Enhanced R Package for GEMPACK .HAR and .SL4 File Processing"

In HARplus: Enhanced R Package for 'GEMPACK' .har and .sl4 Files

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  warning = FALSE,
  message = FALSE,
  eval = requireNamespace("HARplus", quietly = TRUE)
)

Overview of HARplus

The HARplus package enhances GEMPACK users' experience by streamlining .har and .sl4 file processing. It efficiently extracts, organizes, and manages dimension structures, ensuring consistency, optimized memory usage, and simplified data manipulation. The package enables fast data merging, aggregation, and transformation while maintaining a consistent interface across GEMPACK file types. Users can easily pivot data into familiar formats and export results in various formats.

A key feature of HARplus is its flexible subtotal level handling, allowing users to selectively retain "TOTAL" values, decomposed components, or both. This ensures precise data extraction for various economic modeling needs without unnecessary redundancy.

Introduction

This vignette covers key functions of HARplus, highlighting its ability to handle multiple inputs for efficient data processing. Designed for economic modelers and GEMPACK practitioners, it simplifies working with multiple datasets and enhances analytical workflows.

Sample data used in this vignette is obtained from the GTAPv7 model and utilizes publicly available data from the GTAP 9 database. For more details, refer to the GTAP Database Archive.

Acknowledgement

The development of HARplus builds upon the foundational work of the HARr package. I sincerely acknowledge and appreciate the contributions of Maros Ivanic, whose HARr package served as the baseline for this development, advancing efficient data handling for GEMPACK users.

GTAPViz: An Extension of HARplus for Visualization

I have developed another package specifically for visualization, particularly for GTAP users: GTAPViz

---

Installation and Loading Example Data

Before proceeding, ensure that HARplus is installed and loaded:

library(HARplus)

Loading Data .HAR and .SL4

First, load the example data using the following commands <load_harx> and <load_sl4x>:

# Paths to the .har files
har_path1 <- system.file("extdata", "TAR10-WEL.har", package = "HARplus")
har_path2 <- system.file("extdata", "SUBT10-WEL.har", package = "HARplus")

# Paths to the .sl4 files
sl4_path1 <- system.file("extdata", "TAR10.sl4", package = "HARplus")
sl4_path2 <- system.file("extdata", "SUBT10.sl4", package = "HARplus")

# Load the .har files using load_harx()
har_data1 <- load_harx(har_path1)
har_data2 <- load_harx(har_path2)

# Load the .sl4 files using load_sl4x()
sl4_data1 <- load_sl4x(sl4_path1)
sl4_data2 <- load_sl4x(sl4_path2)

Extracting Data

This package allows two main ways to enhance the GEMPACK user experience and improve user-friendliness when extracting data:
1. By selecting the variable/header name using <get_data_by_var>.
2. By using dimension patterns (e.g., REG*COMM) with <get_data_by_dims>.

Both functions have similar options:
- Extract a single variable, multiple variables, or all variables (NULL or "ALL") from the datasets.
- Extract data from a single dataset or multiple datasets.
- Rename the experiment (i.e., dataset) name, which is displayed in the Experiment column (this column is automatically added to handle multiple dataset merges). By default, the name is the dataset name.
- Reporting data level (all, only total, only subtotals). - Rename dimension names, which will be used as column names.
- Merge data across multiple datasets within the same patterns.
- For <get_data_by_var>, variables with the same name (e.g., "qo") from different experiments (exp1 and exp2) will be merged into a single final dataframe, with the input source identified in the Experiment column.
- Similarly, for <get_data_by_dims>, data with matching dimension patterns (e.g., REG*COMM) will be merged.

Let's start with the simplest example as follows:

# Extract data for a single variable
data_qo <- get_data_by_var("qo", sl4_data1)
print(head(data_qo[["sl4_data1"]][["qo"]], 4))

# Extract multiple variables from multiple datasets
data_multiple <- get_data_by_var(c("qo", "qgdp"), sl4_data1, sl4_data2)
print(head(data_multiple[["sl4_data1"]][["qo"]], 4))
print(head(data_multiple[["sl4_data2"]][["qo"]], 4))

# Extract all variables separately from multiple datasets
data_list <- get_data_by_var(NULL, sl4_data1, sl4_data2)
print(names(data_list))

# Extract all variables and merge the same variable from multiple datasets
data_all <- get_data_by_var(NULL, sl4_data1, sl4_data2, merge_data = TRUE)
print(names(data_all))

# Return all value levels
data_all <- get_data_by_var("qo", sl4_data1, sl4_data2, subtotal_level = TRUE)
print(head(data_all[["sl4_data1"]][["qo"]], 4))

# Return only TOTAL, drop subtotal
data_total <- get_data_by_var("qo", sl4_data1, sl4_data2, subtotal_level = FALSE)
print(head(data_total[["sl4_data1"]][["qo"]], 4))

# Return only subtotal, drop TOTAL (result is empty if there in subtotal)
data_decomp <- get_data_by_var("qo", sl4_data1, sl4_data2, subtotal_level = "decomposed")
print(head(data_decomp[["sl4_data2"]][["qo"]], 4))

# Rename specific columns
data_col_renamed <- get_data_by_var("qo", sl4_data1, 
                             rename_cols = c(REG = "Region", COMM = "Commodity"))
str(data_col_renamed)

# Rename experiment names
data_exp_renamed <- get_data_by_var("qo", sl4_data1, sl4_data2, 
                             experiment_names = c("EXP1", "EXP2"))
print(names(data_exp_renamed))

# Merge variable data across multiple datasets with custom experiment names
data_merged <- get_data_by_var(, sl4_data1, sl4_data2,
                            experiment_names = c("EXP1", "EXP2"), 
                            merge_data = TRUE,
                            rename_cols = c(REG = "Region", COMM = "Commodity"))
print(head(data_merged$merged[[1]], 4))

Now, let's delve into <get_data_by_dims>. This command offers an additional feature compared to <get_data_by_var>: it allows for patter_mix. For instance, REGCOMM and COMMREG can be treated as equivalent when merging data (this only applies if merge_data = TRUE). You can experiment with the following commands:

# Merge data by dimensions (e.g., REG*COMM != COMM*REG)
data_no_mix <- get_data_by_dims(NULL, sl4_data1, sl4_data2, 
                                merge_data = TRUE, 
                                pattern_mix = FALSE)

# Merge data while allowing interchangeable dimensions (e.g., REG*COMM = COMM*REG)
data_pattern_mixed <- get_data_by_dims(NULL, sl4_data1, sl4_data2, 
                                       merge_data = TRUE, 
                                       pattern_mix = TRUE)

This flexibility is particularly useful when working with datasets where dimension order does not affect the interpretation of the data.

Grouping Data By Dimension

The <group_data_by_dims> function is a powerful tool that categorizes extracted data into meaningful dimension-based groups. This is a key feature of this package, as it allows GEMPACK users (particularly those working with GTAP model results) to merge data into a structured and useful dataframe.

For example, if users want to retrieve all variables defined by REG, they can simply set REG as a priority while also assigning a new column name for REG, such as REG = "Region". The function will then merge all datasets that contain REG as a dimension element into a single dataframe, ensuring that all relevant data is consolidated.

Unlike <get_data_by_dims>, which focuses on extracting data, <group_data_by_dims> organizes and merges data dynamically based on the structure and priority defined by the user.

It is particularly useful because: - It groups extracted data by dimension levels (1D, 2D, 3D, …). - Users can define priority-based merging (e.g., prioritizing REG first, then COMM). The function first attempts to merge all datasets containing REG, then moves on to merge datasets containing COMM. If a dataset contains REG*COMM, it will be included in the REG output. - It automatically renames dimensions (auto_rename = TRUE) to improve dataset compatibility while preserving all variable information. - It transforms data into a structured long format, simplifying further analysis. - It supports merging grouped data across multiple datasets. - It filters subtotal levels, allowing users to retain "TOTAL" values or decomposed components as needed. - If certain datasets cannot be merged, the function generates a detailed report identifying the root of the issue, enabling users to manually manipulate the data if necessary. - This function primarily focuses on 1D and 2D data, ensuring that higher-dimension data (>2D) is merged only with datasets that share the same pattern.

Understanding Priority List in group_data_by_dims

The priority list in <group_data_by_dims> allows users to control how dimensions are grouped and merged.
By defining priorities, the function ensures that datasets containing high-priority dimensions are merged as much as possible, but only if they share the same structure.

1. Single Priority Example (REG only)

If we prioritize only REG, the function will attempt to merge all datasets containing REG as much as possible, while ensuring that datasets with different structures remain separate.

# Define single priority (Only Region-based grouping)
priority_list <- list("Region" = c("REG"))

# Grouping data with a single priority
grouped_data_single <- group_data_by_dims("ALL", sl4_data1, sl4_data2,
                                          priority = priority_list, auto_rename = TRUE)

# Print structure
print(names(grouped_data_single))
print(names(grouped_data_single[["1D"]]))
print(names(grouped_data_single[["2D"]]))

What happens here? - The function tries to merge all datasets containing REG, but only if they share the same data structure. - If some datasets have different structures, they will remain separate. - The merging happens separately in each data dimension (e.g., 1D, 2D, etc.).

2. Multiple Priorities Example (COMM before REG)

If we prioritize COMM first and REG second, datasets containing COMM will be merged first, followed by REG, as much as possible while ensuring datasets with different structures remain separate.

# Define multiple priority levels: Sector first, then Region
priority_list <- list(
  "Sector" = c("COMM", "ACTS"),
  "Region" = c("REG")
)

# Grouping data with multiple priorities
grouped_data_multiple <- group_data_by_dims("ALL", sl4_data1, 
                                            priority = priority_list, 
                                            auto_rename = TRUE)

# Print structure
print(names(grouped_data_multiple))
print(names(grouped_data_multiple[["1D"]]))
print(names(grouped_data_multiple[["2D"]]))

What happens here?
- The function attempts to merge all datasets containing COMM and ACTS first, both of which are merged into the Sector column before attempting to merge other datasets with REG, as long as they share the same structure.
- Therefore, if datasets contain REG*COMM and REG*ACTS, they will be merged into the Sector dataframe, not the REG dataframe. - If datasets cannot be merged due to structural differences, they are kept separate in their respective dimension groups.

Understanding auto_rename in group_data_by_dims

The auto_rename option in <group_data_by_dims> plays a crucial role in ensuring successful and smooth merging of datasets.
When enabled, it automatically renames lower-priority dimensions to "Dim1", "Dim2", etc., which allows datasets with slightly different dimension names to be merged, while still preserving the original dimension structure.

Without auto_rename, datasets with similar but non-identical dimension names may fail to merge, leading to separate outputs instead of a unified dataset.

Example: Grouping Without auto_rename

By default (auto_rename = FALSE), the function keeps all original dimension names.
This means datasets that contain similar but not identical dimensions (e.g., REG*ENDW and REG*END) cannot be merged.

# Define priority: First by Sector (COMM, ACTS), then by Region (REG)
priority_list <- list(
  "Sector" = c("COMM", "ACTS"),
  "Region" = c("REG")
)

# Grouping data without auto_rename
grouped_data_no_rename <- group_data_by_dims("ALL", sl4_data1, 
                                             priority = priority_list, 
                                             auto_rename = FALSE)

# Print structure
print(names(grouped_data_no_rename))
print(names(grouped_data_no_rename[["1D"]]))
print(names(grouped_data_no_rename[["2D"]]))

Without auto_rename, the number of mergeable dataframes will be less compared to when auto_rename = TRUE, as it does not allow merging across different dimension names, even if their structures are otherwise compatible.

Reshaping and Renaming Data

Pivoting Data

The <pivot_data> function transforms long-format data from SL4 or HAR objects into a wide format, making it more suitable for analysis and visualization. This transformation is particularly useful when working with GEMPACK outputs that need to be reshaped for reporting or further processing.

# Extract data to pivot
data_multiple <- get_data_by_var(c("qo", "pca"), sl4_data1)

# Pivot a single column
pivoted_single <- pivot_data(data_multiple, 
                             pivot_cols = "REG")

# Pivot multiple columns
pivoted_multi <- pivot_data(data_multiple, 
                            pivot_cols = c("COMM", "REG"))

Pivot Data Hierarchy

While pivot_data provides basic pivoting functionality, pivot_data_hierarchy offers enhanced capabilities for creating hierarchical pivot tables similar to those found in spreadsheet applications. This function is particularly useful when you need to maintain dimensional hierarchies in your output or create Excel-ready pivot tables.

Key differences from regular pivoting:

• Maintains hierarchical relationships between pivot columns
• Supports direct Excel export with properly formatted headers
• Preserves dimension order as specified in pivot_cols
• Creates nested column headers reflecting the hierarchy
• Must be exported to XLSX only using the command option <export = TRUE, file_path = "./path/to/output">. This pivot table cannot be exported with the command.

# Create hierarchical pivot without export
pivot_hier <- pivot_data_hierarchy(data_multiple, 
                                  pivot_cols = c("REG", "COMM"))

# Create and export to Excel in one step
pivot_export <- pivot_data_hierarchy(data_multiple, 
                                   pivot_cols = c("REG", "COMM"),
                                   export = TRUE,
                                   file_path = file.path(tempdir(), "pivot_output.xlsx"))

The hierarchy in the resulting pivot table follows the order specified in pivot_cols. For example, when using c("REG", "COMM"), the output will show:

• First level: Region (REG)
• Second level: Commodity (COMM) nested under each region
• Values arranged according to this hierarchy

When exported to Excel, the hierarchical structure is automatically formatted with:

• Properly aligned multi-level column headers
• Bold formatting for header levels
• Clean sheet names for multiple variables

Rename Dimensions

The rename_dims function provides flexible renaming capabilities for dimensions in SL4 or HAR objects. You can rename either dimension names, list names, or both.

# Define a renaming map
mapping_df <- data.frame(
 old = c("REG", "COMM"),
 new = c("Region", "Commodity")
)

# Rename dimensions only
renamed_dims <- rename_dims(sl4_data1, mapping_df)

# Rename both dimensions and list names
renamed_both <- rename_dims(sl4_data1, mapping_df, rename_list_names = TRUE)

The mapping dataframe must have two columns: the first for current names and the second for new names. The function preserves data structure while updating dimension labels according to the specified mapping.

Exporting Data to CSV/STATA/TEXT/RDS/XLSX

The export_data function allows you to export SL4 and HAR data into various formats. Supported formats include: "csv", "xlsx", "stata", "txt", and "rds". You can export single data frames or multi-variable results while preserving their structure.

# Extract data
data_multiple <- get_data_by_var(c("qo", "pca"), sl4_data1)

# Export 
export_data(data_multiple, file.path(tempdir(), "output_directory"), 
           format = c("csv", "xlsx", "stata", "txt", "rds"),
           create_subfolder = TRUE,
           multi_sheet_xlsx = TRUE)

Exploring Data Structure

Since this package is designed to handle multiple inputs with a similar structure, such as simulation outputs from the GTAP model with different shocks or experiments, the first important step is to understand the data structure. This process can also be useful even with a single input, as it helps in analyzing the data shape and dimension size of each variable.

There are a couple of commands available to illustrate the data structure, all of which can be applied to both .har and .sl4 files in the same manner.

Variable Structure

To get a summary of variable names, dimension counts, dimension patterns, and optionally the column and observation counts for one or multiple variables from a single or multiple datasets (return as separate lists):

# (1) Getting all variables from the input file
vars_har_sum <- get_var_structure("ALL", har_data1)
vars_sl4_sum <- get_var_structure(, har_data1)

# (2) Getting selected variables
var_sl4_sum <- get_var_structure(c("pds","pfd","pms"), sl4_data1)
print(head(var_sl4_sum[["sl4_data1"]], 4))


# (3) Including column size and number of observation in the summary
var_sl4_sum <- get_var_structure(c("pds","pfd","pms"), sl4_data1, sl4_data2, 
                                 include_col_size = TRUE)
print(head(var_sl4_sum[["sl4_data1"]], 4))

Understanding the data structure is crucial for aggregating data across multiple experiments (inputs). Even when using the same variables, different experiments may have varying column sizes or data structures due to factors such as subtotal effects and other experimental settings. These discrepancies can lead to errors if the merging process relies solely on variable names without accounting for structural differences.

To compare data structures across multiple experiments, use the following command:

# (1) Comparing all variable structures across experiments 
vars_comparison <- compare_var_structure(
  variables = "ALL", sl4_data1, sl4_data2
)
print(vars_comparison$match[1:2, ])

# (2) Comparing selected variable structures across experiments 
var_comparison <- compare_var_structure(
  variables = c("pds", "pms"), sl4_data1, sl4_data2
)
print(var_comparison$match[1:2, ])

This function returns a list containing: - match: A data frame of variables with identical dimension names and structures across inputs. - diff (if any): A data frame listing variables with structural mismatches. Note: If this list appears, you may need to focus on handling these variables. It is designed to report potential issue-causing variables when merging datasets.

Additionally, the function includes the <keep_unique> option (default: FALSE) which allows users to extract variables with unique names and structures across all inputs. This is particularly useful when inputs contain different sets of variables that need to be combined into a final dataset.

# (3) Extracting unique variable structures
unique_vars <- compare_var_structure(, 
                                     sl4_data1, sl4_data2,
  keep_unique = TRUE
)
print(unique_vars$match[1:10, ])

This function returns a list containing: - match: A data frame displaying distinct variable structures found across inputs. - diff (if any): A data frame detailing how structures differ between inputs.

To sum up, - keep_unique = FALSE → Checks whether variables match across inputs. - keep_unique = TRUE → Extracts all unique variable structures, regardless of whether they match, while reports any variables that do not align.

Dimension Structure

The following commands can be used to retrieve unique dimension names as patterns and their corresponding elements:

# (1) Extracting dimension patterns (e.g., REG*COMM*ACTS)
dims_strg_har <- get_dim_patterns(har_data1, har_data2)
print(dims_strg_har[1:4, ])

# (2) Extracting unique dimension patterns (e.g., REG*COMM*ACTS)
dims_strg_har <- get_dim_patterns(har_data1, har_data2,
                                  keep_unique =TRUE)
print(dims_strg_har[1:4, ])

# (2) Extracting dimension elements e.g., REG, COMM, ACTS
dims_strg_har_uniq <- get_dim_elements(har_data1, har_data2, 
                                       keep_unique =TRUE)
print(dims_strg_har_uniq[1:4, ])

Patterns vs. Elements

Patterns represent structured dimension names (e.g., "REG*COMM*ACTS"), while elements extract individual dimension elements (e.g., "REG", "COMM", "ACTS").

HARplus documentation built on April 4, 2025, 4:14 a.m.