Upsampling categorical variable in time series data in R

Question

I apologize if this is redundant, but I have tried to look for solutions, and have not found anything that appears to be the answer to my question. So, I have time series data for a bunch of variables. Some variables (continuous) were recorded at 1000hz, and the other variables (categorical) is at 500hz. I want to align them in R, such that the categorical variables have the same number of rows as the continuous variables. In other words, I would like to merge the two dataframes and fill in the gaps automatically for the categorical variables, so that my data, which looks like this:

t_emg	CF_01	CF_02
0	-0.01796875	0.01234375
0.001	0.001875	0.05125
0.002	0.00828125	-0.04140625
0.003	-0.0125	-0.005625
0.004	-0.00765625	0.00078125
0.005	-0.01	0.02234375
0.006	0.01515625	-0.01296875
0.007	-0.01375	-0.070625
0.008	0.0096875	0.1534375
0.009	-0.05203125	0.02984375
0.01	0.1234375	0.00765625
0.011	-0.0596875	0.005625
0.012	-0.09296875	-0.001875
0.013	-0.0834375	-0.1128125
0.014	-0.009375	0.0090625
0.015	-0.00609375	0.02
0.016	0.00890625	0.0209375
0.017	0.02671875	0.00375
0.018	-0.001875	-0.0003125
0.019	-0.00171875	0.00171875

and this:

t_kinematics	Phase	stride_stop
0	swing	A
0.002	swing	A
0.004	swing	A
0.006	swing	A
0.008	swing	A
0.01	swing	A
0.012	swing	A
0.014	stance	A
0.016	stance	A
0.018	stance	A

will end up looking like this:

t_emg	CF_01	CF_02	t_kinematics	Phase	stride_stop
0	-0.01796875	0.01234375	0	swing	A
0.001	0.001875	0.05125	0.001	swing	A
0.002	0.00828125	-0.04140625	0.002	swing	A
0.003	-0.0125	-0.005625	0.003	swing	A
0.004	-0.00765625	0.00078125	0.004	swing	A
0.005	-0.01	0.02234375	0.005	swing	A
0.006	0.01515625	-0.01296875	0.006	swing	A
0.007	-0.01375	-0.070625	0.007	swing	A
0.008	0.0096875	0.1534375	0.008	swing	A
0.009	-0.05203125	0.02984375	0.009	swing	A
0.01	0.1234375	0.00765625	0.01	swing	A
0.011	-0.0596875	0.005625	0.011	swing	A
0.012	-0.09296875	-0.001875	0.012	swing	A
0.013	-0.0834375	-0.1128125	0.013	swing	A
0.014	-0.009375	0.0090625	0.014	stance	A
0.015	-0.00609375	0.02	0.015	stance	A
0.016	0.00890625	0.0209375	0.016	stance	A
0.017	0.02671875	0.00375	0.017	stance	A
0.018	-0.001875	-0.0003125	0.018	stance	A
0.019	-0.00171875	0.00171875	0.019	stance	A

Additionally, there are many NAs in the categorical data, and I would like those to be retained. All the information that I can find is more geared towards imbalanced data or imputation of missing values. It is also not really prediction, as I am really just interested in filling in the temporal gaps for the categorical variables.

Answer 1

From your single-frame, we really have two 3-column frames.

quux1 <- structure(list(t_emg = c(0, 0.001, 0.002, 0.003, 0.004, 0.005, 0.006, 0.007, 0.008, 0.009, 0.01, 0.011, 0.012, 0.013, 0.014, 0.015, 0.016, 0.017, 0.018, 0.019), CF_01 = c(-0.01796875, 0.001875, 0.00828125, -0.0125, -0.00765625, -0.01, 0.01515625, -0.01375, 0.0096875, -0.05203125, 0.1234375, -0.0596875, -0.09296875, -0.0834375, -0.009375, -0.00609375, 0.00890625, 0.02671875, -0.001875, -0.00171875), CF_02 = c(0.01234375, 0.05125, -0.04140625, -0.005625, 0.00078125, 0.02234375, -0.01296875, -0.070625,  0.1534375, 0.02984375, 0.00765625, 0.005625, -0.001875, -0.1128125, 0.0090625, 0.02, 0.0209375, 0.00375, -0.0003125, 0.00171875)), class = "data.frame", row.names = c(NA, -20L))
quux2 <- structure(list(t_kinematics = c(0, 0.002, 0.004, 0.006, 0.008, 0.01, 0.012, 0.014, 0.016, 0.018), Phase = c("swing", "swing", "swing", "swing", "swing", "swing", "swing", "stance", "stance", "stance"), stride_stop = c("A", "A", "A", "A", "A", "A", "A", "A", "A", "A")), row.names = c(NA, 10L), class = "data.frame", na.action = structure(11:20, names = c("11", "12", "13", "14", "15", "16", "17", "18", "19", "20"), class = "omit"))

Using this, we'll use findInterval to map the other columns to the time interval.

intvl <- findInterval(quux1$t_emg, quux2$t_kinematics)
out <- cbind(quux1, lapply(quux2[,-1], `[`, intvl))
out
#    t_emg       CF_01       CF_02  Phase stride_stop
# 1  0.000 -0.01796875  0.01234375  swing           A
# 2  0.001  0.00187500  0.05125000  swing           A
# 3  0.002  0.00828125 -0.04140625  swing           A
# 4  0.003 -0.01250000 -0.00562500  swing           A
# 5  0.004 -0.00765625  0.00078125  swing           A
# 6  0.005 -0.01000000  0.02234375  swing           A
# 7  0.006  0.01515625 -0.01296875  swing           A
# 8  0.007 -0.01375000 -0.07062500  swing           A
# 9  0.008  0.00968750  0.15343750  swing           A
# 10 0.009 -0.05203125  0.02984375  swing           A
# 11 0.010  0.12343750  0.00765625  swing           A
# 12 0.011 -0.05968750  0.00562500  swing           A
# 13 0.012 -0.09296875 -0.00187500  swing           A
# 14 0.013 -0.08343750 -0.11281250  swing           A
# 15 0.014 -0.00937500  0.00906250 stance           A
# 16 0.015 -0.00609375  0.02000000 stance           A
# 17 0.016  0.00890625  0.02093750 stance           A
# 18 0.017  0.02671875  0.00375000 stance           A
# 19 0.018 -0.00187500 -0.00031250 stance           A
# 20 0.019 -0.00171875  0.00171875 stance           A

If you really need t_kinematics, then just copy it over from t_emg:

out$t_kinematics <- out$t_emg

---

We could also do a range-join:

library(dplyr)
# library(tidyr) # fill
quux1 |>
  mutate(t_emg_next = lead(t_emg, default = Inf) - 1e-9) |>
  left_join(quux2, join_by(between(y$t_kinematics, x$t_emg, x$t_emg_next))) |>
  tidyr::fill(t_kinematics, Phase, stride_stop) |>
  select(-t_emg_next)
#    t_emg       CF_01       CF_02 t_kinematics  Phase stride_stop
# 1  0.000 -0.01796875  0.01234375        0.000  swing           A
# 2  0.001  0.00187500  0.05125000        0.000  swing           A
# 3  0.002  0.00828125 -0.04140625        0.002  swing           A
# 4  0.003 -0.01250000 -0.00562500        0.002  swing           A
# 5  0.004 -0.00765625  0.00078125        0.004  swing           A
# 6  0.005 -0.01000000  0.02234375        0.004  swing           A
# 7  0.006  0.01515625 -0.01296875        0.006  swing           A
# 8  0.007 -0.01375000 -0.07062500        0.006  swing           A
# 9  0.008  0.00968750  0.15343750        0.008  swing           A
# 10 0.009 -0.05203125  0.02984375        0.008  swing           A
# 11 0.010  0.12343750  0.00765625        0.010  swing           A
# 12 0.011 -0.05968750  0.00562500        0.010  swing           A
# 13 0.012 -0.09296875 -0.00187500        0.012  swing           A
# 14 0.013 -0.08343750 -0.11281250        0.012  swing           A
# 15 0.014 -0.00937500  0.00906250        0.014 stance           A
# 16 0.015 -0.00609375  0.02000000        0.014 stance           A
# 17 0.016  0.00890625  0.02093750        0.016 stance           A
# 18 0.017  0.02671875  0.00375000        0.016 stance           A
# 19 0.018 -0.00187500 -0.00031250        0.018 stance           A
# 20 0.019 -0.00171875  0.00171875        0.018 stance           A

The 1e-9 is to force the between comparison to be right-open; may not be required with your data, I wanted to be certain.