-
Notifications
You must be signed in to change notification settings - Fork 318
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
- Loading branch information
M4Competition
committed
Jun 11, 2018
1 parent
54401c7
commit d5d754c
Showing
164 changed files
with
12,187 additions
and
12,187 deletions.
There are no files selected for viewing
File renamed without changes.
File renamed without changes.
File renamed without changes.
50 changes: 25 additions & 25 deletions
50
036/M4-FPIS-create-datasets-20180523.R → ...otpetr/M4-FPIS-create-datasets-20180523.R
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -1,26 +1,26 @@ | ||
| filenames = c("Yearly", "Quarterly", "Monthly", "Weekly", "Daily", "Hourly") | ||
| horizons = c(6, 8, 18, 13, 14, 48) | ||
| freqs = c(1, 4, 12, 52, 7, 168) | ||
|
|
||
| for (d in 1:6){ | ||
| conn <- file(paste0(filenames[d], "-train.csv"),open="r") | ||
| linn <- readLines(conn) | ||
|
|
||
| series = NULL | ||
| for (i in 2:length(linn)){ | ||
| print(paste(d, i)) | ||
|
|
||
| aline = linn[i] | ||
| aline = strsplit(aline, ",")[[1]] | ||
| n = min(as.numeric(min(which(aline == "")) - 2), length(aline) - 1) | ||
| x = array(NA, n) | ||
| for (i in 1:n){ | ||
| x[i] = as.numeric(substr(aline[i+1], 2, nchar(aline[i+1]) - 1)) | ||
| } | ||
|
|
||
| series = append(series, list(list(n=n, h=horizons[d], frequency=freqs[d], x=ts(x, frequency=freqs[d])))) | ||
| } | ||
|
|
||
| close(conn) | ||
| save(series, file=paste0(filenames[d], ".RData")) | ||
| filenames = c("Yearly", "Quarterly", "Monthly", "Weekly", "Daily", "Hourly") | ||
| horizons = c(6, 8, 18, 13, 14, 48) | ||
| freqs = c(1, 4, 12, 52, 7, 168) | ||
|
|
||
| for (d in 1:6){ | ||
| conn <- file(paste0(filenames[d], "-train.csv"),open="r") | ||
| linn <- readLines(conn) | ||
|
|
||
| series = NULL | ||
| for (i in 2:length(linn)){ | ||
| print(paste(d, i)) | ||
|
|
||
| aline = linn[i] | ||
| aline = strsplit(aline, ",")[[1]] | ||
| n = min(as.numeric(min(which(aline == "")) - 2), length(aline) - 1) | ||
| x = array(NA, n) | ||
| for (i in 1:n){ | ||
| x[i] = as.numeric(substr(aline[i+1], 2, nchar(aline[i+1]) - 1)) | ||
| } | ||
|
|
||
| series = append(series, list(list(n=n, h=horizons[d], frequency=freqs[d], x=ts(x, frequency=freqs[d])))) | ||
| } | ||
|
|
||
| close(conn) | ||
| save(series, file=paste0(filenames[d], ".RData")) | ||
| } |
284 changes: 142 additions & 142 deletions
284
036/M4-FPIS-submission-code-20180525.R → ...otpetr/M4-FPIS-submission-code-20180525.R
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -1,143 +1,143 @@ | ||
| library(forecast) | ||
| library(smooth) | ||
| library(forecTheta) | ||
| library(foreach) | ||
| library(doSNOW) | ||
| cl = registerDoSNOW(makeCluster(4, type = "SOCK")) | ||
|
|
||
| # Datasets sizes | ||
| tsns = c(23000, 24000, 48000, 359, 4227, 414) | ||
|
|
||
| # *** SELECT WHICH SERIES *** | ||
|
|
||
| # The below will produce forecasts for the first 10 series of each frequency | ||
| ids = array(NA, c(2, 6*12)) | ||
| ids[1,] = rep(1:6, each=12) | ||
| ids[2,] = rep(12:1, 6) | ||
|
|
||
| # # The below will produce forecasts for the 13th yearly series, the 45th and the 10001th quarterly, the 30102th monthly and the 11th weekly | ||
| # ids = array(NA, c(2, 5)) | ||
| # ids[1,] = c(1,2,2,3,4) | ||
| # ids[2,] = c(13,45,10001,30102,11) | ||
|
|
||
| # # The below will produce forecasts for all series | ||
| # ids = array(NA, c(2,100000)) | ||
| # ids[1,] = c(rep(1,tsns[1]),rep(2,tsns[2]),rep(3,tsns[3]),rep(4,tsns[4]),rep(5,tsns[5]),rep(6,tsns[6])) | ||
| # ids[2,] = c(1:tsns[1],1:tsns[2],1:tsns[3],1:tsns[4],1:tsns[5],1:tsns[6]) | ||
|
|
||
| # # The below will produce forecasts for Y and Q series | ||
| # ids = array(NA, c(2, sum(tsns[1:2]))) | ||
| # ids[1,] = c(rep(1, tsns[1]), rep(2, tsns[2])) | ||
| # ids[2,] = c(1:tsns[1], 1:tsns[2]) | ||
|
|
||
| # # The below will produce forecasts for M series | ||
| # ids = array(NA, c(2, sum(tsns[3]))) | ||
| # ids[1,] = c(rep(3, tsns[3])) | ||
| # ids[2,] = c(1:tsns[3]) | ||
|
|
||
| # # The below will produce forecasts for W, D, and H series | ||
| # ids = array(NA, c(2, sum(tsns[4:6]))) | ||
| # ids[1,] = c(rep(4, tsns[4]), rep(5, tsns[5]), rep(6, tsns[6])) | ||
| # ids[2,] = c(1:tsns[4], 1:tsns[5], 1:tsns[6]) | ||
|
|
||
| # *** END OF SELECT WHICH SERIES *** | ||
|
|
||
| # Define structures for all forecasts and prediction intervals | ||
| datasets = sort(unique(ids[1,])) | ||
| the_names = list(c(if (any(datasets == 1)){paste0("Y", sort((ids[2,][ids[1,]==1])))}, | ||
| if (any(datasets == 2)){paste0("Q", sort((ids[2,][ids[1,]==2])))}, | ||
| if (any(datasets == 3)){paste0("M", sort((ids[2,][ids[1,]==3])))}, | ||
| if (any(datasets == 4)){paste0("W", sort((ids[2,][ids[1,]==4])))}, | ||
| if (any(datasets == 5)){paste0("D", sort((ids[2,][ids[1,]==5])))}, | ||
| if (any(datasets == 6)){paste0("H", sort((ids[2,][ids[1,]==6])))}), | ||
| paste0("F", 1:48)) | ||
| all_ffcs = array(NA, c(dim(ids)[2], 48), dimnames = the_names) | ||
| all_lower = array(NA, c(dim(ids)[2], 48), dimnames = the_names) | ||
| all_upper = array(NA, c(dim(ids)[2], 48), dimnames = the_names) | ||
|
|
||
| for (d in datasets){ | ||
|
|
||
| # Read dataset (produced by the separate R script) | ||
| if (d==1){ | ||
| frequency = 1; horizon = 6; load("Yearly.RData") | ||
| } else if (d==2){ | ||
| frequency = 4; horizon = 8; load("Quarterly.RData") | ||
| } else if (d==3){ | ||
| frequency = 12; horizon = 18; load("Monthly.RData") | ||
| } else if (d==4){ | ||
| frequency = 52; horizon = 13; load("Weekly.RData") | ||
| } else if (d==5){ | ||
| frequency = 7; horizon = 14; load("Daily.RData") | ||
| } else if (d==6){ | ||
| frequency = 168; horizon = 48; load("Hourly.RData") | ||
| } | ||
| tsn = length(series) | ||
|
|
||
| # Parallel processing of series | ||
| all_fcs <- foreach (tsi=sort((ids[2,][ids[1,]==d])), .combine='rbind', .packages=c('forecast', 'forecTheta', 'smooth')) %dopar% { | ||
|
|
||
| # Record progress | ||
| write(paste(d, tsi), file = paste("progress.txt"), append=FALSE) | ||
|
|
||
| # Retrieve series | ||
| y = ts(as.vector(series[[tsi]]$x), frequency = frequency) | ||
|
|
||
| # Define structures to hold forecasts and prediction intervals for one series | ||
| fcs = array(NA, c(4, horizon)) | ||
| pis = array(NA, c(4, horizon, 2)) | ||
|
|
||
| # ETS | ||
| if (frequency(y) > 24){ | ||
| # condition for W and H | ||
| fcs1 = es(y, h=horizon, intervals="p", level=0.95) | ||
| fcs[1,] = as.vector(fcs1$forecast) | ||
| } else { | ||
| fcs1 = forecast(ets(y), h=horizon, level=95) | ||
| fcs[1,] = as.vector(forecast(fcs1, h=horizon)$mean) | ||
| } | ||
| pis[1,,1] = fcs1$lower | ||
| pis[1,,2] = fcs1$upper | ||
|
|
||
| # CES | ||
| fcs2 = auto.ces(y, h=horizon, intervals="p", level=0.95) | ||
| fcs[2,] = as.vector(fcs2$forecast) | ||
| pis[2,,1] = fcs2$lower | ||
| pis[2,,2] = fcs2$upper | ||
|
|
||
| # AutoARIMA | ||
| fcs3 = forecast(auto.arima(y), h=horizon, level=95) | ||
| fcs[3,] = as.vector(forecast(fcs3, h=horizon)$mean) | ||
| pis[3,,1] = fcs3$lower | ||
| pis[3,,2] = fcs3$upper | ||
|
|
||
| # DOTM | ||
| if (length(as.vector(y)) > 5000){ | ||
| # condition for D2047, D2194 and D4099 | ||
| fcs4 = dotm(ts(tail(as.vector(y), 5000), frequency = frequency(y)), h=horizon, level=95) | ||
| } else { | ||
| fcs4 = dotm(y, h=horizon, level=95) | ||
| } | ||
| fcs[4,] = as.vector(fcs4$mean) | ||
| pis[4,,1] = fcs4$lower | ||
| pis[4,,2] = fcs4$upper | ||
|
|
||
| # bind forecasts and prediction intervals in a singe vector | ||
| fcspis = c(apply(fcs, 2, median), apply(pis[,,1], 2 ,median), apply(pis[,,2], 2 ,median)) | ||
|
|
||
| # set negative values to zero | ||
| fcspis[fcspis < 0] = 0 | ||
|
|
||
| fcspis | ||
|
|
||
| } | ||
|
|
||
| all_ffcs[(sum(ids[1,]<d)+1):(sum(ids[1,]<=d)), 1:horizon] = all_fcs[,1:horizon] | ||
| all_lower[(sum(ids[1,]<d)+1):(sum(ids[1,]<=d)), 1:horizon] = all_fcs[,(horizon+1):(2*horizon)] | ||
| all_upper[(sum(ids[1,]<d)+1):(sum(ids[1,]<=d)), 1:horizon] = all_fcs[,(2*horizon+1):(3*horizon)] | ||
|
|
||
| } | ||
|
|
||
| # Save forecasts and intervals in csv files | ||
| write.csv(all_ffcs, "all-ffcs.csv") | ||
| write.csv(all_lower, "all-lower.csv") | ||
| library(forecast) | ||
| library(smooth) | ||
| library(forecTheta) | ||
| library(foreach) | ||
| library(doSNOW) | ||
| cl = registerDoSNOW(makeCluster(4, type = "SOCK")) | ||
|
|
||
| # Datasets sizes | ||
| tsns = c(23000, 24000, 48000, 359, 4227, 414) | ||
|
|
||
| # *** SELECT WHICH SERIES *** | ||
|
|
||
| # The below will produce forecasts for the first 10 series of each frequency | ||
| ids = array(NA, c(2, 6*12)) | ||
| ids[1,] = rep(1:6, each=12) | ||
| ids[2,] = rep(12:1, 6) | ||
|
|
||
| # # The below will produce forecasts for the 13th yearly series, the 45th and the 10001th quarterly, the 30102th monthly and the 11th weekly | ||
| # ids = array(NA, c(2, 5)) | ||
| # ids[1,] = c(1,2,2,3,4) | ||
| # ids[2,] = c(13,45,10001,30102,11) | ||
|
|
||
| # # The below will produce forecasts for all series | ||
| # ids = array(NA, c(2,100000)) | ||
| # ids[1,] = c(rep(1,tsns[1]),rep(2,tsns[2]),rep(3,tsns[3]),rep(4,tsns[4]),rep(5,tsns[5]),rep(6,tsns[6])) | ||
| # ids[2,] = c(1:tsns[1],1:tsns[2],1:tsns[3],1:tsns[4],1:tsns[5],1:tsns[6]) | ||
|
|
||
| # # The below will produce forecasts for Y and Q series | ||
| # ids = array(NA, c(2, sum(tsns[1:2]))) | ||
| # ids[1,] = c(rep(1, tsns[1]), rep(2, tsns[2])) | ||
| # ids[2,] = c(1:tsns[1], 1:tsns[2]) | ||
|
|
||
| # # The below will produce forecasts for M series | ||
| # ids = array(NA, c(2, sum(tsns[3]))) | ||
| # ids[1,] = c(rep(3, tsns[3])) | ||
| # ids[2,] = c(1:tsns[3]) | ||
|
|
||
| # # The below will produce forecasts for W, D, and H series | ||
| # ids = array(NA, c(2, sum(tsns[4:6]))) | ||
| # ids[1,] = c(rep(4, tsns[4]), rep(5, tsns[5]), rep(6, tsns[6])) | ||
| # ids[2,] = c(1:tsns[4], 1:tsns[5], 1:tsns[6]) | ||
|
|
||
| # *** END OF SELECT WHICH SERIES *** | ||
|
|
||
| # Define structures for all forecasts and prediction intervals | ||
| datasets = sort(unique(ids[1,])) | ||
| the_names = list(c(if (any(datasets == 1)){paste0("Y", sort((ids[2,][ids[1,]==1])))}, | ||
| if (any(datasets == 2)){paste0("Q", sort((ids[2,][ids[1,]==2])))}, | ||
| if (any(datasets == 3)){paste0("M", sort((ids[2,][ids[1,]==3])))}, | ||
| if (any(datasets == 4)){paste0("W", sort((ids[2,][ids[1,]==4])))}, | ||
| if (any(datasets == 5)){paste0("D", sort((ids[2,][ids[1,]==5])))}, | ||
| if (any(datasets == 6)){paste0("H", sort((ids[2,][ids[1,]==6])))}), | ||
| paste0("F", 1:48)) | ||
| all_ffcs = array(NA, c(dim(ids)[2], 48), dimnames = the_names) | ||
| all_lower = array(NA, c(dim(ids)[2], 48), dimnames = the_names) | ||
| all_upper = array(NA, c(dim(ids)[2], 48), dimnames = the_names) | ||
|
|
||
| for (d in datasets){ | ||
|
|
||
| # Read dataset (produced by the separate R script) | ||
| if (d==1){ | ||
| frequency = 1; horizon = 6; load("Yearly.RData") | ||
| } else if (d==2){ | ||
| frequency = 4; horizon = 8; load("Quarterly.RData") | ||
| } else if (d==3){ | ||
| frequency = 12; horizon = 18; load("Monthly.RData") | ||
| } else if (d==4){ | ||
| frequency = 52; horizon = 13; load("Weekly.RData") | ||
| } else if (d==5){ | ||
| frequency = 7; horizon = 14; load("Daily.RData") | ||
| } else if (d==6){ | ||
| frequency = 168; horizon = 48; load("Hourly.RData") | ||
| } | ||
| tsn = length(series) | ||
|
|
||
| # Parallel processing of series | ||
| all_fcs <- foreach (tsi=sort((ids[2,][ids[1,]==d])), .combine='rbind', .packages=c('forecast', 'forecTheta', 'smooth')) %dopar% { | ||
|
|
||
| # Record progress | ||
| write(paste(d, tsi), file = paste("progress.txt"), append=FALSE) | ||
|
|
||
| # Retrieve series | ||
| y = ts(as.vector(series[[tsi]]$x), frequency = frequency) | ||
|
|
||
| # Define structures to hold forecasts and prediction intervals for one series | ||
| fcs = array(NA, c(4, horizon)) | ||
| pis = array(NA, c(4, horizon, 2)) | ||
|
|
||
| # ETS | ||
| if (frequency(y) > 24){ | ||
| # condition for W and H | ||
| fcs1 = es(y, h=horizon, intervals="p", level=0.95) | ||
| fcs[1,] = as.vector(fcs1$forecast) | ||
| } else { | ||
| fcs1 = forecast(ets(y), h=horizon, level=95) | ||
| fcs[1,] = as.vector(forecast(fcs1, h=horizon)$mean) | ||
| } | ||
| pis[1,,1] = fcs1$lower | ||
| pis[1,,2] = fcs1$upper | ||
|
|
||
| # CES | ||
| fcs2 = auto.ces(y, h=horizon, intervals="p", level=0.95) | ||
| fcs[2,] = as.vector(fcs2$forecast) | ||
| pis[2,,1] = fcs2$lower | ||
| pis[2,,2] = fcs2$upper | ||
|
|
||
| # AutoARIMA | ||
| fcs3 = forecast(auto.arima(y), h=horizon, level=95) | ||
| fcs[3,] = as.vector(forecast(fcs3, h=horizon)$mean) | ||
| pis[3,,1] = fcs3$lower | ||
| pis[3,,2] = fcs3$upper | ||
|
|
||
| # DOTM | ||
| if (length(as.vector(y)) > 5000){ | ||
| # condition for D2047, D2194 and D4099 | ||
| fcs4 = dotm(ts(tail(as.vector(y), 5000), frequency = frequency(y)), h=horizon, level=95) | ||
| } else { | ||
| fcs4 = dotm(y, h=horizon, level=95) | ||
| } | ||
| fcs[4,] = as.vector(fcs4$mean) | ||
| pis[4,,1] = fcs4$lower | ||
| pis[4,,2] = fcs4$upper | ||
|
|
||
| # bind forecasts and prediction intervals in a singe vector | ||
| fcspis = c(apply(fcs, 2, median), apply(pis[,,1], 2 ,median), apply(pis[,,2], 2 ,median)) | ||
|
|
||
| # set negative values to zero | ||
| fcspis[fcspis < 0] = 0 | ||
|
|
||
| fcspis | ||
|
|
||
| } | ||
|
|
||
| all_ffcs[(sum(ids[1,]<d)+1):(sum(ids[1,]<=d)), 1:horizon] = all_fcs[,1:horizon] | ||
| all_lower[(sum(ids[1,]<d)+1):(sum(ids[1,]<=d)), 1:horizon] = all_fcs[,(horizon+1):(2*horizon)] | ||
| all_upper[(sum(ids[1,]<d)+1):(sum(ids[1,]<=d)), 1:horizon] = all_fcs[,(2*horizon+1):(3*horizon)] | ||
|
|
||
| } | ||
|
|
||
| # Save forecasts and intervals in csv files | ||
| write.csv(all_ffcs, "all-ffcs.csv") | ||
| write.csv(all_lower, "all-lower.csv") | ||
| write.csv(all_upper, "all-upper.csv") |
File renamed without changes.
File renamed without changes.
Oops, something went wrong.