Fit Grid Partition and estimate cell stats

Split the data, one one side train/fit the partition using fit_partition and then on the other estimate subgroup effects.

Tests whether the object is an estimated_partition object.

fit_estimate_partition(
  y,
  X,
  d = NULL,
  tr_split = 0.5,
  max_splits = Inf,
  max_cells = Inf,
  min_size = 3,
  cv_folds = 5,
  potential_lambdas = NULL,
  partition_i = NA,
  verbosity = 0,
  breaks_per_dim = NULL,
  bucket_min_n = NA,
  bucket_min_d_var = FALSE,
  ctrl_method = "",
  pr_cl = NULL,
  alpha = 0.05,
  bump_samples = 0,
  bump_ratio = 1,
  importance_type = "",
  ...
)

is_estimated_partition(x)

Arguments

y

Nx1 matrix of outcome (label/target) data. With multiple core estimates see Details below.
X

NxK matrix of features (covariates). With multiple core estimates see Details below.
d

(Optional) NxP matrix (with colnames) of treatment data. If all equally important they should be normalized to have the same variance. With multiple core estimates see Details below.
tr_split

Number between 0 and 1 or vector of indexes. If Multiple effect #3 and using vector then pass in list of vectors.
max_splits

Maximum number of splits even if splits continue to improve OOS fit
max_cells

Maximum number of cells even if more splits continue to improve OOS fit
min_size

Minimum cell size when building full grid, cv_tr will use (F-1)/F*min_size, cv_te doesn't use any.
cv_folds

Number of CV Folds or a vector of foldids. If m_mode==DS.MULTI_SAMPLE, then a list with foldids per Dataset. Each must be over the training sample
potential_lambdas

potential lambdas to search through in CV
partition_i

Default NA. Use this to avoid CV
verbosity

0 print no message. 1 prints progress bar for high-level loops. 2 prints detailed output for high-level loops. Nested operations decrease verbosity by 1.
breaks_per_dim

NULL (for all possible breaks); K-length vector with # of break (chosen by quantiles); or K-dim list of vectors giving potential split points for non-categorical variables (can put c(0) for categorical). Similar to 'discrete splitting' in CausalTree though their they do separate split-points for treated and controls.
bucket_min_n

Minimum number of observations needed between different split checks
bucket_min_d_var

Ensure positive variance of d for the observations between different split checks
ctrl_method

Method for determining additional control variables. Empty ("") for nothing, "all", "LassoCV", or "RF"
pr_cl

Default NULL. Parallel cluster. Used for:

  1. CVing the optimal lambda,

  2. fitting full tree (at each split going across dimensions),

  3. fitting trees over the bumped samples

  4. for importance weights to estimate models over limited X domains
alpha

Significance threshold for confidence intervals. Default=0.05
bump_samples

Number of bump bootstraps (default 0), or list of such length where each items is a bootstrap sample. If m_mode==DS.MULTI_SAMPLE then each item is a sublist with such bootstrap samples over each dataset. Each bootstrap sample must be over the train split of the data
bump_ratio

For bootstraps the ratio of sample size to sample (between 0 and 1, default 1)
importance_type

Options: single - (smart) redo full fitting removing each possible dimension interaction - (smart) redo full fitting removing each pair of dimensions "" - Nothing
...

Additional params.
x

an R object

Value

An object with class "estimated_partition".

partition

grid_partition obj defining cuts

cell_stats

Cell stats from est_cell_stats$stats on the est sample

importance_weights

Importance weights for each feature

interaction_weights

Interaction weights for each pair of features

lambda

lambda used

is_obj_val_seq

In-sample objective function values for sequence of partitions

complexity_seq

Complexity #s (# cells-1) for sequence of partitions

partition_i

Index of Partition selected in sequence

split_seq

Sequence of partition_splitss. Note that split i corresponds to partition i+1

index_tr

Index of training sample (we might have generated it). Order N

cv_foldid

CV foldids for the training sample (Size of N_tr)

varnames

varnames (or c("X1", "X2",...) if X doesn't have colnames)

est_plan

Fitted EstimatorPlan used.

full_stat_df

Full sample average stats from est_full_stats

True if x is an estimated_partition

Functions

  • is_estimated_partition: is estimated_partition

Multiple estimates

With multiple core estimates (M) there are 3 options (the first two have the same sample across treatment effects).

  1. DS.MULTI_SAMPLE: Multiple pairs of (Y_m,W_m). y,X,d are then lists of length M. Each element then has the typical size The N_m may differ across m. The number of columns of X will be the same across m.

  2. DS.MULTI_D: Multiple treatments and a single outcome. d is then a NxM matrix.

  3. DS.MULTI_Y: A single treatment and multiple outcomes. y is then a NXM matrix.