estimate_matches: Run lasso estimation on the prepared matrixes
In EBukin/syntheticpanel: Create synthetic panels from cross sectional data

Description Usage Arguments Value Examples

Run lasso estimation on the prepared matrixes

1 2	estimate_matches(x_mat, y_mat, w_mat, x1_mat, n_near = 5, reduced = TRUE, n_folds = 10)

`x_mat, y_mat, w_mat`	matrixes of independent, dependent variables and weights
`x1_mat`	matrixes of independent variables for the prediction sample
`n_near`	number of the nearest observations to derive a random match. If 'n_near' is greater than 'length(match_vector)', minimum out of two is used to create a sample for selecting a random match value.
`reduced`	if TRUE terurns reduced outpur without specific regression details.
`n_folds`	number of folds for cross-validation

In both reduced=TRUE and reduced=FALSE forms, the function returns a list with elements. In the form reduced=TRUE only results of matching are returned:

y0_hat is the vecrtor of predicted values based on the result of the lasso regression with nfolds cross validation and "mse" measure for identifying minimizing value of lambda. It uses x_mat, y_mat, and w_mat for running regression and predictind y0_hat.
y1_hat is the vetor of predicted values produced using the y0_hat regression results and x1_mat independent variables matrix.
match is the dataframe with: columns period_1_id - index of each y1_hat value; column y1_hat - its' value; column period_0_id - position of the nearest match from the y0_hat vector and column y0_hat - value of the nearest match

In the not reduced form the list contains more elements:

items x_mat, y_mat, w_mat and x1_mat from the inputs to the functions.
lambda_cv - result of the cv.glmnet.
fit - result of the glmnet

library(dplyr)
library(purrr)
library(glmnet)
library(syntheticpanel)

# Run estimate of a fake data
XX <- as.matrix(mtcars[, !names(mtcars) %in% "hp"])
YY <- as.matrix(mtcars[, "hp"])
WW <- matrix(rep(1, nrow(mtcars)), ncol = 1, nrow = nrow(mtcars))
XX1 <- as.matrix(mtcars[1:10, !names(mtcars) %in% "hp"])

# Running simple estimation and returning
a <- estimate_matches(x_mat = XX, y_mat = YY, w_mat = WW, x1_mat = XX1, reduced = FALSE, n_near = 5)

# Extract regression coefficients
a$fit %>% coef()
str(a, max.level = 1)

# Developing a sample bootsrtap vector
perm_example <- purrr::map(1:5, ~ sample(1:nrow(YY), nrow(YY), TRUE))

# Running estimations on every single bootstrap permutation vector
a_boot <-
  perm_example %>%
  purrr::map(~ syntheticpanel::estimate_matches(
    x_mat = XX[.x, ],
    y_mat = YY[.x, ],
    w_mat = WW[.x, ],
    x1_mat = XX1,
    reduced = FALSE,
    n_near = 5
  ))

# Exploring the structure
a_boot %>%
  str(max.level = 1)

# Accesssing fit of a specific bootstrap iteration
a_boot[[3]]$fit

# Doing the same with each single bootstrap iteration
a_boot %>%
  map("fit")

# Extracting coefficinets from each single bootstrap iteration
a_boot %>%
  map("fit") %>%
  map(coefficients)

# Combine all coefficients in a table
a_boot %>%
  map("fit") %>%
  map(broom::tidy) %>%
  map(~ select(.x, term, estimate)) %>%
  map2(.y = 1:length(.), ~ rename_at(.x, vars(estimate), list(~ paste0(., "_", .y)))) %>%
  reduce(full_join)