Related
say I have a dataframe
subject stim1 stim2 feedback
1 1003 50 51 1
2 1003 48 50 1
3 1003 49 51 1
4 1003 47 49 1
5 1003 47 46 1
6 1003 46 48 1
10 1003 50 48 1
428 1003 48 51 0
433 1003 46 50 0
434 1003 50 49 0
435 1003 54 59 0
I want to create a new column "transitive_pair" by
group by subject (column 1),
For every row in which feedback==0 (starting index 428, otherwise transitive_pair=NaN).
I want to return a boolean which tells me whether there is any chain of pairings (but only those in which feedback==1) that would transitively link stim1 and stim2 values.
Working out a few examples.
row 428- stim1=48 and stim2=51
48 and 51 are not paired but 51 was paired with 50 (e.g.row 1 ) and 50 was paired with 48 (row 10) so transitive_pair[428]=True
row 433- stim 1=46 and stim2=50
46 and 48 were paired (row 6) and 48 was paired with 50 (row 2) so transitive_pair[433]=True
in row 435, stim1=54, stim2=59
there is no chain of pairs that could link them (59 is not paired with anything while feedback==1) so transitive_pair[435]=False
desired output
subject stim1 stim2 feedback transitive_pair
1 1003 50 51 1 NaN
2 1003 48 50 1 NaN
3 1003 49 51 1 NaN
4 1003 47 49 1 NaN
5 1003 47 46 1 NaN
6 1003 46 48 1 NaN
10 1003 50 48 1 NaN
428 1003 48 51 0 1
433 1003 46 50 0 1
434 1003 50 49 0 1
435 1003 54 59 0 0
any help would be greatly appreciated!!
and putting a recreateble df here
structure(list(subject = c(1003L, 1003L, 1003L, 1003L, 1003L,
1003L, 1003L, 1003L, 1003L, 1003L, 1003L), stim1 = c(50L, 48L,
49L, 47L, 47L, 46L, 50L, 48L, 46L, 50L, 54L), stim2 = c(51L,
50L, 51L, 49L, 46L, 48L, 48L, 51L, 50L, 49L, 59L), feedback = c(1L,
1L, 1L, 1L, 1L, 1L, 1L, 0L, 0L, 0L, 0L), transitive_pair = c(NaN,
NaN, NaN, NaN, NaN, NaN, NaN, 1, 1, 1, 0)), row.names = c(1L,
2L, 3L, 4L, 5L, 6L, 10L, 428L, 433L, 434L, 435L), class = "data.frame")
The columns "stim1" and "stim2" define an undirected graph. Create the graph for feedback == 1, get its connected components and for each row of the data.frame, check if the values of "stim1" and "stim2" belong to the same component. In the end assign NaN to the rows where feedback is 1.
suppressPackageStartupMessages(library(igraph))
inx <- df1$feedback == 1
g <- graph_from_data_frame(df1[inx, c("stim1", "stim2")], directed = FALSE)
plot(g)
g_comp <- components(g)$membership
df1$transitive_pair_2 <- apply(df1[c("stim1", "stim2")], 1, \(x) {
i <- names(g_comp) == x[1]
j <- names(g_comp) == x[2]
if(any(i) & any(j))
g_comp[i] == g_comp[j]
else 0L
})
df1$transitive_pair_2[inx] <- NaN
df1
#> subject stim1 stim2 feedback transitive_pair transitive_pair_2
#> 1 1003 50 51 1 NaN NaN
#> 2 1003 48 50 1 NaN NaN
#> 3 1003 49 51 1 NaN NaN
#> 4 1003 47 49 1 NaN NaN
#> 5 1003 47 46 1 NaN NaN
#> 6 1003 46 48 1 NaN NaN
#> 10 1003 50 48 1 NaN NaN
#> 428 1003 48 51 0 1 1
#> 433 1003 46 50 0 1 1
#> 434 1003 50 49 0 1 1
#> 435 1003 54 59 0 0 0
Created on 2022-07-31 by the reprex package (v2.0.1)
I'm trying to plot the indels length from a file created by vcftools with the feature: "--hist-indel-len". With that file, I wanted to make a plot with the insertion and deletions, if length is negative, is a deletion and if length is positive, is a insertion. The Count column will be at y-axis from 0 to the max value, and the x-axis will be the min length (-15 in that case) to the max length (15 in that case).
The data looks like:
LENGTH COUNT
1 -15 117
2 -14 178
3 -13 198
4 -12 414
5 -11 314
6 -10 451
7 -9 547
8 -8 1114
9 -7 1214
10 -6 2371
11 -5 3822
12 -4 9229
13 -3 17333
14 -2 20373
15 -1 19774
16 0 202129
17 1 22259
18 2 10101
19 3 4940
20 4 2458
21 5 1343
22 6 987
23 7 535
24 8 427
25 9 317
26 10 307
27 11 161
28 12 270
29 13 116
30 14 121
31 15 95
With this data.frame I'm trying to get a plot like:
My attempt was using:
z <- read.csv("/home/userx/out.indel.hist", sep = "\t")
zz <- table(z)
barplot(zz, main="Insertion and Deletions",
xlab="Length", ylab="Count", col=c("darkblue","red"),
legend = rownames(zz), beside=TRUE)
Result:
Any help would be appreciated.
A relatively easy solution using ggplot and the provided data would be to create a grouping variable to color by and plot using geom_col:
library(tidyverse)
create grouping variable:
dat2 %>%
mutate(fill = ifelse(LENGTH <0, "minus", "plus")) -> dat2
ggplot(dat2)+
geom_col(aes(x = LENGTH, y = COUNT, fill = fill))
the data:
structure(list(LENGTH = -15:15, COUNT = c(117L, 178L, 198L, 414L,
314L, 451L, 547L, 1114L, 1214L, 2371L, 3822L, 9229L, 17333L,
20373L, 19774L, 202129L, 22259L, 10101L, 4940L, 2458L, 1343L,
987L, 535L, 427L, 317L, 307L, 161L, 270L, 116L, 121L, 95L)), .Names = c("LENGTH",
"COUNT"), class = "data.frame", row.names = c(NA, -31L))
I have a data frame with 3 ID variables, then several values for each ID.
user Log Pass Value
2 2 123 342
2 2 123 543
2 2 123 231
2 2 124 257
2 2 124 342
4 3 125 543
4 3 125 231
4 3 125 257
4 3 125 342
4 3 125 543
4 3 125 231
4 3 125 257
4 3 125 543
4 3 125 231
4 3 125 257
4 3 125 543
4 3 125 231
4 3 125 257
4 3 125 543
4 3 125 231
4 3 125 257
The start and end of each set of values is sometimes noisy, and I want to be able to delete the first few values. Unfortunately the number of values varies significantly, but it is always the first and last 20% of values that are noisy.
I want to delete the first 20% of rows, with a minimum of 1 row deleted.
So for instance if there are 20 values for user 2 log 2 pass 123 I want to delete the first and last 4 rows. If there are only 3 values for the ID variable I want to delete the first and last row.
The resulting dataset would be:
user Log Pass Value
2 2 123 543
4 3 125 543
4 3 125 231
4 3 125 257
4 3 125 543
4 3 125 231
4 3 125 257
4 3 125 543
4 3 125 231
I've tried fiddling around with nrow but I struggle to figure out how to reference the % of rows by id variable.
Thanks.
Jonathan.
I believe the following can do it.
DATA.
dat <-
structure(list(user = c(2L, 2L, 2L, 2L, 2L, 4L, 4L, 4L, 4L, 4L,
4L, 4L, 4L, 4L, 4L, 4L, 4L, 4L, 4L, 4L, 4L), Log = c(2L, 2L,
2L, 2L, 2L, 3L, 3L, 3L, 3L, 3L, 3L, 3L, 3L, 3L, 3L, 3L, 3L, 3L,
3L, 3L, 3L), Pass = c(123L, 123L, 123L, 124L, 124L, 125L, 125L,
125L, 125L, 125L, 125L, 125L, 125L, 125L, 125L, 125L, 125L, 125L,
125L, 125L, 125L), Value = c(342L, 543L, 231L, 257L, 342L, 543L,
231L, 257L, 342L, 543L, 231L, 257L, 543L, 231L, 257L, 543L, 231L,
257L, 543L, 231L, 257L)), .Names = c("user", "Log", "Pass", "Value"
), class = "data.frame", row.names = c(NA, -21L))
CODE.
fun <- function(x, p = 0.20){
n <- nrow(x)
m <- max(1, round(n*p))
inx <- c(seq_len(m), n - seq_len(m) + 1)
x[-inx, ]
}
result <- do.call(rbind, lapply(split(dat, dat$user), fun))
row.names(result) <- NULL
result
# user Log Pass Value
#1 2 2 123 543
#2 2 2 123 231
#3 2 2 124 257
#4 4 3 125 342
#5 4 3 125 543
#6 4 3 125 231
#7 4 3 125 257
#8 4 3 125 543
#9 4 3 125 231
#10 4 3 125 257
#11 4 3 125 543
#12 4 3 125 231
#13 4 3 125 257
Would something like this help?
For a dataframe df:
df[-c(1:floor(nrow(df)*0.2), (1+ceiling(nrow(df)*0.8)):nrow(df)),]
Just removing the first and last 20%, taking the upper and lower values so that for smaller data frame you keep some of the information:
> df<-data.frame(a=1:100)
> df[-c(1:floor(nrow(df)*0.2),(1+ceiling(nrow(df)*0.8)):nrow(df)),]
[1] 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50
[31] 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
> df<-data.frame(1:3)
> df[-c(1:floor(nrow(df)*0.2),(1+ceiling(nrow(df)*0.8)):nrow(df)),]
[1] 2
You can do this with dplyr...
library(dplyr)
df2 <- df %>% group_by(user, Log, Pass) %>%
filter(n()>2) %>% #remove those with just two elements or fewer
slice(max(2, 1+ceiling(n()*0.2)):min(n()-1, floor(0.8*n())))
df2
user Log Pass Value
1 2 2 123 543
2 4 3 125 543
3 4 3 125 231
4 4 3 125 257
5 4 3 125 543
6 4 3 125 231
7 4 3 125 257
8 4 3 125 543
9 4 3 125 231
Calculate the offset for what you want to retain:
rem <- ceiling( nrow( x ) * .2 ) + 1
Then take out the records you don-t want:
dat <- dat[ rem : ( nrow( dat ) - rem ), ]
Here is an idea using base R that returns the row indices of each user to keep and then subsets on these indices.
idx <- unlist(lapply(split(seq_along(dat[["user"]]), dat[["user"]]), function(x) {
tmp <- max(1, ceiling(.2 * length(x)))
tail(head(x, -tmp), -tmp)}),
use.names=FALSE)
split(seq_along(dat[["user"]]), dat[["user"]]) returns a list of the rows for each user. lapply loops through these rows, calculating the number of rows to drop from each end with split(seq_along(dat[["user"]]), dat[["user"]]), and then dropping them with tail(head(x, -tmp), -tmp)}). Since lapply returns a named list, this is unlisted and the names are dropped.
This returns
idx
2 3 4 10 11 12 13 14 15 16 17
Now subset
dat[idx,]
user Log Pass Value
2 2 2 123 543
3 2 2 123 231
4 2 2 124 257
10 4 3 125 543
11 4 3 125 231
12 4 3 125 257
13 4 3 125 543
14 4 3 125 231
15 4 3 125 257
16 4 3 125 543
17 4 3 125 231
This question already has answers here:
Reshaping multiple sets of measurement columns (wide format) into single columns (long format)
(8 answers)
Closed 5 years ago.
I'd really appreciate some help getting this messy set of new survey data into a usable form. It was collected in a strange way and now I've got strange data to work with. I've looked through tidyr and used those approaches to no end. I suspect my problem is that I'm thinking about this dataset all wrong and I'm blind to some real answer. But given all the things I need to do to this df, I cant figure out where to start and thus where to start googling.
What I need:
For each person to be their own row
Each person retains their GroupID and Treated value
For the variables currently attached to each person individually to become columns (age, weight, height)
Fake (and much smaller):
structure(list(GroupID = 1:5, Treated = c("Y", "Y", "N", "Y",
"N"), person1_age = c(45L, 33L, 71L, 19L, 52L), person1_weight = c(187L,
145L, 136L, 201L, 168L), person1_height = c(69L, 64L, 51L, 70L,
66L), person2_age = c(54L, 20L, 48L, 63L, 26L), person2_weight = c(140L,
122L, 186L, 160L, 232L), person2_height = c(62L, 70L, 65L, 72L,
74L), person3_age = c(21L, 56L, 40L, 59L, 67L), person3_weight = c(112L,
143L, 187L, 194L, 159L), person3_height = c(61L, 69L, 73L, 63L,
72L)), .Names = c("GroupID", "Treated", "person1_age", "person1_weight",
"person1_height", "person2_age", "person2_weight", "person2_height",
"person3_age", "person3_weight", "person3_height"), row.names = c(NA,
5L), class = "data.frame")
Any help or further readings you could point me to would be very much appreciated.
reshape can do this, with the appropriate arguments:
> reshape(x, direction="long", varying=names(x)[3:11], timevar='person', v.names=c('height', 'age', 'weight'), sep='_')
GroupID Treated person height age weight id
1.1 1 Y 1 187 45 69 1
2.1 2 Y 1 145 33 64 2
3.1 3 N 1 136 71 51 3
4.1 4 Y 1 201 19 70 4
5.1 5 N 1 168 52 66 5
1.2 1 Y 2 140 54 62 1
2.2 2 Y 2 122 20 70 2
3.2 3 N 2 186 48 65 3
4.2 4 Y 2 160 63 72 4
5.2 5 N 2 232 26 74 5
1.3 1 Y 3 112 21 61 1
2.3 2 Y 3 143 56 69 2
3.3 3 N 3 187 40 73 3
4.3 4 Y 3 194 59 63 4
5.3 5 N 3 159 67 72 5
This relies on the order of the columns in your original data, for the varying argument, being in increasing order in the original data.
If that's not the case, specify varying manually. Here's what is used above:
> names(x)[3:11]
[1] "person1_age" "person1_weight" "person1_height" "person2_age" "person2_weight" "person2_height"
[7] "person3_age" "person3_weight" "person3_height"
We can also use melt from data.table which can take multiple patterns in the measure argument
library(data.table)
melt(setDT(x), measure = patterns("age$", "weight$", "height$"),
variable.name = "person", value.name = c("age", "weight", "height"))
# GroupID Treated person age weight height
# 1: 1 Y 1 45 187 69
# 2: 2 Y 1 33 145 64
# 3: 3 N 1 71 136 51
# 4: 4 Y 1 19 201 70
# 5: 5 N 1 52 168 66
# 6: 1 Y 2 54 140 62
# 7: 2 Y 2 20 122 70
# 8: 3 N 2 48 186 65
# 9: 4 Y 2 63 160 72
#10: 5 N 2 26 232 74
#11: 1 Y 3 21 112 61
#12: 2 Y 3 56 143 69
#13: 3 N 3 40 187 73
#14: 4 Y 3 59 194 63
#15: 5 N 3 67 159 72
I have a data table that repeats records. I would like to transpose the table but into the unique record names.
Below is a sample of the Data table:
V1 V2 id
ClientID 29 1
CheckID 201 1
PaymentAmount 256 1
Gross 301 1
Net 256 1
Invested 130 1
Invested 53 1
Invested 118 1
ClientID 31 2
CheckID 222 2
PaymentAmount 41 2
Gross 46 2
Net 41 2
Invested 46 2
ClientID 43 3
CheckID 310 3
PaymentAmount 41 3
Gross 46 3
Net 41 3
Invested 46 3
You can see from the table above that the record in X1 called "Investment" can occur more than once for a single ClientID. I'd like to transpose the data so that it looks as such:
ClientID CheckID PaymentAmount Gross Net Invested ID
29 201 256 301 256 130 1
29 201 256 301 256 53 1
29 201 256 301 256 118 1
31 222 41 46 41 46 2
43 310 41 46 41 46 3
43 310 41 46 41 48 3
any support is greatly appreciated!
We can create a sequence column grouped by the "V1", "id" column using data.table, then convert from 'long' to 'wide' format with dcast and replace the NA with the non-NA preceding values using na.locf from zoo.
library(data.table)
library(zoo)
setDT(df1)[, N:= 1:.N , by = .(V1, id)]
dcast(df1, id+N~V1, value.var="V2")[, lapply(.SD, na.locf),
by = id, .SDcols = CheckID:PaymentAmount]
# id CheckID ClientID Gross Invested Net PaymentAmount
#1: 1 201 29 301 130 256 256
#2: 1 201 29 301 53 256 256
#3: 1 201 29 301 118 256 256
#4: 2 222 31 46 46 41 41
#5: 3 310 43 46 46 41 41
data
df1 <- structure(list(V1 = c("ClientID", "CheckID", "PaymentAmount",
"Gross", "Net", "Invested", "Invested", "Invested", "ClientID",
"CheckID", "PaymentAmount", "Gross", "Net", "Invested", "ClientID",
"CheckID", "PaymentAmount", "Gross", "Net", "Invested"), V2 = c(29L,
201L, 256L, 301L, 256L, 130L, 53L, 118L, 31L, 222L, 41L, 46L,
41L, 46L, 43L, 310L, 41L, 46L, 41L, 46L), id = c(1L, 1L, 1L,
1L, 1L, 1L, 1L, 1L, 2L, 2L, 2L, 2L, 2L, 2L, 3L, 3L, 3L, 3L, 3L,
3L)), .Names = c("V1", "V2", "id"), class = "data.frame",
row.names = c(NA, -20L))