I was wondering what the best approach is for missing data in a time series.
I give an example using xts but I would like to know what seems to be the "best" method. Say I have library(xts) xts.ts <- xts(1:4,as.Date(c("1970-01-01", "1970-1-3", "1980-10-10", "2007-8-19")), frequency=52) I would like to turn this into a time series (still could be xts, or converted to ts) that has values for every week starting with the week that includes the start date and ending with the week that includes the end date. If there is data for the week then use it otherwise set it to NA or 0. Remember some years have 52, 53, or rarely 54 full or partial weeks. What to do with the partials at the beginning and ending of the year? This seems to be a fairly common problem and doing it myself is very cumbersome. Does a solution to this kind of problem exist? Once the approach to a weekly period is found I am sure that adjustment to daily, monthly, or quarterly would be relatively straightforward. Thank you. Kevin [[alternative HTML version deleted]] ______________________________________________ [hidden email] mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code. |
Couldn't you use seq.Date() to set up the time index and then just fill as appropriate?
Alternatively, to.weekly if you are starting with a daily series. Michael On Nov 22, 2011, at 4:00 PM, "Kevin Burton" <[hidden email]> wrote: > I was wondering what the best approach is for missing data in a time series. > I give an example using xts but I would like to know what seems to be the > "best" method. Say I have > > > > library(xts) > > xts.ts <- xts(1:4,as.Date(c("1970-01-01", "1970-1-3", "1980-10-10", > "2007-8-19")), frequency=52) > > > > I would like to turn this into a time series (still could be xts, or > converted to ts) that has values for every week starting with the week that > includes the start date and ending with the week that includes the end date. > If there is data for the week then use it otherwise set it to NA or 0. > Remember some years have 52, 53, or rarely 54 full or partial weeks. What to > do with the partials at the beginning and ending of the year? This seems to > be a fairly common problem and doing it myself is very cumbersome. Does a > solution to this kind of problem exist? Once the approach to a weekly period > is found I am sure that adjustment to daily, monthly, or quarterly would be > relatively straightforward. > > > > Thank you. > > > > Kevin > > > > > [[alternative HTML version deleted]] > > ______________________________________________ > [hidden email] mailing list > https://stat.ethz.ch/mailman/listinfo/r-help > PLEASE do read the posting guide http://www.R-project.org/posting-guide.html > and provide commented, minimal, self-contained, reproducible code. ______________________________________________ [hidden email] mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code. |
Thank you for the suggestions.
The only problems I see with 'to.weekly' is converting from the OHLC format and realizing that the date is the last day of the week rather than the first day of the week. Very minor compared to doing the whole thing myself. -----Original Message----- From: R. Michael Weylandt <[hidden email]> [mailto:[hidden email]] Sent: Tuesday, November 22, 2011 3:10 PM To: Kevin Burton Cc: <[hidden email]> Subject: Re: [R] Missing data? Couldn't you use seq.Date() to set up the time index and then just fill as appropriate? Alternatively, to.weekly if you are starting with a daily series. Michael On Nov 22, 2011, at 4:00 PM, "Kevin Burton" <[hidden email]> wrote: > I was wondering what the best approach is for missing data in a time series. > I give an example using xts but I would like to know what seems to be > the "best" method. Say I have > > > > library(xts) > > xts.ts <- xts(1:4,as.Date(c("1970-01-01", "1970-1-3", "1980-10-10", > "2007-8-19")), frequency=52) > > > > I would like to turn this into a time series (still could be xts, or > converted to ts) that has values for every week starting with the week > that includes the start date and ending with the week that includes the > If there is data for the week then use it otherwise set it to NA or 0. > Remember some years have 52, 53, or rarely 54 full or partial weeks. > What to do with the partials at the beginning and ending of the year? > This seems to be a fairly common problem and doing it myself is very > cumbersome. Does a solution to this kind of problem exist? Once the > approach to a weekly period is found I am sure that adjustment to > daily, monthly, or quarterly would be relatively straightforward. > > > > Thank you. > > > > Kevin > > > > > [[alternative HTML version deleted]] > > ______________________________________________ > [hidden email] mailing list > https://stat.ethz.ch/mailman/listinfo/r-help > PLEASE do read the posting guide > http://www.R-project.org/posting-guide.html > and provide commented, minimal, self-contained, reproducible code. ______________________________________________ [hidden email] mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code. |
In reply to this post by Michael Weylandt
Also with to.weekly there seems to be a problem with when the week starts.
For example: >xts.ts <- xts(1:4, c(as.Date("2011-01-01"), as.Date("2011-01-10"), as.Date("2011-10-09"), as.Date("2011-10-10")), frequency=52) > to.weekly(xts.ts) xts.ts.Open xts.ts.High xts.ts.Low xts.ts.Close 2011-01-01 1 1 1 1 2011-01-10 2 2 2 2 2011-10-09 3 3 3 3 2011-10-10 4 4 4 4 > xts.ts <- xts(1:4, c(as.Date("2011-01-01"), as.Date("2011-01-02"), as.Date("2011-10-09"), as.Date("2011-10-10")), frequency=52) > to.weekly(xts.ts) xts.ts.Open xts.ts.High xts.ts.Low xts.ts.Close 2011-01-02 1 2 1 2 2011-10-09 3 3 3 3 2011-10-10 4 4 4 4 So in the first case the week ends on January 1st. But the second indicates that the end of the week is the 2nd but it includes the data from the first. I would expect that the first column should be consistent. Notice that 10-09 and 10-10 are properly considered different weeks because the 9th is a Sunday and the 10th is a Monday (the beginning of the week). -----Original Message----- From: R. Michael Weylandt <[hidden email]> [mailto:[hidden email]] Sent: Tuesday, November 22, 2011 3:10 PM To: Kevin Burton Cc: <[hidden email]> Subject: Re: [R] Missing data? Couldn't you use seq.Date() to set up the time index and then just fill as appropriate? Alternatively, to.weekly if you are starting with a daily series. Michael On Nov 22, 2011, at 4:00 PM, "Kevin Burton" <[hidden email]> wrote: > I was wondering what the best approach is for missing data in a time series. > I give an example using xts but I would like to know what seems to be > the "best" method. Say I have > > > > library(xts) > > xts.ts <- xts(1:4,as.Date(c("1970-01-01", "1970-1-3", "1980-10-10", > "2007-8-19")), frequency=52) > > > > I would like to turn this into a time series (still could be xts, or > converted to ts) that has values for every week starting with the week > that includes the start date and ending with the week that includes the > If there is data for the week then use it otherwise set it to NA or 0. > Remember some years have 52, 53, or rarely 54 full or partial weeks. > What to do with the partials at the beginning and ending of the year? > This seems to be a fairly common problem and doing it myself is very > cumbersome. Does a solution to this kind of problem exist? Once the > approach to a weekly period is found I am sure that adjustment to > daily, monthly, or quarterly would be relatively straightforward. > > > > Thank you. > > > > Kevin > > > > > [[alternative HTML version deleted]] > > ______________________________________________ > [hidden email] mailing list > https://stat.ethz.ch/mailman/listinfo/r-help > PLEASE do read the posting guide > http://www.R-project.org/posting-guide.html > and provide commented, minimal, self-contained, reproducible code. ______________________________________________ [hidden email] mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code. |
In reply to this post by Michael Weylandt
Void of any other suggestions this approach makes sense but for my case I
think I need to use zoo objects rather than xts. If I sequence the data generally I don't know if there will be 365 days in the year or 366. So I have to sequence the dates as: seq(from=as.Date("2011-01-01"), to=as.Date("2011-12-31"), by="day") If I use this sequence with xts I get: > ds <- xts(NA, seq(from=as.Date("2011-01-01"), to=as.Date("2011-12-31"), by="day")) Error in xts(NA, seq(from = as.Date("2011-01-01"), to = as.Date("2011-12-31"), : NROW(x) must match length(order.by) If I leave the 'data' empty I don't get the error but if I try to assign an individual item (fill as appropriate) > ds <- xts(, seq(from=as.Date("2011-01-01"), to=as.Date("2011-12-31"), by="day")) > ds["2011-12-24"] <- 10 > ds Error in structure(coredata(x), names = x.attr$dimnames[[1]]) : 'names' attribute [365] must be the same length as the vector [358] So now I need to remember that I have not filled in all of the data. Also simple dereferencing gives: > ds[1] Error in `[.xts`(ds, 1) : subscript out of bounds With zoo I am able to create a time-series where all of the data is initially NA: > ds <- zoo(NA, seq(from=as.Date("2011-01-01"), to=as.Date("2011-12-31"), by="day")) So I can fill the data as appropriate and the remaining slots will have NA. I may be new with xts but I cannot see a way of creating a useable 'blank' time-series. Also with xts it seems like the frequency is ignored. > ds <- xts(1:365, seq(from=as.Date("2011-01-01"), to=as.Date("2011-12-31"), by="day"), frequency=52) > frequency(ds) [1] 1 Whereas zoo remembers the frequency setting > ds <- zoo(1:365, seq(from=as.Date("2011-01-01"), to=as.Date("2011-12-31"), by="day"), frequency=52) > frequency(ds) [1] 52 But since the ultimate goal is to get the time-series in a 'ts' format (as many functions require 'ts') it seems like even zoo has problems: > as.ts(ds) Time Series: Start = c(14975, 1) End = c(15339, 1) Frequency = 52 [1] 1 NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA [42] NA NA NA NA NA NA NA NA NA NA NA 2 NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA [83] NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA 3 NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA [124] NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA 4 NA NA NA NA NA NA NA [165] NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA [206] . . . . . . So the conversion from zoo to ts maintained the frequency but I am not sure where it decided on the start and end values. Also the conversion seemed to changed the data also. Notice that every period (52 entries) the original data is maintained. In other words if ds is the original zoo time series then ds[1] is 1 and ds[2] is 2 etc. The converted time-series keeps ds[1] but inserts 51 NA's then adds ds[2] etc till the end of the series. That is not what the initial data was. The conversion is inserting data of its own. The conversion to ts from xts seems better behaved: ds <- xts(1:365, seq(from=as.Date("2011-01-01"), to=as.Date("2011-12-31"), by="day"), frequency=52) > as.ts(ds) Time Series: Start = 1 End = 365 Frequency = 1 [1] 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 [43] 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 [85] 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 [127] 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 [169] 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 [211] 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 [253] 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 [295] 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 [337] 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 But alas the frequency is ignored. So this is what I have found out using these two packages. If I want to create a 'blank' data set it seems like zoo is 'better' since I can create a time-series initialized with NA irrespective of the length of the series. However I must be unfamiliar with the conversion because zoo doesn't convert to a regular 'ts' very well. But zoo remembers the frequency setting whereas xts just ignores it. It seems like there is still considerable work to solve the original problem. If I create a time series and fill in the values that are appropriate I still could have NA in the series it seems to.weekly has a problem with NA in the time series: > ds <- xts(rep(NA,365), seq(from=as.Date("2011-01-01"), to=as.Date("2011-12-31"), by="day"), frequency=52) > to.weekly(ds, sum) Error in if (drop.time) x <- .drop.time(x) : argument is not interpretable as logical In addition: Warning message: In to.period(x, "weeks", name = name, ...) : missing values removed from data -----Original Message----- From: R. Michael Weylandt <[hidden email]> [mailto:[hidden email]] Sent: Tuesday, November 22, 2011 3:10 PM To: Kevin Burton Cc: <[hidden email]> Subject: Re: [R] Missing data? Couldn't you use seq.Date() to set up the time index and then just fill as appropriate? Alternatively, to.weekly if you are starting with a daily series. Michael On Nov 22, 2011, at 4:00 PM, "Kevin Burton" <[hidden email]> wrote: > I was wondering what the best approach is for missing data in a time series. > I give an example using xts but I would like to know what seems to be > the "best" method. Say I have > > > > library(xts) > > xts.ts <- xts(1:4,as.Date(c("1970-01-01", "1970-1-3", "1980-10-10", > "2007-8-19")), frequency=52) > > > > I would like to turn this into a time series (still could be xts, or > converted to ts) that has values for every week starting with the week > that includes the start date and ending with the week that includes the > If there is data for the week then use it otherwise set it to NA or 0. > Remember some years have 52, 53, or rarely 54 full or partial weeks. > What to do with the partials at the beginning and ending of the year? > This seems to be a fairly common problem and doing it myself is very > cumbersome. Does a solution to this kind of problem exist? Once the > approach to a weekly period is found I am sure that adjustment to > daily, monthly, or quarterly would be relatively straightforward. > > > > Thank you. > > > > Kevin > > > > > [[alternative HTML version deleted]] > > ______________________________________________ > [hidden email] mailing list > https://stat.ethz.ch/mailman/listinfo/r-help > PLEASE do read the posting guide > http://www.R-project.org/posting-guide.html > and provide commented, minimal, self-contained, reproducible code. ______________________________________________ [hidden email] mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code. |
Why do you need to use a frequency attribute for these data? The point
of the zoo/xts line of time series implementations is that the time stamps are carried through for each observation (unlike ts) and can be irregular. Both classes exist precisely to avoid being forced into a frequency attribute. As far as setting up the time elements, wouldn't this work? Change the start date to get weeks on any desired day d <- seq.Date(from = as.Date("2011-11-26"), by = -7, length.out = 100) xts(rep(NA, length(d)), d) You can avoid the OHLC formatting of to.weekly if you want with the OHLC = FALSE parameter. And if you want to index it by the first of the week rather htan the last, just try this: time(x) <- time(x) - 6 Michael On Tue, Nov 22, 2011 at 6:50 PM, Kevin Burton <[hidden email]> wrote: > Void of any other suggestions this approach makes sense but for my case I > think I need to use zoo objects rather than xts. If I sequence the data > generally I don't know if there will be 365 days in the year or 366. So I > have to sequence the dates as: > > seq(from=as.Date("2011-01-01"), to=as.Date("2011-12-31"), by="day") > > If I use this sequence with xts I get: > >> ds <- xts(NA, seq(from=as.Date("2011-01-01"), to=as.Date("2011-12-31"), > by="day")) > Error in xts(NA, seq(from = as.Date("2011-01-01"), to = > as.Date("2011-12-31"), : > NROW(x) must match length(order.by) > > If I leave the 'data' empty I don't get the error but if I try to assign an > individual item (fill as appropriate) > >> ds <- xts(, seq(from=as.Date("2011-01-01"), to=as.Date("2011-12-31"), > by="day")) >> ds["2011-12-24"] <- 10 >> ds > Error in structure(coredata(x), names = x.attr$dimnames[[1]]) : > 'names' attribute [365] must be the same length as the vector [358] > > So now I need to remember that I have not filled in all of the data. Also > simple dereferencing gives: > >> ds[1] > Error in `[.xts`(ds, 1) : subscript out of bounds > > With zoo I am able to create a time-series where all of the data is > initially NA: > >> ds <- zoo(NA, seq(from=as.Date("2011-01-01"), to=as.Date("2011-12-31"), > by="day")) > > So I can fill the data as appropriate and the remaining slots will have NA. > I may be new with xts but I cannot see a way of creating a useable 'blank' > time-series. > > Also with xts it seems like the frequency is ignored. > >> ds <- xts(1:365, seq(from=as.Date("2011-01-01"), to=as.Date("2011-12-31"), > by="day"), frequency=52) >> frequency(ds) > [1] 1 > > Whereas zoo remembers the frequency setting > >> ds <- zoo(1:365, seq(from=as.Date("2011-01-01"), to=as.Date("2011-12-31"), > by="day"), frequency=52) >> frequency(ds) > [1] 52 > > But since the ultimate goal is to get the time-series in a 'ts' format (as > many functions require 'ts') it seems like even zoo has problems: > >> as.ts(ds) > > Time Series: > Start = c(14975, 1) > End = c(15339, 1) > Frequency = 52 > [1] 1 NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA > NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA > NA NA NA NA NA > [42] NA NA NA NA NA NA NA NA NA NA NA 2 NA NA NA NA NA > NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA > NA NA NA NA NA > [83] NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA > NA NA NA NA NA 3 NA NA NA NA NA NA NA NA NA NA NA NA NA > NA NA NA NA NA > [124] NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA > NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA 4 NA NA > NA NA NA NA NA > [165] NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA > NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA > NA NA NA NA NA > [206] . . . . . . > So the conversion from zoo to ts maintained the frequency but I am not sure > where it decided on the start and end values. Also the conversion seemed to > changed the data also. Notice that every period (52 entries) the original > data is maintained. In other words if ds is the original zoo time series > then ds[1] is 1 and ds[2] is 2 etc. The converted time-series keeps ds[1] > but inserts 51 NA's then adds ds[2] etc till the end of the series. That is > not what the initial data was. The conversion is inserting data of its own. > > The conversion to ts from xts seems better behaved: > > ds <- xts(1:365, seq(from=as.Date("2011-01-01"), to=as.Date("2011-12-31"), > by="day"), frequency=52) >> as.ts(ds) > Time Series: > Start = 1 > End = 365 > Frequency = 1 > [1] 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 > 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 > 37 38 39 40 41 42 > [43] 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 > 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 > 79 80 81 82 83 84 > [85] 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 > 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 > 121 122 123 124 125 126 > [127] 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 > 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 > 163 164 165 166 167 168 > [169] 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 > 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 > 205 206 207 208 209 210 > [211] 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 > 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 > 247 248 249 250 251 252 > [253] 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 > 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 > 289 290 291 292 293 294 > [295] 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 > 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 > 331 332 333 334 335 336 > [337] 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 > 354 355 356 357 358 359 360 361 362 363 364 365 > > But alas the frequency is ignored. > > So this is what I have found out using these two packages. If I want to > create a 'blank' data set it seems like zoo is 'better' since I can create a > time-series initialized with NA irrespective of the length of the series. > However I must be unfamiliar with the conversion because zoo doesn't convert > to a regular 'ts' very well. But zoo remembers the frequency setting > whereas xts just ignores it. > > It seems like there is still considerable work to solve the original > problem. If I create a time series and fill in the values that are > appropriate I still could have NA in the series it seems to.weekly has a > problem with NA in the time series: >> ds <- xts(rep(NA,365), seq(from=as.Date("2011-01-01"), > to=as.Date("2011-12-31"), by="day"), frequency=52) >> to.weekly(ds, sum) > Error in if (drop.time) x <- .drop.time(x) : > argument is not interpretable as logical > In addition: Warning message: > In to.period(x, "weeks", name = name, ...) : > missing values removed from data > > > -----Original Message----- > From: R. Michael Weylandt <[hidden email]> > [mailto:[hidden email]] > Sent: Tuesday, November 22, 2011 3:10 PM > To: Kevin Burton > Cc: <[hidden email]> > Subject: Re: [R] Missing data? > > Couldn't you use seq.Date() to set up the time index and then just fill as > appropriate? > > Alternatively, to.weekly if you are starting with a daily series. > > Michael > > On Nov 22, 2011, at 4:00 PM, "Kevin Burton" <[hidden email]> > wrote: > >> I was wondering what the best approach is for missing data in a time > series. >> I give an example using xts but I would like to know what seems to be >> the "best" method. Say I have >> >> >> >> library(xts) >> >> xts.ts <- xts(1:4,as.Date(c("1970-01-01", "1970-1-3", "1980-10-10", >> "2007-8-19")), frequency=52) >> >> >> >> I would like to turn this into a time series (still could be xts, or >> converted to ts) that has values for every week starting with the week >> that includes the start date and ending with the week that includes the > end date. >> If there is data for the week then use it otherwise set it to NA or 0. >> Remember some years have 52, 53, or rarely 54 full or partial weeks. >> What to do with the partials at the beginning and ending of the year? >> This seems to be a fairly common problem and doing it myself is very >> cumbersome. Does a solution to this kind of problem exist? Once the >> approach to a weekly period is found I am sure that adjustment to >> daily, monthly, or quarterly would be relatively straightforward. >> >> >> >> Thank you. >> >> >> >> Kevin >> >> >> >> >> [[alternative HTML version deleted]] >> >> ______________________________________________ >> [hidden email] mailing list >> https://stat.ethz.ch/mailman/listinfo/r-help >> PLEASE do read the posting guide >> http://www.R-project.org/posting-guide.html >> and provide commented, minimal, self-contained, reproducible code. > > ______________________________________________ [hidden email] mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code. |
In reply to this post by rkevinburton
On Tue, Nov 22, 2011 at 6:50 PM, Kevin Burton <[hidden email]> wrote:
> Void of any other suggestions this approach makes sense but for my case I > think I need to use zoo objects rather than xts. If I sequence the data > generally I don't know if there will be 365 days in the year or 366. So I > have to sequence the dates as: > > seq(from=as.Date("2011-01-01"), to=as.Date("2011-12-31"), by="day") > > If I use this sequence with xts I get: > >> ds <- xts(NA, seq(from=as.Date("2011-01-01"), to=as.Date("2011-12-31"), > by="day")) > Error in xts(NA, seq(from = as.Date("2011-01-01"), to = > as.Date("2011-12-31"), : > NROW(x) must match length(order.by) > > If I leave the 'data' empty I don't get the error but if I try to assign an > individual item (fill as appropriate) > >> ds <- xts(, seq(from=as.Date("2011-01-01"), to=as.Date("2011-12-31"), > by="day")) >> ds["2011-12-24"] <- 10 >> ds > Error in structure(coredata(x), names = x.attr$dimnames[[1]]) : > 'names' attribute [365] must be the same length as the vector [358] > > So now I need to remember that I have not filled in all of the data. Also > simple dereferencing gives: > >> ds[1] > Error in `[.xts`(ds, 1) : subscript out of bounds > > With zoo I am able to create a time-series where all of the data is > initially NA: > >> ds <- zoo(NA, seq(from=as.Date("2011-01-01"), to=as.Date("2011-12-31"), > by="day")) > > So I can fill the data as appropriate and the remaining slots will have NA. > I may be new with xts but I cannot see a way of creating a useable 'blank' > time-series. > > Also with xts it seems like the frequency is ignored. > >> ds <- xts(1:365, seq(from=as.Date("2011-01-01"), to=as.Date("2011-12-31"), > by="day"), frequency=52) >> frequency(ds) > [1] 1 > > Whereas zoo remembers the frequency setting > >> ds <- zoo(1:365, seq(from=as.Date("2011-01-01"), to=as.Date("2011-12-31"), > by="day"), frequency=52) >> frequency(ds) > [1] 52 > > But since the ultimate goal is to get the time-series in a 'ts' format (as > many functions require 'ts') it seems like even zoo has problems: The problem is that you seem to want a fixed number of periods per year but there is not a constant of 52 weeks nor 365 days in a year. You are going to have give up something since your apparent criteria conflict with reality. For example, you could use months in which case there are exactly 12 or you could stick more than 7 days into the first or last week of the year so that there are exactly 52 weeks in a year but they don't all have the same number of days, etc. -- Statistics & Software Consulting GKX Group, GKX Associates Inc. tel: 1-877-GKX-GROUP email: ggrothendieck at gmail.com ______________________________________________ [hidden email] mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code. |
I admit it isnt reality but I was hoping through judicious use of these functions I could approximate reality. For example in the years where there are more than 53 weeks in a year I would be happy if there were a way to recognize this and drop the last week of data. If there were less than 53 I would "pad" the year with an extra dummy week. This is just about the same as your suggestion of putting more than 7 days in the first and last weeks. But i still need this kind of date manipulation to even know how many days to add in to make the approximation viable. This kind of best approximation to reality seems better than to settle for the resolution of a month just because it is consistent. Daily would be too much data and even then there would be an approximation due to leap years.
On Nov 26, 2011, at 3:13 PM, Gabor Grothendieck <[hidden email]> wrote: > On Tue, Nov 22, 2011 at 6:50 PM, Kevin Burton <[hidden email]> wrote: >> Void of any other suggestions this approach makes sense but for my case I >> think I need to use zoo objects rather than xts. If I sequence the data >> generally I don't know if there will be 365 days in the year or 366. So I >> have to sequence the dates as: >> >> seq(from=as.Date("2011-01-01"), to=as.Date("2011-12-31"), by="day") >> >> If I use this sequence with xts I get: >> >>> ds <- xts(NA, seq(from=as.Date("2011-01-01"), to=as.Date("2011-12-31"), >> by="day")) >> Error in xts(NA, seq(from = as.Date("2011-01-01"), to = >> as.Date("2011-12-31"), : >> NROW(x) must match length(order.by) >> >> If I leave the 'data' empty I don't get the error but if I try to assign an >> individual item (fill as appropriate) >> >>> ds <- xts(, seq(from=as.Date("2011-01-01"), to=as.Date("2011-12-31"), >> by="day")) >>> ds["2011-12-24"] <- 10 >>> ds >> Error in structure(coredata(x), names = x.attr$dimnames[[1]]) : >> 'names' attribute [365] must be the same length as the vector [358] >> >> So now I need to remember that I have not filled in all of the data. Also >> simple dereferencing gives: >> >>> ds[1] >> Error in `[.xts`(ds, 1) : subscript out of bounds >> >> With zoo I am able to create a time-series where all of the data is >> initially NA: >> >>> ds <- zoo(NA, seq(from=as.Date("2011-01-01"), to=as.Date("2011-12-31"), >> by="day")) >> >> So I can fill the data as appropriate and the remaining slots will have NA. >> I may be new with xts but I cannot see a way of creating a useable 'blank' >> time-series. >> >> Also with xts it seems like the frequency is ignored. >> >>> ds <- xts(1:365, seq(from=as.Date("2011-01-01"), to=as.Date("2011-12-31"), >> by="day"), frequency=52) >>> frequency(ds) >> [1] 1 >> >> Whereas zoo remembers the frequency setting >> >>> ds <- zoo(1:365, seq(from=as.Date("2011-01-01"), to=as.Date("2011-12-31"), >> by="day"), frequency=52) >>> frequency(ds) >> [1] 52 >> >> But since the ultimate goal is to get the time-series in a 'ts' format (as >> many functions require 'ts') it seems like even zoo has problems: > > The problem is that you seem to want a fixed number of periods per > year but there is not a constant of 52 weeks nor 365 days in a year. > You are going to have give up something since your apparent criteria > conflict with reality. For example, you could use months in which > case there are exactly 12 or you could stick more than 7 days into the > first or last week of the year so that there are exactly 52 weeks in a > year but they don't all have the same number of days, etc. > > -- > Statistics & Software Consulting > GKX Group, GKX Associates Inc. > tel: 1-877-GKX-GROUP > email: ggrothendieck at gmail.com ______________________________________________ [hidden email] mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code. |
In reply to this post by Michael Weylandt
I was just trying to be complete. Why is the frequency argument and
attribute available? -----Original Message----- From: R. Michael Weylandt [mailto:[hidden email]] Sent: Saturday, November 26, 2011 2:40 PM To: Kevin Burton Cc: [hidden email] Subject: Re: [R] Missing data? Why do you need to use a frequency attribute for these data? The point of the zoo/xts line of time series implementations is that the time stamps are carried through for each observation (unlike ts) and can be irregular. Both classes exist precisely to avoid being forced into a frequency attribute. As far as setting up the time elements, wouldn't this work? Change the start date to get weeks on any desired day d <- seq.Date(from = as.Date("2011-11-26"), by = -7, length.out = 100) xts(rep(NA, length(d)), d) You can avoid the OHLC formatting of to.weekly if you want with the OHLC = FALSE parameter. And if you want to index it by the first of the week rather htan the last, just try this: time(x) <- time(x) - 6 Michael On Tue, Nov 22, 2011 at 6:50 PM, Kevin Burton <[hidden email]> wrote: > Void of any other suggestions this approach makes sense but for my > case I think I need to use zoo objects rather than xts. If I sequence > the data generally I don't know if there will be 365 days in the year > or 366. So I have to sequence the dates as: > > seq(from=as.Date("2011-01-01"), to=as.Date("2011-12-31"), by="day") > > If I use this sequence with xts I get: > >> ds <- xts(NA, seq(from=as.Date("2011-01-01"), >> to=as.Date("2011-12-31"), > by="day")) > Error in xts(NA, seq(from = as.Date("2011-01-01"), to = > as.Date("2011-12-31"), : > NROW(x) must match length(order.by) > > If I leave the 'data' empty I don't get the error but if I try to > assign an individual item (fill as appropriate) > >> ds <- xts(, seq(from=as.Date("2011-01-01"), to=as.Date("2011-12-31"), > by="day")) >> ds["2011-12-24"] <- 10 >> ds > Error in structure(coredata(x), names = x.attr$dimnames[[1]]) : > 'names' attribute [365] must be the same length as the vector [358] > > So now I need to remember that I have not filled in all of the data. > Also simple dereferencing gives: > >> ds[1] > Error in `[.xts`(ds, 1) : subscript out of bounds > > With zoo I am able to create a time-series where all of the data is > initially NA: > >> ds <- zoo(NA, seq(from=as.Date("2011-01-01"), >> to=as.Date("2011-12-31"), > by="day")) > > So I can fill the data as appropriate and the remaining slots will have > I may be new with xts but I cannot see a way of creating a useable 'blank' > time-series. > > Also with xts it seems like the frequency is ignored. > >> ds <- xts(1:365, seq(from=as.Date("2011-01-01"), >> to=as.Date("2011-12-31"), > by="day"), frequency=52) >> frequency(ds) > [1] 1 > > Whereas zoo remembers the frequency setting > >> ds <- zoo(1:365, seq(from=as.Date("2011-01-01"), >> to=as.Date("2011-12-31"), > by="day"), frequency=52) >> frequency(ds) > [1] 52 > > But since the ultimate goal is to get the time-series in a 'ts' format > (as many functions require 'ts') it seems like even zoo has problems: > >> as.ts(ds) > > Time Series: > Start = c(14975, 1) > End = c(15339, 1) > Frequency = 52 > [1] 1 NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA > NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA > NA NA NA NA NA NA NA > [42] NA NA NA NA NA NA NA NA NA NA NA 2 NA NA NA NA > NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA > NA NA NA NA NA NA NA > [83] NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA > NA NA NA NA NA NA 3 NA NA NA NA NA NA NA NA NA NA NA > NA NA NA NA NA NA NA > [124] NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA > NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA 4 > NA NA NA NA NA NA NA > [165] NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA > NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA > NA NA NA NA NA NA NA > [206] . . . . . . > So the conversion from zoo to ts maintained the frequency but I am > not sure where it decided on the start and end values. Also the > conversion seemed to changed the data also. Notice that every period > (52 entries) the original data is maintained. In other words if ds is > the original zoo time series then ds[1] is 1 and ds[2] is 2 etc. The > converted time-series keeps ds[1] but inserts 51 NA's then adds ds[2] > etc till the end of the series. That is not what the initial data was. > > The conversion to ts from xts seems better behaved: > > ds <- xts(1:365, seq(from=as.Date("2011-01-01"), > to=as.Date("2011-12-31"), by="day"), frequency=52) >> as.ts(ds) > Time Series: > Start = 1 > End = 365 > Frequency = 1 > [1] 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 > 17 > 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 > 36 > 37 38 39 40 41 42 > [43] 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 > 59 > 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 > 78 > 79 80 81 82 83 84 > [85] 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 > 101 > 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 > 119 120 > 121 122 123 124 125 126 > [127] 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 > 143 > 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 > 161 162 > 163 164 165 166 167 168 > [169] 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 > 185 > 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 > 203 204 > 205 206 207 208 209 210 > [211] 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 > 227 > 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 > 245 246 > 247 248 249 250 251 252 > [253] 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 > 269 > 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 > 287 288 > 289 290 291 292 293 294 > [295] 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 > 311 > 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 > 329 330 > 331 332 333 334 335 336 > [337] 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 > 353 > 354 355 356 357 358 359 360 361 362 363 364 365 > > But alas the frequency is ignored. > > So this is what I have found out using these two packages. If I want > to create a 'blank' data set it seems like zoo is 'better' since I can > create a time-series initialized with NA irrespective of the length of the > However I must be unfamiliar with the conversion because zoo doesn't > convert to a regular 'ts' very well. But zoo remembers the frequency > setting whereas xts just ignores it. > > It seems like there is still considerable work to solve the original > problem. If I create a time series and fill in the values that are > appropriate I still could have NA in the series it seems to.weekly has > a problem with NA in the time series: >> ds <- xts(rep(NA,365), seq(from=as.Date("2011-01-01"), > to=as.Date("2011-12-31"), by="day"), frequency=52) >> to.weekly(ds, sum) > Error in if (drop.time) x <- .drop.time(x) : > argument is not interpretable as logical In addition: Warning > message: > In to.period(x, "weeks", name = name, ...) : > missing values removed from data > > > -----Original Message----- > From: R. Michael Weylandt <[hidden email]> > [mailto:[hidden email]] > Sent: Tuesday, November 22, 2011 3:10 PM > To: Kevin Burton > Cc: <[hidden email]> > Subject: Re: [R] Missing data? > > Couldn't you use seq.Date() to set up the time index and then just > fill as appropriate? > > Alternatively, to.weekly if you are starting with a daily series. > > Michael > > On Nov 22, 2011, at 4:00 PM, "Kevin Burton" <[hidden email]> > wrote: > >> I was wondering what the best approach is for missing data in a time > series. >> I give an example using xts but I would like to know what seems to be >> the "best" method. Say I have >> >> >> >> library(xts) >> >> xts.ts <- xts(1:4,as.Date(c("1970-01-01", "1970-1-3", "1980-10-10", >> "2007-8-19")), frequency=52) >> >> >> >> I would like to turn this into a time series (still could be xts, or >> converted to ts) that has values for every week starting with the >> week that includes the start date and ending with the week that >> includes the > end date. >> If there is data for the week then use it otherwise set it to NA or 0. >> Remember some years have 52, 53, or rarely 54 full or partial weeks. >> What to do with the partials at the beginning and ending of the year? >> This seems to be a fairly common problem and doing it myself is very >> cumbersome. Does a solution to this kind of problem exist? Once the >> approach to a weekly period is found I am sure that adjustment to >> daily, monthly, or quarterly would be relatively straightforward. >> >> >> >> Thank you. >> >> >> >> Kevin >> >> >> >> >> [[alternative HTML version deleted]] >> >> ______________________________________________ >> [hidden email] mailing list >> https://stat.ethz.ch/mailman/listinfo/r-help >> PLEASE do read the posting guide >> http://www.R-project.org/posting-guide.html >> and provide commented, minimal, self-contained, reproducible code. > > ______________________________________________ [hidden email] mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code. |
Back compatibility with other time series best I can tell, but to be
honest, I'm not even sure how it plays into that. Perhaps it's just an artifact in the signature. It doesn't seem to have a role in the xts constructor. E.g., identical(xts(1:5, Sys.Date()+1:5, frequency = 1), xts(1:5, Sys.Date()+1:5, frequency = 3)) Michael On Sun, Nov 27, 2011 at 4:51 PM, Kevin Burton <[hidden email]> wrote: > I was just trying to be complete. Why is the frequency argument and > attribute available? > > -----Original Message----- > From: R. Michael Weylandt [mailto:[hidden email]] > Sent: Saturday, November 26, 2011 2:40 PM > To: Kevin Burton > Cc: [hidden email] > Subject: Re: [R] Missing data? > > Why do you need to use a frequency attribute for these data? The point of > the zoo/xts line of time series implementations is that the time stamps are > carried through for each observation (unlike ts) and can be irregular. Both > classes exist precisely to avoid being forced into a frequency attribute. > > As far as setting up the time elements, wouldn't this work? Change the start > date to get weeks on any desired day > > d <- seq.Date(from = as.Date("2011-11-26"), by = -7, length.out = 100) > xts(rep(NA, length(d)), d) > > You can avoid the OHLC formatting of to.weekly if you want with the OHLC = > FALSE parameter. And if you want to index it by the first of the week rather > htan the last, just try this: > > time(x) <- time(x) - 6 > > Michael > > On Tue, Nov 22, 2011 at 6:50 PM, Kevin Burton <[hidden email]> > wrote: >> Void of any other suggestions this approach makes sense but for my >> case I think I need to use zoo objects rather than xts. If I sequence >> the data generally I don't know if there will be 365 days in the year >> or 366. So I have to sequence the dates as: >> >> seq(from=as.Date("2011-01-01"), to=as.Date("2011-12-31"), by="day") >> >> If I use this sequence with xts I get: >> >>> ds <- xts(NA, seq(from=as.Date("2011-01-01"), >>> to=as.Date("2011-12-31"), >> by="day")) >> Error in xts(NA, seq(from = as.Date("2011-01-01"), to = >> as.Date("2011-12-31"), : >> NROW(x) must match length(order.by) >> >> If I leave the 'data' empty I don't get the error but if I try to >> assign an individual item (fill as appropriate) >> >>> ds <- xts(, seq(from=as.Date("2011-01-01"), to=as.Date("2011-12-31"), >> by="day")) >>> ds["2011-12-24"] <- 10 >>> ds >> Error in structure(coredata(x), names = x.attr$dimnames[[1]]) : >> 'names' attribute [365] must be the same length as the vector [358] >> >> So now I need to remember that I have not filled in all of the data. >> Also simple dereferencing gives: >> >>> ds[1] >> Error in `[.xts`(ds, 1) : subscript out of bounds >> >> With zoo I am able to create a time-series where all of the data is >> initially NA: >> >>> ds <- zoo(NA, seq(from=as.Date("2011-01-01"), >>> to=as.Date("2011-12-31"), >> by="day")) >> >> So I can fill the data as appropriate and the remaining slots will have > NA. >> I may be new with xts but I cannot see a way of creating a useable 'blank' >> time-series. >> >> Also with xts it seems like the frequency is ignored. >> >>> ds <- xts(1:365, seq(from=as.Date("2011-01-01"), >>> to=as.Date("2011-12-31"), >> by="day"), frequency=52) >>> frequency(ds) >> [1] 1 >> >> Whereas zoo remembers the frequency setting >> >>> ds <- zoo(1:365, seq(from=as.Date("2011-01-01"), >>> to=as.Date("2011-12-31"), >> by="day"), frequency=52) >>> frequency(ds) >> [1] 52 >> >> But since the ultimate goal is to get the time-series in a 'ts' format >> (as many functions require 'ts') it seems like even zoo has problems: >> >>> as.ts(ds) >> >> Time Series: >> Start = c(14975, 1) >> End = c(15339, 1) >> Frequency = 52 >> [1] 1 NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA >> NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA >> NA NA NA NA NA NA NA >> [42] NA NA NA NA NA NA NA NA NA NA NA 2 NA NA NA NA >> NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA >> NA NA NA NA NA NA NA >> [83] NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA >> NA NA NA NA NA NA 3 NA NA NA NA NA NA NA NA NA NA NA >> NA NA NA NA NA NA NA >> [124] NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA >> NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA 4 >> NA NA NA NA NA NA NA >> [165] NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA >> NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA >> NA NA NA NA NA NA NA >> [206] . . . . . . >> So the conversion from zoo to ts maintained the frequency but I am >> not sure where it decided on the start and end values. Also the >> conversion seemed to changed the data also. Notice that every period >> (52 entries) the original data is maintained. In other words if ds is >> the original zoo time series then ds[1] is 1 and ds[2] is 2 etc. The >> converted time-series keeps ds[1] but inserts 51 NA's then adds ds[2] >> etc till the end of the series. That is not what the initial data was. > The conversion is inserting data of its own. >> >> The conversion to ts from xts seems better behaved: >> >> ds <- xts(1:365, seq(from=as.Date("2011-01-01"), >> to=as.Date("2011-12-31"), by="day"), frequency=52) >>> as.ts(ds) >> Time Series: >> Start = 1 >> End = 365 >> Frequency = 1 >> [1] 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 >> 17 >> 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 >> 36 >> 37 38 39 40 41 42 >> [43] 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 >> 59 >> 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 >> 78 >> 79 80 81 82 83 84 >> [85] 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 >> 101 >> 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 >> 119 120 >> 121 122 123 124 125 126 >> [127] 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 >> 143 >> 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 >> 161 162 >> 163 164 165 166 167 168 >> [169] 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 >> 185 >> 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 >> 203 204 >> 205 206 207 208 209 210 >> [211] 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 >> 227 >> 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 >> 245 246 >> 247 248 249 250 251 252 >> [253] 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 >> 269 >> 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 >> 287 288 >> 289 290 291 292 293 294 >> [295] 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 >> 311 >> 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 >> 329 330 >> 331 332 333 334 335 336 >> [337] 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 >> 353 >> 354 355 356 357 358 359 360 361 362 363 364 365 >> >> But alas the frequency is ignored. >> >> So this is what I have found out using these two packages. If I want >> to create a 'blank' data set it seems like zoo is 'better' since I can >> create a time-series initialized with NA irrespective of the length of the > series. >> However I must be unfamiliar with the conversion because zoo doesn't >> convert to a regular 'ts' very well. But zoo remembers the frequency >> setting whereas xts just ignores it. >> >> It seems like there is still considerable work to solve the original >> problem. If I create a time series and fill in the values that are >> appropriate I still could have NA in the series it seems to.weekly has >> a problem with NA in the time series: >>> ds <- xts(rep(NA,365), seq(from=as.Date("2011-01-01"), >> to=as.Date("2011-12-31"), by="day"), frequency=52) >>> to.weekly(ds, sum) >> Error in if (drop.time) x <- .drop.time(x) : >> argument is not interpretable as logical In addition: Warning >> message: >> In to.period(x, "weeks", name = name, ...) : >> missing values removed from data >> >> >> -----Original Message----- >> From: R. Michael Weylandt <[hidden email]> >> [mailto:[hidden email]] >> Sent: Tuesday, November 22, 2011 3:10 PM >> To: Kevin Burton >> Cc: <[hidden email]> >> Subject: Re: [R] Missing data? >> >> Couldn't you use seq.Date() to set up the time index and then just >> fill as appropriate? >> >> Alternatively, to.weekly if you are starting with a daily series. >> >> Michael >> >> On Nov 22, 2011, at 4:00 PM, "Kevin Burton" <[hidden email]> >> wrote: >> >>> I was wondering what the best approach is for missing data in a time >> series. >>> I give an example using xts but I would like to know what seems to be >>> the "best" method. Say I have >>> >>> >>> >>> library(xts) >>> >>> xts.ts <- xts(1:4,as.Date(c("1970-01-01", "1970-1-3", "1980-10-10", >>> "2007-8-19")), frequency=52) >>> >>> >>> >>> I would like to turn this into a time series (still could be xts, or >>> converted to ts) that has values for every week starting with the >>> week that includes the start date and ending with the week that >>> includes the >> end date. >>> If there is data for the week then use it otherwise set it to NA or 0. >>> Remember some years have 52, 53, or rarely 54 full or partial weeks. >>> What to do with the partials at the beginning and ending of the year? >>> This seems to be a fairly common problem and doing it myself is very >>> cumbersome. Does a solution to this kind of problem exist? Once the >>> approach to a weekly period is found I am sure that adjustment to >>> daily, monthly, or quarterly would be relatively straightforward. >>> >>> >>> >>> Thank you. >>> >>> >>> >>> Kevin >>> >>> >>> >>> >>> [[alternative HTML version deleted]] >>> >>> ______________________________________________ >>> [hidden email] mailing list >>> https://stat.ethz.ch/mailman/listinfo/r-help >>> PLEASE do read the posting guide >>> http://www.R-project.org/posting-guide.html >>> and provide commented, minimal, self-contained, reproducible code. >> >> > > ______________________________________________ [hidden email] mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code. |
In reply to this post by rkevinburton
On Sun, Nov 27, 2011 at 4:08 PM, Kevin Burton <[hidden email]> wrote:
> I admit it isnt reality but I was hoping through judicious use of these functions I could approximate reality. For example in the years where there are more than 53 weeks in a year I would be happy if there were a way to recognize this and drop the last week of data. If there were less than 53 I would "pad" the year with an extra dummy week. This is just about the same as your suggestion of putting more than 7 days in the first and last weeks. But i still need this kind of date manipulation to even know how many days to add in to make the approximation viable. This kind of best approximation to reality seems better than to settle for the resolution of a month just because it is consistent. Daily would be too much data and even then there would be an approximation due to leap years. > OK. As you are willing to regard days past the 364th as part of the last week of the year then we can do this. Create a zoo object z as test data. Then convert its time scale to year + week/52 where 0 is the first week of the year and we replace any week that is greater than 51 with 51. Then we aggregate z by week taking the last data point in the week and convert it to ts. Because of the way we constructed it the frequency will be 52. library(zoo) # test data z <- zoo(1:100, Sys.Date() + 1:100) yr.wk <- with(as.POSIXlt(time(z)), year + 1900 + pmin(yday %/% 7, 51) / 52) z.wk <- aggregate(z, yr.wk, tail, 1) z.ts <- as.ts(z.wk) frequency(z.ts) # 52 -- Statistics & Software Consulting GKX Group, GKX Associates Inc. tel: 1-877-GKX-GROUP email: ggrothendieck at gmail.com ______________________________________________ [hidden email] mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code. |
This has been very helpful. Thank you.
At the risk of further confirming my ignorance and taxing your patience I would like to add another question. How would I modify this code so that each week starts with the same day of the week regardless of the year? I would add this stipulation so that for multiple years I always get the same 'week-number' like > format(as.Date("2011-11-27"), "%W-%w") [1] "47-0" The convention (at least for US culture) seems to be that the week starts with Sunday (it is index 0 for day of week). So it would be convenient if the code was modified so that each 'week' began on Sunday. The partial at the beginning would just start with the day of week that was at the start. I still would want to aggregate that 'week-number's that are greater than 51 like you have shown. Thanks again. Kevin -----Original Message----- From: Gabor Grothendieck [mailto:[hidden email]] Sent: Sunday, November 27, 2011 4:24 PM To: Kevin Burton Cc: [hidden email] Subject: Re: [R] Missing data? On Sun, Nov 27, 2011 at 4:08 PM, Kevin Burton <[hidden email]> wrote: > I admit it isnt reality but I was hoping through judicious use of these functions I could approximate reality. For example in the years where there are more than 53 weeks in a year I would be happy if there were a way to recognize this and drop the last week of data. If there were less than 53 I would "pad" the year with an extra dummy week. This is just about the same as your suggestion of putting more than 7 days in the first and last weeks. But i still need this kind of date manipulation to even know how many days to add in to make the approximation viable. This kind of best approximation to reality seems better than to settle for the resolution of a month just because it is consistent. Daily would be too much data and even then there would be an approximation due to leap years. > OK. As you are willing to regard days past the 364th as part of the last week of the year then we can do this. Create a zoo object z as test data. Then convert its time scale to year + week/52 where 0 is the first week of the year and we replace any week that is greater than 51 with 51. Then we aggregate z by week taking the last data point in the week and convert it to ts. Because of the way we constructed it the frequency will be 52. library(zoo) # test data z <- zoo(1:100, Sys.Date() + 1:100) yr.wk <- with(as.POSIXlt(time(z)), year + 1900 + pmin(yday %/% 7, 51) / 52) z.wk <- aggregate(z, yr.wk, tail, 1) z.ts <- as.ts(z.wk) frequency(z.ts) # 52 -- Statistics & Software Consulting GKX Group, GKX Associates Inc. tel: 1-877-GKX-GROUP email: ggrothendieck at gmail.com ______________________________________________ [hidden email] mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code. |
On Sun, Nov 27, 2011 at 8:10 PM, Kevin Burton <[hidden email]> wrote:
> This has been very helpful. Thank you. > > At the risk of further confirming my ignorance and taxing your patience I > would like to add another question. How would I modify this code so that > each week starts with the same day of the week regardless of the year? I > would add this stipulation so that for multiple years I always get the same > 'week-number' like > >> format(as.Date("2011-11-27"), "%W-%w") > [1] "47-0" > > The convention (at least for US culture) seems to be that the week starts > with Sunday (it is index 0 for day of week). So it would be convenient if > the code was modified so that each 'week' began on Sunday. The partial at > the beginning would just start with the day of week that was at the start. I > still would want to aggregate that 'week-number's that are greater than 51 > like you have shown. It would be the same except replace the calculation of yr.wk with: tt <- time(z) yr.wk <- as.numeric(format(tt, "%Y")) + pmin(as.numeric(format(tt, "%W")), 51)/52 This puts the 1st Sunday of the year and the days prior to it in week 0, the next 7 days are in week 1 and so on. All days after the 51st Sunday are either in week 51 or are forced to be. -- Statistics & Software Consulting GKX Group, GKX Associates Inc. tel: 1-877-GKX-GROUP email: ggrothendieck at gmail.com ______________________________________________ [hidden email] mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code. |
Free forum by Nabble - Free Resume Builder | Edit this page |