I have a data.frame consisting of about 300k rows with 24 rows for each ID - each row representing an hourly observation of that ID. My problem lies in that for some IDs the observation ends before the 24 hours has gone by - yet still have 24 rows with the remaining rows having NA in their 3 observation variables.
In a simplified table would be something like this
ID HOUR OBS_1 OBS_2 OBS_3 MISC MISC_2
1 0 29 32 34 19 21
1 1 21 12 NA 19 21
1 2 NA 24 NA 19 21
1 3 NA NA NA 19 21
1 4 NA NA NA 19 21
2 0 41 16 21 13 24
2 1 NA NA NA 13 24
2 2 11 30 41 13 24
2 3 21 NA NA 13 24
2 4 24 35 21 13 24
2 5 NA NA NA 13 24
2 6 NA NA NA 13 24
3 0 NA NA NA 35 46
3 1 23 34 24 35 46
3 2 NA 26 NA 35 46
3 3 NA NA 24 35 46
3 4 12 29 42 35 46
3 5 NA NA NA 35 46
3 6 NA NA NA 35 46
In the table, each ID would represent a scenario that should be handled appropriately:
ID 1: Ordinary with observations starting from hour 0 and observation ending at hour 3 - and thus row with hour 3 and 4 for that group should be removed
ID 2: Has an hour (1) where all three observation variables are set at NA, but observation is resumed and ends at hour 5 - and thus row 2 should be kept (due to faulty registration and not end of observation) and rows with hour 5 and 6 should be removed.
ID 3: Starts out with an row with NA in all three observation variables, but observation begins then next hour and ends at hour 5. This is akin to the scenario for ID 2, but this time occurring at the very start (instead of in the middle of the observations). However, this still represent a faulty registration and should be kept and rows from hour 5 and 6 in this group should be removed.
Conceptually, I would think a possible solution would be do a group_by ID and then for R to go through the rows in a group in reverse (from bottom and up) until it encounters a row where "OBS_1", "OBS_2" and "OBS_3" are not all NA and remove the rows examined before reaching to this row and then move on to examine the next group.
Any help would be greatly appreciated!
If your MISC and MISC_2 values are consistent for each ID, you could
filter all rows that have na values then fill back in the missing data with complete and fill.
library(dplyr)
library(tidyr)
df %>% filter(!(is.na(OBS_1)&is.na(OBS_2)&is.na(OBS_3))) %>%
group_by(ID) %>%
complete(HOUR=0:max(HOUR)) %>%
fill(MISC,MISC_2) %>% fill(MISC,MISC_2,.direction = "up")
# A tibble: 13 x 7
# Groups: ID [3]
# ID HOUR OBS_1 OBS_2 OBS_3 MISC MISC_2
# <int> <int> <int> <int> <int> <int> <int>
# 1 1 0 29 32 34 19 21
# 2 1 1 21 12 NA 19 21
# 3 1 2 NA 24 NA 19 21
# 4 2 0 41 16 21 13 24
# 5 2 1 NA NA NA 13 24
# 6 2 2 11 30 41 13 24
# 7 2 3 21 NA NA 13 24
# 8 2 4 24 35 21 13 24
# 9 3 0 NA NA NA 35 46
# 10 3 1 23 34 24 35 46
# 11 3 2 NA 26 NA 35 46
# 12 3 3 NA NA 24 35 46
# 13 3 4 12 29 42 35 46
This filters only the missing values if the no observation for the day are existing after this and keeps all missing observations that do not indicate the end of the observations for the day. These also allow for your other variables to vary during the day because it just removes them if the end of observations is reached.
df %>% arrange(rev(as.numeric(rownames(.)))) %>%
group_by(ID) %>%
mutate(rowNum = 1:n(),
naObs = cumsum((is.na(OBS_1) & is.na(OBS_2) & is.na(OBS_3))),
missingBlock = naObs != rowNum) %>%
slice(min(which(missingBlock)):n()) %>%
ungroup() %>%
arrange(rev(as.numeric(rownames(.)))) %>%
select(-rowNum, -naObs, -missingBlock)
Related
replicate <- c(1,1,1,1,2,2,2,2,2,2,2)
level <- c("low","low","mid","mid","low","low","mid","mid","up","up","up")
high.density <- c(14,54,82,24,12,11,12,NA,40,NA,20)
low.density <- c(36,31,10,NA,28,45,17,24,10,5,2)
df <- data.frame(replicate, level, high.density, low.density)
The data looks like the following
replicate level high.density low.density
1 low 14 36
1 low 54 31
1 mid 82 10
1 mid 24 NA
2 low 12 28
2 low 11 45
2 mid 12 17
2 mid NA 24
2 up 40 10
2 up NA 5
2 up 20 2
I am trying to create data-sets where I make various combinations between high.density and low.density with NAs. Rule: I can only have combinations for same replicate and level. So each data-set will be unique in those positions. For instance,
replicate level high.density low.density
1 low 14 36
1 low 54 31
1 mid 82 NA #only 1 change
1 mid 24 10 #only 1 change
2 low 12 28
2 low 11 45
2 mid 12 17
2 mid NA 24
2 up 40 10
2 up NA 5
Next data-set
replicate level high.density low.density
1 low 14 36
1 low 54 31
1 mid 82 10 #original dataset
1 mid 24 NA #original dataset
2 low 12 28
2 low 11 45
2 mid NA 17 #only 1 change
2 mid 12 24 #only 1 change
2 up 40 10
2 up NA 5
2 up 20 2
Next iteration
replicate level high.density low.density
1 low 14 36
1 low 54 31
1 mid 82 NA #change here
1 mid 24 10 #change here
2 low 12 28
2 low 11 45
2 mid NA 17 #change here
2 mid 12 24 #change here
2 up 40 10
2 up NA 5
2 up 20 2
Edit: For replicate 2, "up" the iterations should look like the following. Only two combinations are possible.
2 up NA 10
2 up 40 5
2 up 20 2
2 up 40 10
2 up 20 5
2 up NA 2
Even though I could not manage to get the thing running myself, I think the for loops should be of the following form:
for(i in seq_len(dim(df)[1])) {
for(j in seq_len(length(df$replicate))) {
for(k in seq_len(length(df$level))) {
if(!is.na(df$high.density)) | is.na(df$low.density))
}
}
}
Could anyone provide some direction?
I want to take differences for each pair of consecutive columns but for an arbitrary number of columns. For example...
df <- as.tibble(data.frame(group = rep(c("a", "b", "c"), each = 4),
subgroup = rep(c("adam", "boy", "charles", "david"), times = 3),
iter1 = 1:12,
iter2 = c(13:22, NA, 24),
iter3 = c(25:35, NA)))
I want to calculate the differences by column. I would normally use...
df %>%
mutate(diff_iter2 = iter2 - iter1,
diff_iter3 = iter3 - iter2)
But... I'd like to:
accomodate an arbitrary number of columns and
treat NAs such that:
if the number we're subtracting from is NA, then the result should be NA. E.g. NA - 11 = NA
if the number we're subtracting is NA, then that NA is effectively treated as a 0. E.g. 35 - NA = 35
The result should look like this...
group subgroup iter1 iter2 iter3 diff_iter2 diff_iter3
<chr> <chr> <int> <dbl> <int> <dbl> <dbl>
1 a adam 1 13 25 12 12
2 a boy 2 14 26 12 12
3 a charles 3 15 27 12 12
4 a david 4 16 28 12 12
5 b adam 5 17 29 12 12
6 b boy 6 18 30 12 12
7 b charles 7 19 31 12 12
8 b david 8 20 32 12 12
9 c adam 9 21 33 12 12
10 c boy 10 22 34 12 12
11 c charles 11 NA 35 NA 35
12 c david 12 24 NA 12 NA
Originally, this df was in long format but the problem was that I believe the lag() function operates on position within groups and all the groups aren't the same because some have missing records (hence the NA in the wider table shown above).
Starting with long format would do but then please assume the records shown above with NA values would not exist in that longer dataframe.
Any help is appreciated.
An option in tidyverse would be - loop across the columns of 'iter' other than the iter1, then get the column value by replacing the column name (cur_column()) substring by subtracting 1 (as.numeric(x) -1) with str_replace, then replace the NA elements with 0 (replace_na) based on the OP's logic, subtract from the looped column and create new columns by adding prefix in .names ("diff_{.col}" - {.col} will be the original column name)
library(dplyr)
library(stringr)
library(tidyr)
df <- df %>%
mutate(across(iter2:iter3, ~
. - replace_na(get(str_replace(cur_column(), '\\d+',
function(x) as.numeric(x) - 1)), 0), .names = 'diff_{.col}'))
-output
df
# A tibble: 12 × 7
group subgroup iter1 iter2 iter3 diff_iter2 diff_iter3
<chr> <chr> <int> <dbl> <int> <dbl> <dbl>
1 a adam 1 13 25 12 12
2 a boy 2 14 26 12 12
3 a charles 3 15 27 12 12
4 a david 4 16 28 12 12
5 b adam 5 17 29 12 12
6 b boy 6 18 30 12 12
7 b charles 7 19 31 12 12
8 b david 8 20 32 12 12
9 c adam 9 21 33 12 12
10 c boy 10 22 34 12 12
11 c charles 11 NA 35 NA 35
12 c david 12 24 NA 12 NA
Find the columns whose names start with iter, ix, and then take all but the first as df1, all but the last as df2 and replace the NAs in df2 with 0. Then subtract them and cbind df to that. No packages are used.
ix <- grep("^iter", names(df))
df1 <- df[tail(ix, -1)]
df2 <- df[head(ix, -1)]
df2[is.na(df2)] <- 0
cbind(df, diff = df1 - df2)
giving:
group subgroup iter1 iter2 iter3 diff.iter2 diff.iter3
1 a adam 1 13 25 12 12
2 a boy 2 14 26 12 12
3 a charles 3 15 27 12 12
4 a david 4 16 28 12 12
5 b adam 5 17 29 12 12
6 b boy 6 18 30 12 12
7 b charles 7 19 31 12 12
8 b david 8 20 32 12 12
9 c adam 9 21 33 12 12
10 c boy 10 22 34 12 12
11 c charles 11 NA 35 NA 35
12 c david 12 24 NA 12 NA
I have two dataframes of this format.
df1:
id x y
1 2 3
2 4 5
3 6 7
4 8 9
5 1 1
df2:
id id2 v v2
1 t 11 21
1 b 12 22
2 t 13 23
2 b 14 24
3 t 15 25
3 b 16 26
4 b 17 27
Hence, sometimes, the id in main 'df' will appear twice (maximum) sometimes once, and sometimes not at all. The expected result would be:
df_merged:
id x y v.t v2.t v.b v2.b
1 2 3 11 21 12 22
2 4 5 13 23 24 24
3 6 7 15 25 16 26
4 8 9 NA NA 17 27
5 1 1 NA NA NA NA
I have used merge but due to the fact that id2 in df2 doesn't match, I get two instances of id in df_merged like so:
id x y v v2
1 ...
1 ...
Thanks in advance!
We can start by adjusting df2 to the right format then do a normal joining.
librar(dplyr)
library(tidyr)
df2 %>% gather(key,val,-id,-id2) %>% #Transfer from wide to long format for v and v2
mutate(new_key=paste0(key,'.',id2)) %>% #Create a new id2 as new_key
select(-id2,-key) %>% #de-select the unnessary columns
spread(new_key,val) %>% #Transfer back to wide foramt with right foramt for id
right_join(df1) %>% #right join df1 "To includes all rows in df1" using id
select(id,x,y,v.t,v2.t,v.b,v2.b) #rearrange columns name
Joining, by = "id"
id x y v.t v2.t v.b v2.b
1 1 2 3 11 21 12 22
2 2 4 5 13 23 14 24
3 3 6 7 15 25 16 26
4 4 8 9 NA NA 17 27
5 5 1 1 NA NA NA NA
You can solve this just using merge. Split df2 based on whether id2 equals b or t. Merge these two new objects with df1, and finally merge them together. The code includes one additional step to also include data found in df1 but not df2.
dfb <- merge(df1, df2[df2$id2=='b',], by='id')
dft <- merge(df1, df2[df2$id2=='t',], by='id')
dfRest <- df1[!df1$id %in% df2$id,]
dfAll <- merge(dfb[,c('id','x','y','v','v2')], dft[,c('id','v','v2')], by='id', all.x=T)
merge(dfAll, dfRest, all.x=T, all.y=T)
id x y v.x v2.x v.y v2.y
1 1 2 3 12 22 11 21
2 2 4 5 14 24 13 23
3 3 6 7 16 26 15 25
4 4 8 9 17 27 NA NA
5 5 1 1 NA NA NA NA
I have data for spot price and day-ahead price for hour 2 and hour 3. They are as below. They are from 2015-12-31 to 2011-01-01 all the way down.
> head(da2)
Date Price Hour
43802 2015-12-31 12.56 2
43778 2015-12-30 23.59 2
43754 2015-12-29 17.07 2
> head(sp2)
# A tibble: 6 x 3
Date Hour Price
<dttm> <chr> <dbl>
1 2015-12-31 2 17.15
2 2015-12-30 2 26.23
3 2015-12-29 2 23.01
> head(da3)
Date Price Hour
43803 2015-12-31 10.46 3
43779 2015-12-30 23.55 3
43755 2015-12-29 16.52 3
> head(sp3)
# A tibble: 6 x 3
Date Hour Price
<dttm> <chr> <dbl>
1 2015-12-31 3 12.96
2 2015-12-30 3 25.65
3 2015-12-29 3 23.59
I tried to put da2$Price and sp2$Price together, and again the same for hour 3.
But unfortunately, I get this.
> rpdf2<-data.frame(da2$Date,da2$Price,sp2$Price)
Error in data.frame(da2$Date, da2$Price, sp2$Price) :
arguments imply differing number of rows: 1826, 1822
> rpdf3<-data.frame(da3$Date,da3$Price,sp3$Price)
Error in data.frame(da3$Date, da3$Price, sp3$Price) :
arguments imply differing number of rows: 1821, 1825
So I applied > setdiff(paste(da2$Date),paste(sp2$Date))
Then I found
[1] "2014-03-30" "2013-03-31" "2012-03-25" "2011-03-27"
It was okay. But when I did setdiff(paste(da3$Date),paste(sp3$Date)), It shows me character(0).
There must be 4 observations difference. But I cannot find those four. Can anyone help me with this situation? Thank you.
When setdiff(da3$Date,sp3$Date)
result is
[1] 16800.04 16799.04 16798.04 16797.04 16796.04 16795.04 16794.04 16793.04 16792.04 16791.04 16790.04 16789.04 16788.04 16787.04 16786.04 16785.04 16784.04
[18] 16783.04 16782.04 16781.04 16780.04 16779.04 16778.04 16777.04 16776.04 16775.04 16774.04 16773.04 16772.04 16771.04 16770.04 16769.04 16768.04 16767.04
[35] 16766.04 16765.04 16764.04 16763.04 16762.04 16761.04 16760.04 16759.04 16758.04 16757.04 16756.04 16755.04 16754.04 16753.04 16752.04 16751.04 16750.04
[52] 16749.04 16748.04 16747.04 16746.04 16745.04 16744.04 16743.04 16742.04 16741.04 16740.04 16739.04 16738.04 16737.04 16736.04 16735.04 16734.04 16733.04
[69] 16732.04 16731.04 16730.04 16729.04 16728.04 16727.04 16726.04 16725.04 16724.04 16723.04 16722.04 16721.04 16720.04 16719.04 16718.04 16717.04 16716.04
[86] 16715.04 16714.04 16713.04 16712.04 16711.04 16710.04 16709.04 16708.04 16707.04 16706.04 16705.04 16704.04 16703.04 16702.04 16701.04 16700.04 16699.04
and so further.
One way (of many) to tackle this is instead of looking directly for the differences is to find a way to join your tables which will work regardless. To do so you simply need to generate a complete sequence of all dates from the first date on your list to the last, then left-join these to each of your daily and spot price data frames in turn. Missing date rows in each table will show as NA columns in the resulting joined table.
Example sequence, shortened to one month only for this exemplar. You'd start it at 2011-01-01 instead.
somedates = seq(as.Date("2015-12-01"), as.Date("2015-12-31"), by = "day")
Generate some test data each with four randomly missed dates to simulate your da2, da3, sp2 and sp3 tables:
library(dplyr)
set.seed(0)
da2 = data.frame(Date = sample(somedates, 27)) %>%
mutate(hour = 2, price = 20)
set.seed(1)
da3 = data.frame(Date = sample(somedates, 27)) %>%
mutate(hour = 3, price = 21)
set.seed(2)
sp2 = data.frame(Date = sample(somedates, 27)) %>%
mutate(hour = 2, price = 19)
set.seed(3)
sp3 = data.frame(Date = sample(somedates, 27)) %>%
mutate(hour = 3, price = 18)
Joining the da2, da3, sp2 and sp3 tables
With the test data generated, joining the tables to the complete sequence of dates (as a data frame) is straightforward. (NB I haven't replaced the joined column names with more meaningful versions in the result below).
all =
left_join(data.frame(Date = somedates), da2, by = "Date") %>%
left_join(da3, by = "Date") %>%
left_join(sp2, by = "Date") %>%
left_join(sp3, by = "Date")
Results from the test data joined
>all
Date hour.x price.x hour.y price.y hour.x.x price.x.x hour.y.y price.y.y
1 2015-12-01 2 20 3 21 2 19 3 18
2 2015-12-02 2 20 3 21 2 19 3 18
3 2015-12-03 NA NA 3 21 2 19 3 18
4 2015-12-04 2 20 3 21 2 19 3 18
5 2015-12-05 2 20 3 21 2 19 3 18
6 2015-12-06 2 20 3 21 2 19 3 18
7 2015-12-07 2 20 3 21 2 19 NA NA
8 2015-12-08 2 20 3 21 2 19 3 18
9 2015-12-09 2 20 3 21 NA NA 3 18
10 2015-12-10 2 20 3 21 NA NA 3 18
11 2015-12-11 2 20 3 21 2 19 3 18
12 2015-12-12 NA NA 3 21 2 19 3 18
13 2015-12-13 2 20 NA NA 2 19 NA NA
14 2015-12-14 2 20 3 21 2 19 3 18
15 2015-12-15 2 20 3 21 2 19 3 18
16 2015-12-16 2 20 3 21 2 19 3 18
17 2015-12-17 2 20 3 21 2 19 3 18
18 2015-12-18 2 20 NA NA 2 19 3 18
19 2015-12-19 NA NA 3 21 2 19 3 18
20 2015-12-20 2 20 NA NA NA NA 3 18
21 2015-12-21 2 20 3 21 2 19 3 18
22 2015-12-22 2 20 3 21 2 19 3 18
23 2015-12-23 2 20 3 21 2 19 3 18
24 2015-12-24 2 20 3 21 2 19 NA NA
25 2015-12-25 2 20 3 21 2 19 3 18
26 2015-12-26 2 20 3 21 2 19 3 18
27 2015-12-27 2 20 3 21 2 19 3 18
28 2015-12-28 2 20 3 21 2 19 3 18
29 2015-12-29 2 20 3 21 2 19 3 18
30 2015-12-30 2 20 3 21 NA NA 3 18
31 2015-12-31 NA NA NA NA 2 19 NA NA
Edit I note the numeric dates you posted as a result of your set join have a 0.04 time component as well as the whole-number date. You will need to add this to the date sequence to get the join to work. I have now tested this and without adding the time component you'd have to convert each date to a whole number. This can be done fairly simply though:
da2$Date = trunc.Date(da2$Date, "days")
da3$Date = trunc.Date(da3$Date, "days")
sp2$Date = trunc.Date(sp2$Date, "days")
sp3$Date = trunc.Date(sp3$Date, "days")
You'd do this before the joins.
I have a data frame as follows
Identifier V1 Location V2
1 12 A 21
1 12 B 24
2 20 B 15
2 20 C 18
2 20 B 23
3 43 A 10
3 43 B 17
3 43 A 18
3 43 B 20
3 43 C 25
3 43 A 30
I’d like to re-cast it with a single row for each Identifier and one column for each value in the current location column. I don’t care about the data in V1 but I need the data in V2 and these will become the values in the new columns.
Note that for the Location column there are repeated values for Identifiers 2 and 3.
I ASSUME that the first task is to make the values in the Location column unique.
I used the following (the data frame is called “Test”)
L<-length(Test$Identifier)
for (i in 1:L)
{
temp<-Test$Location[Test$Identifier==i]
temp1<-make.unique(as.character(temp), sep="-")
levels(Test$Location)=c(levels(Test$Location),temp1)
Test$Location[Test$Identifier==i]=temp1
}
This produces
Identifier V1 Location V2
1 12 A 21
1 12 B 24
2 20 B 15
2 20 C 18
2 20 B-1 23
3 43 A 10
3 43 B 17
3 43 A-1 18
3 43 B-1 20
3 43 C 25
3 50 A-2 30
Then using
cast(Test, Identifier ~ Location)
gives
Identifier A B C B-1 A-1 A-2
1 21 24 NA NA NA NA
2 NA 15 18 23 NA NA
3 10 17 25 20 18 30
And this is more or less what I want.
My questions are
Is this the right way to handle the problem?
I know R-people don’t use the “for” construction so is there a more R-elegant (relegant?) way to do this? I should mention that the real data set has over 160,000 rows and starts with over 50 unique values in the Location vector and the function takes just over an hour to run. Anything quicker would be good. I should also mention that the cast function had to be run on 20-30k rows of the output at a time despite increasing the memory limit. All the cast outputs were then merged
Is there a way to sort the columns in the output so that (here) they are A, A-1, A-2, B, B-1, C
Please be gentle with your replies!
Usually your original format is much better than your desired result. However, you can do this easily using the split-apply-combine approach, e.g., with package plyr:
DF <- read.table(text="Identifier V1 Location V2
1 12 A 21
1 12 B 24
2 20 B 15
2 20 C 18
2 20 B 23
3 43 A 10
3 43 B 17
3 43 A 18
3 43 B 20
3 43 C 25
3 43 A 30", header=TRUE, stringsAsFactors=FALSE)
#note that I make sure that there are only characters and not factors
#use as.character if you have factors
library(plyr)
DF <- ddply(DF, .(Identifier), transform, Loc2 = make.unique(Location, sep="-"))
library(reshape2)
DFwide <- dcast(DF, Identifier ~Loc2, value.var="V2")
# Identifier A B B-1 C A-1 A-2
#1 1 21 24 NA NA NA NA
#2 2 NA 15 23 18 NA NA
#3 3 10 17 20 25 18 30
If column order is important to you (usually it isn't):
DFwide[, c(1, order(names(DFwide)[-1])+1)]
# Identifier A A-1 A-2 B B-1 C
#1 1 21 NA NA 24 NA NA
#2 2 NA NA NA 15 23 18
#3 3 10 18 30 17 20 25
For reference, here's the equivalent of #Roland's answer in base R.
Use ave to create the unique "Location" columns....
DF$Location <- with(DF, ave(Location, Identifier,
FUN = function(x) make.unique(x, sep = "-")))
... and reshape to change the structure of your data.
## If you want both V1 and V2 in your "wide" dataset
## "dcast" can't directly do this--you'll need `recast` if you
## wanted both columns, which first `melt`s and then `dcast`s....
reshape(DF, direction = "wide", idvar = "Identifier", timevar = "Location")
## If you only want V2, as you indicate in your question
reshape(DF, direction = "wide", idvar = "Identifier",
timevar = "Location", drop = "V1")
# Identifier V2.A V2.B V2.C V2.B-1 V2.A-1 V2.A-2
# 1 1 21 24 NA NA NA NA
# 3 2 NA 15 18 23 NA NA
# 6 3 10 17 25 20 18 30
Reordering the columns can be done the same way that #Roland suggested.