every two quarters
#
#Since the seasonal compenent tails off, it probably involves an AR(P)
#
#the speed (slow) of this decay suggests a seasonal difference
#
auto.arima(log(ausbeer))
#
#arima(1,1,2)(0,1,1)[4] interesting
#
#I'll start with trying a arima(0,1,1)(0,1,1) since the inbetween seasonal parts
#
#of the PACF are tailing off, a MA(q) must be involved
#
sarima(log(ausbeer), 0,1,1, 0,1,1, 4)
#
#ACF still looks like more than white noise, lets take a closer look
#
#The ACF is cutting off after the first lag so let's try MA(2)
#
sarima(log(ausbeer), 0,1,2, 0,1,1, 4)
#
#the model looks great...normaility, good Ljung-Box stat, ACF looks good too
#
#I'll take a closer look at the ACF and PACF to make sure it's a good fit
#
good_model = sarima(log(ausbeer), 0,1,2, 0,1,1, 4)
#
acf2(good_model$fit$residuals)
#
#looks like all the information was captured by the model, just white noise left
#
#Time to do a little forecasting...say, for the next year (2008 Q4 to 2009 Q3)
#
par(mfrow=c(1,1))
#
#doing the forecast and the graph of the time series w/ forecasted values
#
fore_cast = sarima.for(log(ausbeer), 4, 0,1,2, 0,1,1, 4)
#
#transforming the predicted values from logged values
#
exp(fore_cast$pred)
#
#prediction intervals
#
U = exp(fore_cast$pred + fore_cast$se) 
#
L = exp(fore_cast$pred - fore_cast$se)
#
for(i in 1:4){
#
  print(c(L[i],U[i])) 
#
}#
#frequency domain
#
par(mfrow=c(1,1))
#
library(itsmr)
#
#have to make the series stationary to do frequency analysis
#
ausbeer_sta = diff(log(ausbeer))
#
#Creating the periodogram, which is a estimator of spectral density
#
ausbeer.per = spec.pgram(ausbeer_sta, taper=0, log='no')
#
which(ausbeer.per$spec == max(ausbeer.per$spec))
#
#length that spec.pgram used
#
nextn(length(ausbeer))
#
#frequency is multiples of 1/4, remember only goes up to folding freq so double
#
#multiply the max obs by two since the spec.pgram only goes up 
#
#to the folding freq
#
#peaks at 1*(1/4) and 2*(1/4)
#
#one cycle every four months and another every two months
#
abline(v=1, lty="dotted")
#
which(ausbeer.per$spec == max(ausbeer.per$spec[-(53:55)]))
#
abline(v=2, lty="dotted")
#time domain
library(fpp)
library(astsa)
library(forecast)
str(ausbeer)
plot(ausbeer)
#the mean is not constant, the first diff would be useful
plot(diff(ausbeer))
#the autovariance of the series looks like it depends on time, a log might help
plot(diff(log(ausbeer)))
#the plot of that looks better, I'll try some modeling now
acf2(diff(log(ausbeer)))
#definitely a seasonal component, at least one every two quarters
#Since the seasonal compenent tails off, it probably involves an AR(P)
#the speed (slow) of this decay suggests a seasonal difference
auto.arima(log(ausbeer))
#arima(1,1,2)(0,1,1)[4] interesting
#I'll start with trying a arima(0,1,1)(0,1,1) since the inbetween seasonal parts
#of the PACF are tailing off, a MA(q) must be involved
sarima(log(ausbeer), 0,1,1, 0,1,1, 4)
#ACF still looks like more than white noise, lets take a closer look
#The ACF is cutting off after the first lag so let's try MA(2)
sarima(log(ausbeer), 0,1,2, 0,1,1, 4)
#the model looks great...normality, good Ljung-Box stat, ACF looks good too
#I'll take a closer look at the ACF and PACF to make sure it's a good fit
good_model = sarima(log(ausbeer), 0,1,2, 0,1,1, 4)
acf2(good_model$fit$residuals)
#looks like all the information was captured by the model, just white noise left
#Time to do a little forecasting...say, for the next year (2008 Q4 to 2009 Q3)
par(mfrow=c(1,1))
#doing the forecast and the graph of the time series w/ forecasted values
fore_cast = sarima.for(log(ausbeer), 4, 0,1,2, 0,1,1, 4)
#transforming the predicted values from logged values
exp(fore_cast$pred)
#prediction intervals
U = exp(fore_cast$pred + fore_cast$se) 
L = exp(fore_cast$pred - fore_cast$se)
for(i in 1:4){
  print(c(L[i],U[i])) 
}

#frequency domain
par(mfrow=c(1,1))
library(itsmr)
#have to make the series stationary to do frequency analysis
ausbeer_sta = diff(log(ausbeer))
#Creating the periodogram, which is an estimator of spectral density
ausbeer.per = spec.pgram(ausbeer_sta, taper=0, log='no')
which(ausbeer.per$spec == max(ausbeer.per$spec))
#length that spec.pgram used
nextn(length(ausbeer))
#frequency is multiples of 1/4, remember only goes up to folding 
#peaks at 1*(1/4) and 2*(1/4)
#one cycle every four months and another every two months
abline(v=1, lty="dotted")
which(ausbeer.per$spec == max(ausbeer.per$spec[-(53:55)]))
abline(v=2, lty="dotted")
U = qchisq(0.025, 2)
#
L = qchisq(0.975, 2)
#
c(2*ausbeer.per$spec[54]/L, 2*ausbeer.per$spec[54]/U)
#
c(2*ausbeer.per$spec[108]/L, 2*ausbeer.per$spec[108]/U)
#
abline(h=2*ausbeer.per$spec[54]/L, lty="dotted", col='blue')#higher line
#
abline(h=2*ausbeer.per$spec[108]/L, lty="dotted", col='red')#lower line
#
#ausbeer pgram at freq 1/4 = 54/216 and 2/4 = 108/216
#
#looks like both dominant frequencies are significant
ls()
getwd()
setwd('/Users/nicholaslipanovich/Documents/GitHub/MachineLearnCourseProjects')
setwd('/Users/nicholaslipanovich/Documents/GitHub/MachineLearningCourseProjects')
j = read.csv('1987')
j = read.csv('1987.csv')
length(j[,1])
h = j[800000,]
?write.csv
write.csv(h, '1987.csv')
j = read.csv('1987.csv')
length(j[,1])
j = read.csv('1987.csv')
h = j[1:800000,]
length(h[,1])
length(h[1,])
write.csv(h, '1987.csv')