I have a data-set in which each column is a variable and each row is an observation (like time series data. It looks like this (I apologize for the format, but I can't show the data):
I'd like to know if a person or group is saying the same thing(s) over time. I'm familiar with n-grams, but it's not quite what I need. Any help would be appreciated.
This is the output I'd like:
Sorry for all the edits poor comments; still getting used to the website.
If you want to see the frequence of each comments related to each Person and a new column Ready you can do this with the following code :
set.seed(123456)
### I use the same data as the previous example, thank you for providing this !
data <-data.frame(date = Sys.Date() - sample(100),
Group = c("Cars","Trucks") %>% sample(100,replace=T),
Reporting_person = c("A","B","C") %>% sample(100,replace=T),
Comments = c("Awesome","Meh","NC") %>% sample(100,replace=T),
Ready = as.character(c("Yes","No") %>% sample(100,replace=T))
)
library(dplyr)
data %>%
group_by(Reporting_person,Ready) %>%
count(Comments) %>%
mutate(prop = prop.table(n))
If what you are asking is to see if a change occurs in the comments over time and to see if that change is correlated with an event (like Ready) you can use the following code:
library(dplyr)
### Creating a column comments at time + plus
new = data %>%
arrange(Reporting_person,Group,date) %>%
group_by(Group,Reporting_person) %>%
mutate(comments_plusone=lag(Comments))
new = na.omit(new)
### Creating the change column 1 is a change , 0 no change
new$Change = as.numeric(new$Comments != new$comments_plusone)
### Get the correlation between Change and the events...
### Chi-test to test if correlation between the event and the change
### Not that using Pearson correlation is not pertinent here :
tbl <- table(new$Ready,new$Change)
chi2 = chisq.test(tbl, correct=F)
c(chi2$statistic, chi2$p.value)
sqrt(chi2$statistic / sum(tbl))
You should get no significative correlation with this example. As you can clearly see when you illustrate the table.
plot(tbl)
Not that using cor function is not appropriate working with two binary variable.
Here a post in this topic.... Correlation between two binary
Frequence of change by change of State
Following your comments, I am adding this code:
newR = data %>%
arrange(Reporting_person,Group,date) %>%
group_by(Group,Reporting_person) %>%
mutate(Ready_plusone=lag(Ready))
newR = na.omit(newR)
###------------------------Add the column to the new data frame
### Creating the REady change column 1 is a change , 0 no change
### Creating the change of state , I use this because you seem to have more than 2 levels.
new$State_change = paste(newR$Ready,newR$Ready_plusone,sep="_")
### Getting the frequency of Change by Change of State(Ready Yes-no..no-yes..)
result <- new %>%
group_by(Reporting_person,State_change) %>%
count(Change) %>%
mutate(Frequence = prop.table(n))%>%
filter(Change==1)
### Tidyr is a great library for reshape data, you want the wide format of the previous long
### dataframe... However doing this will generate a lot of NA so If I were you I would get
### the result format instead of the following but this could be helpful for future need so here you go.
library(tidyr)
final = as.data.frame(spread(result, key = State_change, value = prop))[,c(1,4:7)]
Hope this help :)
Something like this ?
df <-data.frame(date = Sys.Date() - sample(10),
Group = c("Cars","Trucks") %>% sample(10,replace=T),
Reporting_person = c("A","B","C") %>% sample(10,replace=T),
Comments = c("Awesome","Meh","NC") %>% sample(10,replace=T))
# date Group Reporting_person Comments
# 1 2017-06-08 Trucks B Awesome
# 2 2017-06-05 Trucks A Awesome
# 3 2017-06-14 Cars B Meh
# 4 2017-06-06 Cars B Awesome
# 5 2017-06-11 Cars A Meh
# 6 2017-06-07 Cars B NC
# 7 2017-06-09 Cars A NC
# 8 2017-06-10 Cars A NC
# 9 2017-06-13 Trucks C Awesome
# 10 2017-06-12 Trucks B NC
aggregate(date ~ .,df,length)
# Group Reporting_person Comments date
# 1 Trucks A Awesome 1
# 2 Cars B Awesome 1
# 3 Trucks B Awesome 1
# 4 Trucks C Awesome 1
# 5 Cars A Meh 1
# 6 Cars B Meh 1
# 7 Cars A NC 2
# 8 Cars B NC 1
# 9 Trucks B NC 1
Related
I'm having a hard time making rounded percentages that add up to 100% within groups.
Consider the following example:
# Loading main library used
library(dplyr)
# Creating the basic data frame
df = data.frame(group = c('A','A','A','A','B','B','B','B'),
categories = c('Cat1','Cat2','Cat3','Cat4','Cat1','Cat2','Cat3','Cat4'),
values = c(2200,4700,3000,2000,2900,4400,2200,1000))
print(df)
# group categories values
# 1 A Cat1 2200
# 2 A Cat2 4700
# 3 A Cat3 3000
# 4 A Cat4 2000
# 5 B Cat1 2900
# 6 B Cat2 4400
# 7 B Cat3 2200
# 8 B Cat4 1000
df_with_shares = df %>%
# Calculating group totals and adding them back to the main df
left_join(df %>%
group_by(group) %>%
summarize(group_total = sum(values)),
by='group') %>%
# Calculating each category's share within the groups
mutate(group_share = values / group_total,
group_share_rounded = round(group_share,2))
# Summing the rounded shares within groups
rounded_totals = df_with_shares %>%
group_by(group) %>%
summarize(total_share = sum(group_share_rounded))
print(rounded_totals)
# # A tibble: 2 x 2
# group total_share
# <chr> <int>
# 1 A 0.99
# 2 B 1.01
# Note how the totals do not add up to 100% as expected
I am aware of a few generic solutions to the "rounding percentages to add up to 100%" problem, as explained in this SO post. I was even able to make a little R implementation of one of those approaches, as seen here. This is what it would look like if I just applied that R approach to this problem:
df_with_rounded_shares = df %>%
mutate(
percs = values / sum(values),
percs_cumsum = cumsum(percs),
percs_cumsum_round = round(percs_cumsum, 2),
percs_cumsum_round_offset = replace_na(lag(percs_cumsum_round,1),0),
percs_rounded_final = percs_cumsum_round - percs_cumsum_round_offset)
However, the method I devised in the thread above does not work as I would like. It just calculates the shares of the values column across the whole dataset. In other words, it does not take into consideration the grouping variable representing the multiple groups in the data, each of which need their rounded values to add up to 100% independently from every other group.
What can I do to generate a column of rounded percentages that add up to 100% by group?
PS: While writing this question I actually found something that worked, so I'll answer my own question below. I know it's super simple, but I think it's still worth having a direct answer here on SO addressing this issue.
The method devised in your implementation (from here) just needs a few small tweaks to make it work.
First, include a group_by statement before calculating the new columns. Also, you need to use a summarize statement instead of the mutate statement you have now.
In essence, this is what it'll look like:
# Modified version of your implementation of the rounding procedure.
# The new procedure below accommodates for grouping variables.
df_with_rounded_shares_by_group = df %>%
group_by(group) %>%
summarize(
group_share = values / sum(values),
group_share_cumsum = cumsum(group_share),
group_share_cumsum_round = round(group_share_cumsum, 2),
group_share_cumsum_round_offset = replace_na(lag(group_share_cumsum_round,1),0),
group_share_rounded_final = group_share_cumsum_round - group_share_cumsum_round_offset) %>%
# Removing unnecessary temporary columns
select(-group_share_cumsum, -group_share_cumsum_round, -group_share_cumsum_round_offset)
# Verifying if the results add up to 100% within each group
rounded_totals = df_with_rounded_shares_by_group %>%
group_by(group) %>%
summarize(total_share = sum(group_share_rounded_final))
print(rounded_totals)
# # A tibble: 2 x 2
# group total_share
# <chr> <dbl>
# 1 A 1
# 2 B 1
# Yep, they all add up to 100% as expected!
Btw, apologies for the ridiculously long column names. I just made them enormous to make it clear what each step was really doing.
I have a dataframe as below:
**df**
Cust_name time freq
Andrew 0 4
Dillain 1 2
Alma 2 3
Andrew 1 4
Kiko 2 1
Sarah 2 8
Sarah 0 3
I want to calculate the sum of frequency by the time range provided for each cust_name. Example: If I select time range 0 to 2 for Andrew, it will give me sum of freq: 4+4= 8. And for Sarah, it will give me 8+3=11. I have tried it in the following ways just to get the time range, but do not know how to do the rest, as I am very new to R:
df[(df$time>=0 & df$time<=2),]
You can do this with dplyr.
To make your code reproducible, you should add the creation of your dataframe in your post. Copy and pasting everything is time consuming.
library(dplyr)
df <- data.frame(
cust_name = c('Andrew', 'Dillain', 'Alma', 'Andrew', 'Kiko', 'Sarah', 'Sarah'),
time = c(0,1,2,1,2,2,0),
freq = c(4,2,3,4,1,8,3)
)
df %>%
filter(time >=0, time <=2) %>%
group_by(cust_name) %>%
summarise(sum_freq = sum(freq))
I'm new to R, so please go easy on me... I have some longitudinal data that looks like
Basically, I'm trying to find a way to get a table with a) the number of unique cases that have all complete data and b) the number of unique cases that have at least one incomplete or missing data. The end results would ideally be
df<- df %>% group_by(Location)
df1<- df %>% group_by(any(Completion_status=='Incomplete' | 'Missing'))
Not sure about what you want, because it seems there are something of inconsistent between your request and the desired output, however lets try, it seems you need a kind of frequency table, that you can manage with basic R. At the bottom of the answer you can find some data similar to yours.
# You have two cases, the Complete, and the other, so here a new column about it:
data$case <- ifelse(data$Completion_status =='Complete','Complete', 'MorIn')
# now a frequency table about them: if you want a data.frame, here we go
result <- as.data.frame.matrix(table(data$Location,data$case))
# now the location as a new column rather than the rownames
result$Location <- rownames(result)
# and lastly a data.frame with the final results: note that you can change the names
# of the columns but if you want spaces maybe a tibble is better
result <- data.frame(Location = result$Location,
`Number.complete` = result$Complete,
`Number.incomplete.missing` = result$MorIn)
result
Location Number.complete Number.incomplete.missing
1 London 0 1
2 Los Angeles 0 1
3 Paris 3 1
4 Phoenix 0 2
5 Toronto 1 1
Or if you prefere a dplyr chain:
data %>%
mutate(case = ifelse(data$Completion_status =='Complete','Complete', 'MorIn')) %>%
do( as.data.frame.matrix(table(.$Location,.$case))) %>%
mutate(Location = rownames(.)) %>%
select(3,1,2) %>%
`colnames<-`(c("Location","Number of complete ", "Number of incomplete or"))
Location Number of complete Number of incomplete or
1 London 0 1
2 Los Angeles 0 1
3 Paris 3 1
4 Phoenix 0 2
5 Toronto 1 1
With data:
# here your data (next time try to put them in an usable way in the question)
data <- data.frame( ID = c("A1","A1","A2","A2","B1","C1","C2","D1","D2","E1"),
Location = c('Paris','Paris','Paris','Paris','London','Toronto','Toronto','Phoenix','Phoenix','Los Angeles'),
Completion_status = c('Complete','Complete','Incomplete','Complete','Incomplete','Missing',
'Complete','Incomplete','Incomplete','Missing'))
I am trying to calculate the families sizes from a data frame, which also contains two types of events : family members who died, and those who left the family. I would like to take into account these two parameters in order to compute the actual family size.
Here is a reproductive example of my problem, with 3 families only :
family <- factor(rep(c("001","002","003"), c(10,8,15)), levels=c("001","002","003"), labels=c("001","002","003"), ordered=TRUE)
dead <- c(0,0,0,0,1,0,1,0,0,0,0,0,0,0,0,0,0,1,0,0,0,0,0,1,0,1,0,0,0,0,0,0,0)
left <- c(0,0,0,0,0,1,0,0,0,1,1,0,0,0,1,1,0,0,0,1,1,1,0,0,0,0,0,0,1,1,1,0,0)
DF <- data.frame(family, dead, left) ; DF
I could count N = total family members (in each family) in a second dataframe DF2, by simply using table()
DF2 <- with(DF, data.frame(table(family)))
colnames(DF2)[2] <- "N" ; DF2
family N
1 001 10
2 002 8
3 003 15
But i can not find a proper way to get the actual number of people (for example, creating a new variable N2 into DF2) , calculated by substracting to N the number of members who died or left the family. I suppose i have to relate the two dataframes DF and DF2 in a way. i have looked for other related questions in this site but could not find the right answer...
If anyone has a good idea, it would be great !
Thank you in advance..
Deni
Logic : First we want to group_by(family) and then calculate 2 numbers : i) total #obs in each group ii) subtract the sum(dead) + sum(left) from this total .
In dplyr package : n() helps us get the total #observations in each group
In data.table : .N does the same above job
library(dplyr)
DF %>% group_by(family) %>% summarise( total = n(), current = n()-sum(dead,left, na.rm = TRUE))
# family total current
# (fctr) (int) (dbl)
#1 001 10 6
#2 002 8 4
#3 003 15 7
library(data.table)
# setDT() is preferred if incase your data was a data.frame. else just DF.
setDT(DF)[, .(total = .N, current = .N - sum(dead, left, na.rm = TRUE)), by = family]
# family total current
#1: 001 10 6
#2: 002 8 4
#3: 003 15 7
Here is a base R option
do.call(data.frame, aggregate(dl~family, transform(DF, dl = dead + left),
FUN = function(x) c(total=length(x), current=length(x) - sum(x))))
Or a modified version is
transform(aggregate(. ~ family, transform(DF, total = 1,
current = dead + left)[c(1,4:5)], FUN = sum), current = total - current)
# family total current
#1 001 10 6
#2 002 8 4
#3 003 15 7
I finally found another which works fine (from another post), allowing to compute everything from the original DF table. This uses the ddply function :
DF <- ddply(DF,.(family),transform,total=length(family))
DF <- ddply(DF,.(family),transform,actual=length(family)-sum(dead=="1")-sum(left=="1"))
DF
Thanks a lot to everyone who helped ! Deni
Let's say I have:
Person Movie Rating
Sally Titanic 4
Bill Titanic 4
Rob Titanic 4
Sue Cars 8
Alex Cars **9**
Bob Cars 8
As you can see, there is a contradiction for Alex. All the same movies should have the same ranking, but there was a data error entry for Alex. How can I use R to solve this? I've been thinking about it for a while, but I can't figure it out. Do I have to just do it manually in excel or something? Is there a command on R that will return all the cases where there are data contradictions between two columns?
Perhaps I could have R do a boolean check if all the Movie cases match the first rating of its first iteration? For all that returns "no," I can go look at it manually? How would I write this function?
Thanks
Here's a data.table solution
Define the function
Myfunc <- function(x) {
temp <- table(x)
names(temp)[which.max(temp)]
}
library(data.table)
Create a column with the correct rating (by reference)
setDT(df)[, CorrectRating := Myfunc(Rating), Movie][]
# Person Movie Rating CorrectRating
# 1: Sally Titanic 4 4
# 2: Bill Titanic 4 4
# 3: Rob Titanic 4 4
# 4: Sue Cars 8 8
# 5: Alex Cars 9 8
# 6: Bob Cars 8 8
Or If you want to remove the "bad" ratings
df[Rating == CorrectRating][]
# Person Movie Rating CorrectRating
# 1: Sally Titanic 4 4
# 2: Bill Titanic 4 4
# 3: Rob Titanic 4 4
# 4: Sue Cars 8 8
# 5: Bob Cars 8 8
It looks like, within each group defined by "Movie", you're looking for any instances of Rating that are not the same as the most common value.
You can solve this using dplyr (which is good at "group by one column, then perform an operation within each group), along with the "Mode" function defined in this answer that finds the most common item in a vector:
Mode <- function(x) {
ux <- unique(x)
ux[which.max(tabulate(match(x, ux)))]
}
library(dplyr)
dat %>% group_by(Movie) %>% filter(Rating != Mode(Rating))
This finds all the cases where a row does not agree with the rest of the group. If you instead want to remove them, you can do:
newdat <- dat %>% group_by(Movie) %>% filter(Rating == Mode(Rating))
If you want to fix them, do
newdat <- dat %>% group_by(Movie) %>% mutate(Rating = Mode(Rating))
You can test the above with a reproducible version of your data:
dat <- data.frame(Person = c("Sally", "Bill", "Rob", "Sue", "Alex", "Bob"),
Movie = rep(c("Titanic", "Cars"), each = 3),
Rating = c(4, 4, 4, 8, 9, 8))
If the goal is to see if all the values within a group are the same (or if there are some differences) then this can be a simple application of tapply (or aggregate, etc.) used with a function like var (or compute the range). If all the values are the same then the variance and range will be 0. If it is any other value (outside of rounding error) then there must be a value that is different. The which function can help identify the group/individual.
tapply(dat$Rating, dat$Movie, FUN=var)
which(.Last.value > 0.00001)
tapply(dat$Rating, dat$Movie, FUN=function(x)diff(range(x)))
which(.Last.value != 0)
which( abs(dat$Rating - ave(dat$Rating, dat$Movie)) > 0)
which.max( abs(dat$Rating - ave(dat$Rating, dat$Movie)) )
dat[.Last.value,]
I would add a variable for mode so I can see if there is anything weird going on with the data, like missing data, text, many different answers instead of the rare anomaly,etc. I used "x" as your dataset
# one of many functions to find mode, could use any other
modefunc <- function(x){
names(table(x))[table(x)==max(table(x))]
}
# add variable for mode split by Movie
x$mode <- ave(x = x$Rating,x$Movie,FUN = modefunc)
# do whatever you want with the records that are different
x[x$Rating != x$mode, ]
If you want another function for mode, try other functions for mode