impute_simulated: Imputation algorithm tester on simulated data

Description Usage Arguments Details Value Examples

View source: R/impute_simulated.R

Description

impute_simulated tests the imputation quality of all missing data imputation algorithms on matrices with various missing data patterns, using various metrics

Usage

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
impute_simulated(
  rownum,
  colnum,
  cormat,
  n.iter = 10,
  MD_pattern,
  NA_fraction,
  min_PDM = 10,
  assumed_pattern = NA
)

Arguments

rownum

Number of rows (samples) in the original dataframe (Rows output from the get_data function)

colnum

Number of rows (variables) in the original dataframe (Columns output from the get_data function)

cormat

Correlation matrix of the original dataframe (Corr_matrix output from the get_data function)

n.iter

Number of iterations to perform with default 10.

MD_pattern

Missing data pattern in the original dataset (MD_Pattern output from the get_data function)

NA_fraction

Fraction of missingness in the original dataset (Fraction_missingness output from the get_data function)

min_PDM

All patterns with number of observations less than this number will be removed from the missing data generation. This argument is necessary to be carefully set, as the function will fail or generate erroneous missing data patterns with very complicated missing data patterns. The default is 10, but for large datasets this number needs to be set higher to avoid errors.

assumed_pattern

Vector of missingness types (must be same length as missingness fraction per variable). If this input is specified, the function will spike in missing datapoints in a MAP pattern as well.

Details

This function tests the imputation accuracy of the a curated list of missing data imputation algorithms (16 algorithms at the moment) by comparing the original simulated matrix with no missingness and the imputed matrices generated by the algorithm using the matrices with MCAR, MAR, MNAR and (optionally) MAP missingness patterns. The function calculates root-mean-square error (RMSE), mean absolute error (MAE), Kolmogorov-Smirnov test statistics D (KS) between the imputed datapoints and the original datapoints (that were subsequently set to missing) for each missing data imputation algorithm. The function will calculate average computation time per method as well. The function will automatically detect whether there is a MAP matrix in the list and calculate metrics for all matrices provided in the list. Important! All statistics output by this function are calculated for ALL missing values across the dataset, not by variable.

Value

Imputation_metrics_raw

Raw RMSE, MAE, KS and computation time values per method, per missingness pattern, per iteration)

Imputation_metrics_means

RMSE, MAE, KS and computation time means per method and missingness pattern

Plot_TIME

Boxplot of computation time values per missing data imputation algorithm

Plot_RMSE

Faceted boxplot of RMSE values per missingness pattern and missing data imputation algorithm

Plot_MAE

Faceted boxplot of MAE values per missingness pattern and missing data imputation algorithm

Plot_KS

Faceted boxplot of KS values per missingness pattern and missing data imputation algorithm

Examples

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
## in case there is no assumed missingness pattern per variable
# wrap <- impute_simulated(rownum = metadata$Rows,
#        colnum = metadata$Columns,
#        cormat = metadata$Corr_matrix,
#        MD_pattern = metadata$MD_Pattern,
#        NA_fraction = metadata$Fraction_missingness,
#        min_PDM = 10,
#        n.iter = 50)

## in case there is a pre-defined assumed pattern
# wrap <- impute_simulated(rownum = metadata$Rows,
#        colnum = metadata$Columns,
#        cormat = metadata$Corr_matrix,
#        MD_pattern = metadata$MD_Pattern,
#        NA_fraction = metadata$Fraction_missingness,
#        min_PDM = 10,
#        assumed_pattern = c('MAR','MAR','MCAR','MCAR',
#                          'MNAR','MCAR','MAR','MNAR',
#                          'MCAR','MNAR','MCAR'),
#        n.iter = 50)

missCompare documentation built on Dec. 1, 2020, 9:09 a.m.