R 中的结构化和交叉引用时限数据点

Question

This is a two-part question, the first concerns how to structure my data, the second concerns asks for suggestions regarding my research design.

I have three sets of data that look like this:

Sample of data set 1

start time	stop time	char
0	9.719	A
9.719	11.735	B
11.735	14.183	A
14.183	16.554	C
16.554	18.482	A
18.482	19.553	B

They are csv format and were exported from a video-annotation software tool. Each row represents an annotation, the first column shows when the annotation begins, and the second column shows when the annotation ends. The values in the third column are references to a particular character that is being depicted and/or talked about in that particular annotation.

Data sets 2 and 3, look the same, although the values for all three columns will are different. Importantly , data sets 2 and 3 come from the same recording. Thus I have three "channels" in which the same characters are represented/discussed but not always at the same time.

If the data were to successfully imported into R and visualized on a time scale, it should look something like this:

The Y axis would be the three different data sets or 'channels' and the X axis would be the duration of the entire recording. The data points thus plotted here would represent the annotations and when they begin and end.

Question #1

How do I appropriately structure my data so that a particular value (A,B, or C), is bound to its specific start and stop times? I imagine that embedded vectors are involved but I'm not sure how to set it up.

Once that is accomplished, what would be the appropriate plot to use to visualize the data and confirm that it's showing what I want to show? Something like a mosaic plot perhaps?

Question #2

When the data is set up appropriately, I want to investigate when the character values (A, B,C) align or do not align with the same values in the other channels (1, 2, 3). How would I go about doing that? I suppose I would need one of the three channels (1, 2, or 3) to serve as a sort of anchor point. Thus, something like:

"For every A value in channel 1, what values overlap with it in channels 2 and 3?"

I would also like to have the ability to point to query a specific data point in a given channel and pull up its value as well as the values of the data points in the other channels that co-occur with it. Thus, I should be able to pull up data point #15 in channel 2 and get its value (A, B, or C) as well as the number of data points that co-occur within the window of data point #15's duration in the other channels (and their values).

This question is a bit exploratory in nature, and any and all answers, suggestions, feedback to either question would be most appreciated.

Answer 1

Not sure about question 2 off-hand, but for question 1...

I think the current format is fine, though in R you will likely want it in one frame (instead of 3) where the dataset name (and/or number, extracted) is a column. For instance, if your file above is in file1.csv , then the others might also be in file2.csv and file3.csv . Reading and combining them can be done with:

allfiles <- lapply(setNames(nm = list.files(pattern="csv$")), read.csv)
alldat <- dplyr::bind_rows(allfiles, .id = "dataset")
alldat
#      dataset start.time stop.time char
# 1  file1.csv      0.000     9.719    A
# 2  file1.csv      9.719    11.735    B
# 3  file1.csv     11.735    14.183    A
# 4  file1.csv     14.183    16.554    C
# 5  file1.csv     16.554    18.482    A
# 6  file1.csv     18.482    19.553    B
# 7  file2.csv      0.000    11.693    A
# 8  file2.csv     11.693    12.310    B
# 9  file2.csv     12.310    13.912    A
# 10 file2.csv     13.912    15.406    C
# 11 file2.csv     15.406    16.988    A
# 12 file2.csv     16.988    19.553    B
# 13 file3.csv      0.000     7.777    A
# 14 file3.csv      7.777    12.920    B
# 15 file3.csv     12.920    15.449    A
# 16 file3.csv     15.449    15.920    C
# 17 file3.csv     15.920    20.042    A
# 18 file3.csv     20.042    19.553    B

(I jittered the first dataset into files 2 and 3.)

From here, plotting with ggplot2 is not too difficult:

library(ggplot2)
ggplot(alldat, aes(fill = char, color = char)) +
  geom_rect(aes(xmin = start.time, xmax = stop.time, ymin = -0.5, ymax = 0.5)) +
  geom_text(aes(x = pmin(start.time, (start.time+stop.time)/2),
                y = 0, label = char),
            hjust = -0.5, vjust = 0.5,
            inherit.aes = FALSE) +
  scale_x_continuous(name = "Time (min)") +
  facet_grid(dataset ~ .) +
  theme(axis.text.y=element_blank(),
        axis.ticks.y=element_blank() )

The plot could be improved by nuancing the hjust= (horizontal justification, ie, -0.5 shifts the letters half a letter to the right of the start.time value) in the narrow bands. Other areas of improvement are mostly addressed by theme(..) , eg, removing the y-axis minor grid lines in the background, limiting the x-axis expansion, placement (or removal) of the legend, all of which are standard ggplot2 operations and should be easy enough to research and apply.