Error when passing data in tibble format.

[khataei]

Environment:

• OS: Ubuntu 20.04
• Tester: Javad Khataei
• Software version: R 4.1.2 from RStudio docker image

Description:

When the input data is passed in tibble format, it throws the following error.
Error: Must subset rows with a valid subscript vector. x Can't convert from <double> to <integer> due to loss of precision.

Steps to reproduce:

library(activityCounts)  
bad_input =  dplyr::as_tibble(activityCounts::sampleXYZ)  
activityCounts::counts(bad_input)

Expected Result:

A data frame with counts:

                  Time  x  y  z
1   2022-01-16 17:57:49  4 93 13
2   2022-01-16 17:57:50 21 50 14
3   2022-01-16 17:57:51 17 22 15
4   2022-01-16 17:57:52 22 39 23
5   2022-01-16 17:57:53 15 25 16
...

Actual Result:

Error: Must subset rows with a valid subscript vector.
x Can't convert from <double> to <integer> due to loss of precision.
Run `rlang::last_error()` to see where the error occurred.
In addition: Warning message:
In counts(data = dplyr::as_tibble(input), hertz = 100, x_axis = 2, 

acknowledgement:

This bug report is created based on Karl Brown feedback.

[khataei]

First Solution:

Convert input data to dataframe before passing it to the function.

library(activityCounts)  
bad_input =  dplyr::as_tibble(activityCounts::sampleXYZ)  
good_input = as.data.farame(bad_input)
activityCounts::counts(bad_input)

Second solution:

A new version of the package is pushed to the dev branch. Installing it via devtools package fixes this issue.

Steps:

• Restart R
• Remove the CRAN version of ActivityCounts by running:
  • remove.packages('activityCounts')
• Install 'devtools' package to install ActivityCounts from Github:
  • install.packages('devtools')
• Install that latest version of ActivityCounts from Github dev branch.
  • devtools::install_git('https://github.com/walkabillylab/activityCounts', force=T, ref = 'dev')
• Restart R

Note:

• We will keep this issue open until we submit the modified version to CRAN.

[AaronRendahl]

So I was having this issue too (and used the same workaround), but I now really think this issue is because the downsampling isn't doing the right thing; it currently takes the code to downsample from seewave::resamp (which is called directly in the version on the jbrond version), which looks like this:

r <- f/g
wave1 <- wave[seq(1, n, by = r), 1]

But what the seq command does is to create a list of evenly spaced numbers from 1 to n, which means they're not integers -- but are being used to index the wave variable anyway! And a tibble correctly won't let you do this but a data.frame will. A data frame silently converts non-integers to integers using floor, like this (demonstrating with a numeric vector, which does the same thing:

Suppose we downsample by a ratio of 1.3, over a range of 1 to 11; then the locations that we really want to see are these:

k <- seq(1, 11, by=1.3)
k
## [1]  1.0  2.3  3.6  4.9  6.2  7.5  8.8 10.1

But the locations we actually get are these:

(1:11)[k]
## [1]  1  2  3  4  6  7  8 10

This doesn't feel right to me...

[khataei]

Hey @jranaraki. Any comment on this?

[AaronRendahl]

For comparison, the original matlab code uses the resample function, which "applies an FIR Antialiasing Lowpass Filter to x and compensates for the delay introduced by the filter" (Documentation at mathworks.com) and the python code uses resampy.resample which uses a "Kaiser" filter (Documentation at readthedocs.io).