How to get the actual "matched dataset"?

[PetrichorSjj]

I'm using the code below

library(ggplot2)
library(CausalGPS)
library(memoise)
library(data.table)

# Load the data
data_file <- "xxx.csv"
data <- read.csv(data_file)
seed_val <- 249
set.seed(seed_val)

# set up the params
example <- "matching_1"

#------To reproduce the same results, do not change below this line.------------

if (example == "matching_1") {
  p_q1 <- 0.01
  p_q2 <- 0.99
  delta_n <- 1.7
} else if (example == "matching_2"){
  p_q1 <- 0.05
  p_q2 <- 0.95
  delta_n <- 0.9
} else if (example == "matching_3"){
  p_q1 <- 0.10
  p_q2 <- 0.90
  delta_n <- 1.9
} else {
  stop("Example type is not defined.")
}

sc_name <- example
ci_appr <- "matching"
gps_density <- "normal"

max_it <- 10
scale_val <- 1

# the range of my exposure factor is [1, 300]
exposure_min <- min(data$fields_num)  
exposure_max <- max(data$fields_num)  

# calculate the number of bins（Note: it's 79)
L <- ceiling((exposure_max - exposure_min) / (2 * delta_n))

# setup the confounders
confounders <- c("records_num", "views_num", "tables_num")

# add id column
data$id <- 1:nrow(data)

# generate pseudo population
start_time <- proc.time()
ps_pop_obj_1 <- generate_pseudo_pop(data[, c("id", "fields_num")],
                                    data[, c("id", confounders)],
                                    ci_appr = ci_appr,
                                    gps_density = gps_density,
                                    bin_seq = NULL,
                                    exposure_trim_qtls = c(p_q1,
                                                           p_q2),
                                    use_cov_transform = TRUE,
                                    params = list(xgb_max_depth = c(3,4,5),
                                                  xgb_nrounds = seq(10, 40, 1)),
                                    sl_lib = c("m_xgboost"),
                                    nthread = 10,
                                    covar_bl_method = "absolute",
                                    covar_bl_trs = 0.1,
                                    covar_bl_trs_type= "maximal",
                                    max_attempt = max_it,
                                    dist_measure = "l1",
                                    delta_n = delta_n,
                                    scale = scale_val)
end_time <- proc.time()
end_time - start_time

# covariate balance test
plot(ps_pop_obj_1, include_details = TRUE)

# get the matched set?（I'm not sure if it's correct, but the result is weird)
# the max value of fields_num(which is the exposure factor of my dataset is 43 and the min value is 6, which is very different from the original range [1, 300], I don't know the reason.
data <- merge(data[, c("id", "ttu")], ps_pop_obj_1$pseudo_pop, by = "id")

# estimate erf and plot
start_time <- proc.time()
m_Y <- as.numeric(data$ttu)
m_w <- as.numeric(data$fields_num)
counter_weight <- as.numeric(data$counter_weight)
erf_obj <- estimate_npmetric_erf(m_Y,
                                 m_w,
                                 counter_weight,
                                 bw_seq = seq(0.2,2,0.2),
                                 w_vals = seq(10, 40, 1),
                                 nthread = 1)
end_time <- proc.time()
end_time - start_time
plot(erf_obj)

• Question 1: How to get the matched dataset(I think it should be a LxN dimensions dataset, but the shape（of dataset I get from the code below） is (41092, 14)

data <- merge(data[, c("id", "ttu")], ps_pop_obj_1$pseudo_pop, by = "id")

• Question 2: How to get the erf graph? When I ran the code below, I got the following error:
  Error in checkForRemoteErrors(val) :
  one node produced an error: NA/NaN/Inf in foreign function call (arg 4)

start_time <- proc.time()
m_Y <- as.numeric(data$ttu)
m_w <- as.numeric(data$fields_num)
counter_weight <- as.numeric(data$counter_weight)
erf_obj <- estimate_npmetric_erf(m_Y,
                                 m_w,
                                 counter_weight,
                                 bw_seq = seq(0.2,2,0.2),
                                 w_vals = seq(10, 40, 1),
                                 nthread = 1)
end_time <- proc.time()
end_time - start_time
plot(erf_obj)

• Question 3：Why the range of the exposure factor (In this case, it's the "fields_num"）changed so much?
  I think if the original range is [1, 300], and I get 79 bins using the caliper, what I can get is something like this:
• medium exposure value of the first bin + outcome value
• medium exposure value of the second bin + outcome value
  ...
• medium exposure value of the last bin + outcome value
  so I can get 79 pairs of exposure value and outcome value, and then I use kernel smoother to get the erf graph.

I have read the papers below carefully, but I still can't get the answer of the questions above, I'm sorry if I missed them.
https://arxiv.org/pdf/1812.06575
https://arxiv.org/pdf/2310.00561

【Reference】
The result of summary(ps_pop_obj_1)
Number of data samples: 41092
Number of iterations: 1
Effective sample size:
Achieved: 3583.31032217326
Min recommended: 4109.2

As you can see above, I have 50000 samples, so I think the "Number of data samples" means part of the samples are trimmed.
I don't know what is the "Effective sample size" is.

But the covariate balance test is passed.