triact package for R: Analyzing the lying behavior of cows from accelerometer data


This vignette contains executable examples for the intended use of the R package triact. Most of the functionalities are presented with default parameters. We recommend that you also read the help pages to learn more about the many parameters that can be used to customize the behavior of the methods in triact. Furthermore, background and rationale of the implemented analyses are described in detail in the following publication:


Simmler M., Brouwers S. P., 2023. triact package for R: Analyzing the lying behavior of cows from accelerometer data, under review


Setup


Since triact is typically used with accelerometer data with sampling frequency of ≥1 Hz, it is advisable to set R’s global option digits.secs to >=1 in order to enable the printing of fractional seconds.

options(digits.secs = 1)


Via the global option triact_table the type of tables returned by triact can be specified. Options are "data.frame" (the default), "tibble", and "data.table".

options(triact_table = "data.frame")



Getting help

All functionalities of the triact R package are documented on the help page of the Triact R6 class.

?Triact



Inspecting the example data

The triact R package includes two raw data files from triaxial accelerometers (MSR145, MSR Electronics, Switzerland) attached to the left hind leg of dairy cows. The sampling frequency was 5 Hz. Each file represents one day of recording of one cow.

input_dir <-  system.file("extdata", package = "triact") 

files <- list.files(input_dir)

print(files)
## [1] "cow01_5hz.csv" "cow02_5hz.csv"


Inspecting one of the files reveals a file header and the semicolon-separated data starting after the line with "*Data“. This is an example of what files imported by triact might look like. However, triact can handle any kind of delimiter-separated text files, with or without an arbitrary file header (which is ignored during import).

cat(paste(readLines(file.path(input_dir, files[1]), n = 30), collapse = "\n"))
## *CREATOR
## msr_cutter.exe;[V6.06.02]
## msr2csv.exe;[V6.06.02]
## 
## *STARTTIME
## 2021-06-29;06:00:00;
## 
## *MODUL
## NAME;MSR314553;MSR314553;MSR314553
## *NAME
## NAME;6A;6A;6A
## 
## ID;[C26113 V5.66];[C26113 V5.66];[C26113 V5.66]
## 
## *CHANNEL
## TIME;ACC x;ACC y;ACC z
## 
## *UNIT
## ;G;G;G
## 
## *LIMITS
## ALARM;;;
## RECRD;;;
## LIMIT1;;;
## LIMIT2;;;
## 
## *DATA
## 2021-06-29 06:00:00.055;-0.048;1.032;-0.063
## 2021-06-29 06:00:00.258;-0.048;1.000;-0.063
## 2021-06-29 06:00:00.461;-0.048;1.000;-0.063



Importing data

Importing from raw files

The typical triact workflow starts by creating a new object of the Triact class.

my_triact <- Triact$new()


Acceleration data is then imported into the Triact object (here named ‘my_triact’). The $load_files() method can be used to import any raw data files, which are delimiter-separated text files. If you are starting with a new file format you may want to select one or a few (small) files to find out how to specify the method’s argument to enable correct import from the file(s). Once this is done, you move on to process all your files. Examine your file format in a plain text editor or in R (as above).

my_triact$load_files(input = input_dir,
                     id_substring = c(1, 5),
                     timeFwdUpRight_cols = c(1, -2, 3, -4),
                     time_format = "%Y-%m-%d %H:%M:%OS",
                     tz = "Europe/Zurich",
                     skip = "DATA",
                     sep = ";",
                     header = FALSE,
                     dec = ".")


The parameters as used above in the call of $load_files() have the following effects:

  • With id_substring = c(1, 5) we specify which part of the filenames represents the unique identifier of the cows, by indicating the start and end character positions c(start, end). The example files are named “cow01_5hz.csv” and “cow02_5hz.csv”. The substring from the first to the fifth character can here serve as unique identifier, therefore c(1, 5). Alternatively, we could have used a perl-like regular expression that matches the substring ("^\\w{5}", see ?regex).

  • With timeFwdUpRight_cols = c(1, -2, 3, -4) we map the columns as found in the files to the time, and the forward, up, and right accelerations as understood by triact. Fig. 1a shows the accelerometer used to collect the example data with the axis directions as defined by the manufacturer (XYZ). Fig. 1b shows the directions as used in triact. To collect the example data, the accelerometer was mounted to the outside of the left hind leg with the Y axis pointing in Up direction, X in opposite direction of forward (directed backwards), and Z in opposite direction of right (directed left). In timeFwdUpRight_cols = c(1, -2, 3, -4), the first number indicates in which column in the file the timestamp (data-time) is located, here in the first column. The second number indicates which column in the file maps to forward acceleration, here the second column, but with negative mathematical sign as X was pointing in the opposite direction of forward (hence -2). The third number indicates which column in the file maps to the up axis, here the third (the Y data). The last number indicates which column in the file maps to the right acceleration, here the fourth, but with opposite mathematical sign as Z was pointing in the opposite direction of right (hence -4).

  • With time_format = "%Y-%m-%d %H:%M:%OS" we specify the format of the timestamps as found in the files. The syntax is described in ?strptime.

  • With tz = "Europe/Zurich" we specify the time zone of the timestamps in the files. This is usually irrelevant if you do not work across time zones as your system’s time zone is used as default.

  • With skip = "DATA" we specify the line in the files to start reading data, here using a (sub)string of that line (see file inspection above). Alternatively we could have used an integer indicating the number of lines to skip before reading data.

  • With sep = ";" we specify the separator character that separates columns in the files.

  • With header = FALSE we specify whether the first column of the data (after considering skip) contains column names.

  • Finally, with dec = "." we specify the decimal separator.

(a) Axis directions of the MSR145 accelerometer (MSR Electronics, Switzerland). (b) Directions as used in the triact R package. Shown is the situation with the accelerometer on the outside of the left hind leg as used to collect the example data.
(a) Axis directions of the MSR145 accelerometer (MSR Electronics, Switzerland). (b) Directions as used in the triact R package. Shown is the situation with the accelerometer on the outside of the left hind leg as used to collect the example data.


Of the optional arguments that specify details about your file format, you may not need to specify many, because by default most of them are inferred from the data. In the case of the example data, you can reduce the arguments to:

my_triact$load_files(input = input_dir,
                     id_substring = c(1, 5),
                     timeFwdUpRight_cols = c(1, -2, 3, -4),
                     skip = "DATA")


Once the data has been imported, you can use the `$data’ field to inspect the imported raw data and (later) added analyses at any time during the workflow.

head(my_triact$data)
##      id                time acc_fwd acc_up acc_right
## 1 cow01 2021-06-29 06:00:00   0.048  1.032     0.063
## 2 cow01 2021-06-29 06:00:00   0.048  1.000     0.063
## 3 cow01 2021-06-29 06:00:00   0.048  1.000     0.063
## 4 cow01 2021-06-29 06:00:00   0.048  1.032     0.063
## 5 cow01 2021-06-29 06:00:00   0.048  1.032     0.031
## 6 cow01 2021-06-29 06:00:01   0.079  1.032     0.031
str(my_triact$data)
## 'data.frame':    354462 obs. of  5 variables:
##  $ id       : Factor w/ 2 levels "cow01","cow02": 1 1 1 1 1 1 1 1 1 1 ...
##  $ time     : POSIXct, format: "2021-06-29 06:00:00.0" "2021-06-29 06:00:00.2" ...
##  $ acc_fwd  : num  0.048 0.048 0.048 0.048 0.048 0.079 0.048 0.048 0.079 0.079 ...
##  $ acc_up   : num  1.03 1 1 1.03 1.03 ...
##  $ acc_right: num  0.063 0.063 0.063 0.063 0.031 0.031 0.031 0.031 0.031 0.031 ...

⚠️ Warning

Correct mapping of your data to the body relative direction as used in triact is necessary for correct determination of the lying/standing posture and the lying laterality (lying side). Therefore, special care should be taken when specifying the parameter timeFwdUpRight_cols.

If you are unsure of the directions of your accelerometer’s XYZ data, you can determine them by experimentation. If the axis in question is pointing up, it should measure +1 g at rest. Note that accelerometers measure proper acceleration which is relative to free fall, so the gravitational components points skywards (reaction force of the ground).


ℹ NOTE 1

If you suspect that you have accidentally mounted some of your accelerometers 180° rotated in the sagittal plane, but you don’t know which raw data files are affected, you should call the $check_orientation() method after loading the data into the Triact object. This method will identify such “upside down” data and correct it accordingly by multiplying the up and forward axes by -1.


ℹ NOTE 2

You can import multiple files from the same cow, i.e., multiple files with the same id as extracted from file names according to what you specify via the id_substring parameter. But the data with the same cow id must follow each other without gaps in time. If you have gaps, you should use unique ids for the time series without gaps, e.g. you may need to use the cow identifier in combination with the date as a unique id.


ℹ NOTE 3

It can take a long time to import many files. You can speed things up by setting the parallel parameter of $load_files() to > 1, which allows files to be read in parallel. See ?Triact for more information.



Importing from a data.frame

Alternatively to importing from raw data files with load_files(), you can read your files with your own routine and then use the $load_table() method to import a data.frame into the Triact object. Column names and data types of the data.frame must be exactly as described in ?Triact. An example data.frame, cows_5hz, is provided by the triact package.

A description of the example data.frame can be found on its help page.

?cows_5hz
str(cows_5hz)
## 'data.frame':    354462 obs. of  5 variables:
##  $ id       : Factor w/ 2 levels "cow01","cow02": 1 1 1 1 1 1 1 1 1 1 ...
##  $ time     : POSIXct, format: "2021-06-29 06:00:00.0" "2021-06-29 06:00:00.2" ...
##  $ acc_fwd  : num  0.048 0.048 0.048 0.048 0.048 0.079 0.048 0.048 0.079 0.079 ...
##  $ acc_up   : num  1.03 1 1 1.03 1.03 ...
##  $ acc_right: num  0.063 0.063 0.063 0.063 0.031 0.031 0.031 0.031 0.031 0.031 ...


Create new Triact object and import cows_5hz.

my_triact <- Triact$new()
my_triact$load_table(cows_5hz)



Adding analyses

Calling add_… methods triggers analyses of lying behavior and the calculation of proxies for the level of physical activity. The results of the analyses are obtained for each time point of your accelerometer data and added in new columns to the tabular data in the Triact object.


Detecting standing and lying posture

The $add_lying() method performs the classification into lying and standing posture. The results are (silently) added to the data in the Triact object as a logical column named lying, where TRUE indicates lying and FALSE standing. Additionally, lying and standing bouts are uniquely numbered (per cow or ID) in column bout_nr.

my_triact$add_lying()
str(my_triact$data)
## 'data.frame':    354462 obs. of  7 variables:
##  $ id       : Factor w/ 2 levels "cow01","cow02": 1 1 1 1 1 1 1 1 1 1 ...
##  $ time     : POSIXct, format: "2021-06-29 06:00:00.0" "2021-06-29 06:00:00.2" ...
##  $ acc_fwd  : num  0.048 0.048 0.048 0.048 0.048 0.079 0.048 0.048 0.079 0.079 ...
##  $ acc_up   : num  1.03 1 1 1.03 1.03 ...
##  $ acc_right: num  0.063 0.063 0.063 0.063 0.031 0.031 0.031 0.031 0.031 0.031 ...
##  $ bout_nr  : num  1 1 1 1 1 1 1 1 1 1 ...
##  $ lying    : logi  FALSE FALSE FALSE FALSE FALSE FALSE ...

ℹ NOTE

The $add_lying() method comes with many parameters that allow you to tweak the underlying algorithm. With the default parameters, you can expect good results if your accelerometer sampling frequency is ≥1 Hz. See ?Triact and Simmler & Brouwers (2023) for explanations.


Detecting lying laterality

The $add_side() method performs the determination of lying laterality (lying side). The results are (silently) added to the data in the Triact object as a factor column named side, with levels ‘L’ (left) and ‘R’ (right). During standing posture, side is NA (not available). Crucial for correct determination of the lying side is the correct specification of which hind leg the accelerometer was mounted on (parameter left_leg = TRUE for left, or FALSE for right).

my_triact$add_side(left_leg = TRUE)
str(my_triact$data)
## 'data.frame':    354462 obs. of  8 variables:
##  $ id       : Factor w/ 2 levels "cow01","cow02": 1 1 1 1 1 1 1 1 1 1 ...
##  $ time     : POSIXct, format: "2021-06-29 06:00:00.0" "2021-06-29 06:00:00.2" ...
##  $ acc_fwd  : num  0.048 0.048 0.048 0.048 0.048 0.079 0.048 0.048 0.079 0.079 ...
##  $ acc_up   : num  1.03 1 1 1.03 1.03 ...
##  $ acc_right: num  0.063 0.063 0.063 0.063 0.031 0.031 0.031 0.031 0.031 0.031 ...
##  $ bout_nr  : num  1 1 1 1 1 1 1 1 1 1 ...
##  $ lying    : logi  FALSE FALSE FALSE FALSE FALSE FALSE ...
##  $ side     : Factor w/ 2 levels "L","R": NA NA NA NA NA NA NA NA NA NA ...


Calculating proxies for the level of physical activity

The $add_activity() method performs the calculation of proxies for the level of physical activity of the cow(s). By default, the L2 norm of the vector of the dynamic body acceleration (DBA) is calculated. It is ‘adjusted’ to a value of zero during lying bouts (prefix Adj), i.e., periods when cows are lying are considered as ‘inactive’ by definition. The results are (silently) added to the data in the Triact object as numeric column named AdjL2DBA.

my_triact$add_activity()
str(my_triact$data)
## 'data.frame':    354462 obs. of  9 variables:
##  $ id       : Factor w/ 2 levels "cow01","cow02": 1 1 1 1 1 1 1 1 1 1 ...
##  $ time     : POSIXct, format: "2021-06-29 06:00:00.0" "2021-06-29 06:00:00.2" ...
##  $ acc_fwd  : num  0.048 0.048 0.048 0.048 0.048 0.079 0.048 0.048 0.079 0.079 ...
##  $ acc_up   : num  1.03 1 1 1.03 1.03 ...
##  $ acc_right: num  0.063 0.063 0.063 0.063 0.031 0.031 0.031 0.031 0.031 0.031 ...
##  $ bout_nr  : num  1 1 1 1 1 1 1 1 1 1 ...
##  $ lying    : logi  FALSE FALSE FALSE FALSE FALSE FALSE ...
##  $ side     : Factor w/ 2 levels "L","R": NA NA NA NA NA NA NA NA NA NA ...
##  $ AdjL2DBA : num  0.0453 0.032 0.032 0.0453 0.032 ...

ℹ NOTE 1

You can select several proxies for the level of physical activity, namely the L1 and L2 norms of DBA and Jerk. Use the norm and dynamic_measure parameters to do this. See Simmler & Brouwers (2023) for a discussion of these proxies.


ℹ NOTE 2

Calculating DBA-based proxies involves a filtering step that can be tuned with several parameters (see ?Triact). With the default parameters, you can expect good results if your accelerometer sampling frequency is ≥1 Hz.



Summarizing results

The $summarize… methods are used to summarize the analyses added to the Triact object per time period, representing either the standing/lying bouts or regular intervals, e.g. 1 h or 24 h.


Summarizing per lying/standing bout

With $summarize_bouts() a summary is created for the individual lying and standing bouts, with duration, mean activity, and lying side (for a lying bout). There is a help page with a description of the output.

?bout_summary
bouts_summary <- my_triact$summarize_bouts()


In the output you can see that the first bout per cow is not completely observed (startTime is missing) and therefor NAs are returned for measures that depend on complete observation of the bout, e.g. duration.

head(bouts_summary)
##      id bout_nr           startTime             endTime  duration lying side
## 1 cow01       1                <NA> 2021-06-29 06:17:40        NA FALSE <NA>
## 2 cow01       2 2021-06-29 06:17:40 2021-06-29 07:07:16  49.59968  TRUE    L
## 3 cow01       3 2021-06-29 07:07:16 2021-06-29 09:11:33 124.28197 FALSE <NA>
## 4 cow01       4 2021-06-29 09:11:33 2021-06-29 09:46:51  35.30985  TRUE    R
## 5 cow01       5 2021-06-29 09:46:51 2021-06-29 12:44:45 177.90022 FALSE <NA>
## 6 cow01       6 2021-06-29 12:44:45 2021-06-29 14:18:29  93.73198  TRUE    R
##   meanAdjL2DBA
## 1           NA
## 2   0.00000000
## 3   0.20761060
## 4   0.00000000
## 5   0.08440727
## 6   0.00000000


If only the lying bouts are of interest, the bout_type parameter can be specified accordingly.

bouts_summary <- my_triact$summarize_bouts(bout_type = "lying")


You can see that the last lying bout of cow01 and the first lying bout of cow02 were incompletely observed (startTime and endTime missing, respectively). Again, NAs are returned for measures that depend on complete observation of the bout.

head(bouts_summary)
##      id bout_nr           startTime             endTime  duration lying side
## 1 cow01       2 2021-06-29 06:17:40 2021-06-29 07:07:16  49.59968  TRUE    L
## 2 cow01       4 2021-06-29 09:11:33 2021-06-29 09:46:51  35.30985  TRUE    R
## 3 cow01       6 2021-06-29 12:44:45 2021-06-29 14:18:29  93.73198  TRUE    R
## 4 cow01       8 2021-06-29 14:23:35                <NA>        NA  TRUE    L
## 5 cow02       1                <NA> 2021-06-29 07:07:26        NA  TRUE    R
## 6 cow02       3 2021-06-29 12:56:43 2021-06-29 14:38:01 101.29500  TRUE    R
##   meanAdjL2DBA
## 1            0
## 2            0
## 3            0
## 4           NA
## 5           NA
## 6            0

ℹ NOTE

When calling the $summarize_bouts() method with parameter calc_for_incomplete = TRUE, a complete summary is also returned for the incompletely observed bouts (first and last bout for each cow). It does this by simply assuming that the bouts were completely observed. Use this option with caution.



Summarizing per regular intervals

With $summarize_intervals() the summary is obtained per regular intervals, by default per hour. There is a help page with a description of the output.

?interval_summary
int_summary <- my_triact$summarize_intervals()


The NAs in the output are a result of incompletely observed intervals (first and last interval of each cow). The NaN on the other hand do not indicate missing information: For example, if the cow was not standing in the interval, the mean activity during standing is not zero, but cannot be calculated (thus NaN, “not a number”).

head(int_summary)
##      id           startTime          centerTime             endTime duration
## 1 cow01 2021-06-29 06:00:00 2021-06-29 06:30:00 2021-06-29 07:00:00       NA
## 2 cow01 2021-06-29 07:00:00 2021-06-29 07:30:00 2021-06-29 08:00:00 59.99968
## 3 cow01 2021-06-29 08:00:00 2021-06-29 08:30:00 2021-06-29 09:00:00 59.99970
## 4 cow01 2021-06-29 09:00:00 2021-06-29 09:30:00 2021-06-29 10:00:00 60.00307
## 5 cow01 2021-06-29 10:00:00 2021-06-29 10:30:00 2021-06-29 11:00:00 59.99970
## 6 cow01 2021-06-29 11:00:00 2021-06-29 11:30:00 2021-06-29 12:00:00 59.99968
##   durationStanding durationLying durationLyingLeft durationLyingRight
## 1               NA            NA                NA                 NA
## 2         52.72782      7.271868          7.271868            0.00000
## 3         59.99970      0.000000          0.000000            0.00000
## 4         24.69321     35.309862          0.000000           35.30986
## 5         59.99970      0.000000          0.000000            0.00000
## 6         59.99968      0.000000          0.000000            0.00000
##   meanAdjL2DBA meanAdjL2DBALying meanAdjL2DBAStanding
## 1           NA                NA                   NA
## 2   0.23734730                 0           0.27008066
## 3   0.16785237               NaN           0.16785237
## 4   0.04852252                 0           0.11790694
## 5   0.10169885               NaN           0.10169885
## 6   0.06039537               NaN           0.06039537
str(int_summary)
## 'data.frame':    20 obs. of  12 variables:
##  $ id                  : Factor w/ 2 levels "cow01","cow02": 1 1 1 1 1 1 1 1 1 1 ...
##  $ startTime           : POSIXct, format: "2021-06-29 06:00:00" "2021-06-29 07:00:00" ...
##  $ centerTime          : POSIXct, format: "2021-06-29 06:30:00" "2021-06-29 07:30:00" ...
##  $ endTime             : POSIXct, format: "2021-06-29 07:00:00" "2021-06-29 08:00:00" ...
##  $ duration            : num  NA 60 60 60 60 ...
##  $ durationStanding    : num  NA 52.7 60 24.7 60 ...
##  $ durationLying       : num  NA 7.27 0 35.31 0 ...
##  $ durationLyingLeft   : num  NA 7.27 0 0 0 ...
##  $ durationLyingRight  : num  NA 0 0 35.3 0 ...
##  $ meanAdjL2DBA        : num  NA 0.2373 0.1679 0.0485 0.1017 ...
##  $ meanAdjL2DBALying   : num  NA 0 NaN 0 NaN NaN 0 0 0 NA ...
##  $ meanAdjL2DBAStanding: num  NA 0.27 0.168 0.118 0.102 ...


The intervals can be specified quite flexibly: In case of 30 min intervals and starting 10 min after the full hour you can specify interval and lag_in_s parameters accordingly.

int_summary <- my_triact$summarize_intervals(interval = "30 min",
                                             lag_in_s = 10 * 60)
head(int_summary)
##      id           startTime          centerTime             endTime duration
## 1 cow01 2021-06-29 05:40:00 2021-06-29 05:55:00 2021-06-29 06:10:00       NA
## 2 cow01 2021-06-29 06:10:00 2021-06-29 06:25:00 2021-06-29 06:40:00 30.00152
## 3 cow01 2021-06-29 06:40:00 2021-06-29 06:55:00 2021-06-29 07:10:00 29.99812
## 4 cow01 2021-06-29 07:10:00 2021-06-29 07:25:00 2021-06-29 07:40:00 30.00152
## 5 cow01 2021-06-29 07:40:00 2021-06-29 07:55:00 2021-06-29 08:10:00 29.99813
## 6 cow01 2021-06-29 08:10:00 2021-06-29 08:25:00 2021-06-29 08:40:00 30.00150
##   durationStanding durationLying durationLyingLeft durationLyingRight
## 1               NA            NA                NA                 NA
## 2         7.671342      22.33017          22.33017                  0
## 3         2.728640      27.26948          27.26948                  0
## 4        30.001517       0.00000           0.00000                  0
## 5        29.998133       0.00000           0.00000                  0
## 6        30.001500       0.00000           0.00000                  0
##   meanAdjL2DBA meanAdjL2DBALying meanAdjL2DBAStanding
## 1           NA                NA                   NA
## 2   0.03410266                 0            0.1333706
## 3   0.05575711                 0            0.6129823
## 4   0.27305669               NaN            0.2730567
## 5   0.20804566               NaN            0.2080457
## 6   0.17560383               NaN            0.1756038
str(int_summary)
## 'data.frame':    42 obs. of  12 variables:
##  $ id                  : Factor w/ 2 levels "cow01","cow02": 1 1 1 1 1 1 1 1 1 1 ...
##  $ startTime           : POSIXct, format: "2021-06-29 05:40:00" "2021-06-29 06:10:00" ...
##  $ centerTime          : POSIXct, format: "2021-06-29 05:55:00" "2021-06-29 06:25:00" ...
##  $ endTime             : POSIXct, format: "2021-06-29 06:10:00" "2021-06-29 06:40:00" ...
##  $ duration            : num  NA 30 30 30 30 ...
##  $ durationStanding    : num  NA 7.67 2.73 30 30 ...
##  $ durationLying       : num  NA 22.3 27.3 0 0 ...
##  $ durationLyingLeft   : num  NA 22.3 27.3 0 0 ...
##  $ durationLyingRight  : num  NA 0 0 0 0 ...
##  $ meanAdjL2DBA        : num  NA 0.0341 0.0558 0.2731 0.208 ...
##  $ meanAdjL2DBALying   : num  NA 0 0 NaN NaN NaN NaN 0 0 NaN ...
##  $ meanAdjL2DBAStanding: num  NA 0.133 0.613 0.273 0.208 ...


With bouts = TRUE you can request that, additionally, the bouts within the intervals are summarized. For measures such as the number of lying bouts or mean lying bout duration, a weighted mean is calculated with the weights being the proportion of the individual bout overlapping with the respective interval. With side = TRUE you can additionally request a differentiation of all results by lying side. Again, NAs in the output are a result of incompletely observed intervals (first and last interval of each cow). Additionally you will sometimes find NAs in the output for summaries of bouts per interval (bouts = TRUE) that result from incompletely observed bouts (first and last bout for each cow). As bouts can span several intervals, this can affect more than just the first and the last interval. Please be aware about the difference between NA and NaN described above.

int_summary <- my_triact$summarize_intervals(bouts = TRUE,
                                             side = TRUE)
str(int_summary)
## 'data.frame':    20 obs. of  22 variables:
##  $ id                         : Factor w/ 2 levels "cow01","cow02": 1 1 1 1 1 1 1 1 1 1 ...
##  $ startTime                  : POSIXct, format: "2021-06-29 06:00:00" "2021-06-29 07:00:00" ...
##  $ centerTime                 : POSIXct, format: "2021-06-29 06:30:00" "2021-06-29 07:30:00" ...
##  $ endTime                    : POSIXct, format: "2021-06-29 07:00:00" "2021-06-29 08:00:00" ...
##  $ duration                   : num  NA 60 60 60 60 ...
##  $ durationStanding           : num  NA 52.7 60 24.7 60 ...
##  $ durationLying              : num  NA 7.27 0 35.31 0 ...
##  $ durationLyingLeft          : num  NA 7.27 0 0 0 ...
##  $ durationLyingRight         : num  NA 0 0 35.3 0 ...
##  $ meanAdjL2DBA               : num  NA 0.2373 0.1679 0.0485 0.1017 ...
##  $ meanAdjL2DBALying          : num  NA 0 NaN 0 NaN NaN 0 0 0 NA ...
##  $ meanAdjL2DBAStanding       : num  NA 0.27 0.168 0.118 0.102 ...
##  $ meanAdjL2DBALyingLeft      : num  NA 0 NaN NaN NaN NaN NaN NaN 0 NA ...
##  $ meanAdjL2DBALyingRight     : num  NA NaN NaN 0 NaN NaN 0 0 0 NA ...
##  $ nBoutsStanding             : num  NA 0.424 0.483 0.167 0.337 ...
##  $ nBoutsLying                : num  NA 0.147 0 1 0 ...
##  $ nBoutsLyingLeft            : num  NA 0.147 0 0 0 ...
##  $ nBoutsLyingRight           : num  NA 0 0 1 0 ...
##  $ wMeanDurationStandingBout  : num  NA 124 124 148 178 ...
##  $ wMeanDurationLyingBout     : num  NA 49.6 NaN 35.3 NaN ...
##  $ wMeanDurationLyingBoutLeft : num  NA 49.6 NaN NaN NaN ...
##  $ wMeanDurationLyingBoutRight: num  NA NaN NaN 35.3 NaN ...

ℹ NOTE

When calling the $summarize_intervals() method with parameter calc_for_incomplete = TRUE, a complete summary will also be returned for the incompletely observed intervals (first and last interval for each cow) and for any parameter using information of incompletely observed bouts (which can affect more than just the first and last interval). It does this by simply assuming that intervals and bouts were completely observed. Use this option with caution.


Visualizing results


The triact package does not come with visualization capabilities. But the data can easily be accessed and plotted with base R or packages dedicated to graphics (e.g. ggplot2). The following example shows how to access the data of a single cow (here with id “cow01”) and to visualize the lying behavior.

cow_id = "cow01"

data_01 <- my_triact$data[my_triact$data$id == cow_id, ]

plot(!lying ~ time, data = data_01,
     type = "l", ylab = "", yaxt = "n", bty = "n")

lines(ifelse(side == "R", 0, NA) ~ time, data = data_01, col = "orange")
lines(ifelse(side == "L", 0, NA) ~ time, data = data_01, col = "purple")

axis(2, at = c(0, 1),
     labels = c("lying", "standing"),
     las = 1,
     lwd = 0)

legend(x = "center",
       legend = c("right", "left"),
       col = c("orange", "purple"),
       lwd = 1, bty = "n", title = "lying side")


Extracting posture transitions

Using $extract_liedown() and $extract_standup(), the raw acceleration data (and added analyses) of the posture transitions, i.e., lying-to-standing and standing-to-lying, can be extracted.

With default parameters, only the time of the transition, bout nr of the lying bout, and lying side (if available) are returned.

st_ups <- my_triact$extract_standup()
print(st_ups)
##      id                time bout_nr side
## 1 cow01 2021-06-29 07:07:16       2    L
## 2 cow01 2021-06-29 09:46:51       4    R
## 3 cow01 2021-06-29 14:18:29       6    R
## 4 cow01 2021-06-29 15:29:43       8    L
## 5 cow02 2021-06-29 07:07:26       1    R
## 6 cow02 2021-06-29 14:38:00       3    R
## 7 cow02 2021-06-29 15:43:00       5    L


When specifying ‘sec_before’ and ’sec_after`, time series around the exact moment of posture transition as detected by triact are returned. The result is a list with tables (one table per posture transition).

l_downs <- my_triact$extract_liedown(sec_before = 3, sec_after = 3)


head(l_downs[[1]])
##      id                time acc_fwd acc_up acc_right bout_nr lying side
## 1 cow01 2021-06-29 06:17:37   0.238  0.968     0.031       1 FALSE <NA>
## 2 cow01 2021-06-29 06:17:37   0.175  0.968     0.031       1 FALSE <NA>
## 3 cow01 2021-06-29 06:17:37   0.175  1.000     0.000       1 FALSE <NA>
## 4 cow01 2021-06-29 06:17:38   0.143  1.000     0.063       1 FALSE <NA>
## 5 cow01 2021-06-29 06:17:38   0.143  1.000     0.063       1 FALSE <NA>
## 6 cow01 2021-06-29 06:17:38   0.111  1.032     0.031       1 FALSE <NA>
##     AdjL2DBA
## 1 0.09929753
## 2 0.09897980
## 3 0.13270268
## 4 0.09400532
## 5 0.09400532
## 6 0.14817895