• Data Release Preparation Tool
  • Description
  • Installation
  • Usage
    • CLI
    • Recipe Definition
  • Example
  • Thanks

⚠️ This is currently at beta development stage and likely has a lot of bugs. Please use the issue tracker to report an bugs or feature requests.

Command-line tool for preparing a dataset for publishing by dropping, renaming, scaling, and obfuscating columns defined in a recipe.

After performing the operations defined in the recipe the tool generates the transformed dataset version and a CSV report listing the performed actions.

The tool can be installed using pip:

pip install drpt

Usage: drpt [OPTIONS] RECIPE_FILE INPUT_FILE

Options:
  -d, --dry-run           Generate only the report without the release dataset
  -v, --verbose           Verbose [Not implemented]
  -n, --nrows TEXT        Number of rows to read from a CSV file. Doesn't work
                          with parquet files.
  -l, --limits-file PATH  Limits file
  -o, --output-dir PATH   Output directory. The default output directory is
                          the same as the location of the recipe_file.
  --version               Show the version and exit.
  --help                  Show this message and exit.

The recipe is a JSON formatted file that includes what operations should be performed on the dataset. For versioning purposes, the recipe also contains a version key which is appended in the generated filenames and the report.

Default recipe:

{
  "version": "",
  "actions": {
    "drop": [],
    "drop-constant-columns": false,
    "obfuscate": [],
    "disable-scaling": false,
    "skip-scaling": [],
    "sort-by": [],
    "rename": []
  }
}

The currently supported actions, performed in this order, are as follows:

• drop: Column deletion
• drop-constant-columns: Drops all columns that containt only one unique value
• obfuscate: Column obfuscation, where the listed columns are treated as categorical variables and then integer coded.
• Scaling: By default all columns are Min/Max scaled
  • disable-scaling: Can be used to disable scaling for all columns
  • skip-scaling: By default all columns are Min/Max scaled, except those excluded (skip-scaling)
• sort-by: Sort rows by the listed columns
• rename: Column renaming

All column definitions above support regular expressions.

The drop action is defined as a list of column names to be dropped.

This is a boolean action, which when set to true will drop all the columns that have only a single unique value.

The obfuscate action is defined as a list of column names to be obfuscated.

By default, the tool Min/Max scales all numerical columns. This behavior can be disabled for all columns by setting the disable-scaling action to true. If scaling must be disabled for only a set of columns these columns can be defined using the skip-scaling action, as a list of column names.

This is a list of column names by which to sort the rows. The order in the list denotes the sorting priority.

The rename action is defined as a list of objects whose key is the original name (or regular expression), and their value is the target name. When the target uses matched groups from the regular expression those can be provided with their group number prepended with an escaped backslash (\\1) [see example below].

{
  //...
  "rename": [{"original_name": "target_name"}]
  //...
}

Input CSV file:

test1,test2,test3,test4,test5,test6,test7,test8,test9,foo.bar.test,foo.bar.test2,const
1.1,1,one,2,0.234,0.3,-1,a,e,1,1,1
2.2,2,two,2,0.555,0.4,0,b,f,2,2,1
3.3,3,three,4,0.1,5,1,c,g,3,3,1
2.22,2,two,4,1,0,2.5,d,h,4,4,1

Recipe:

{
  "version": "1.0",
  "actions": {
    "drop": ["test2", "test[8-9]"],
    "drop-constant-columns": true,
    "obfuscate": ["test3"],
    "skip-scaling": ["test4"],
    "sort-by": ["test4", "test3"],
    "rename": [
      { "test1": "test1_renamed" },
      { "test([3-4])": "test\\1_regex_renamed" },
      { "foo[.]bar[.].*": "foo" }
    ]
  }
}

Generated CSV file:

test3_regex_renamed,test4_regex_renamed,test1_renamed,test5,test6,test7,foo_1,foo_2
0,2,0.0,0.1488888888888889,0.06,0.0,0.0,0.0
2,2,0.5000000000000001,0.5055555555555556,0.08,0.2857142857142857,0.3333333333333333,0.3333333333333333
1,4,1.0,0.0,1.0,0.5714285714285714,0.6666666666666666,0.6666666666666666
2,4,0.5090909090909091,1.0,0.0,1.0,1.0,1.0

Report:

,action,column,details
0,recipe_version,,1.0
1,drpt_version,,0.6.3
2,DROP,test2,
3,DROP,test8,
4,DROP,test9,
5,DROP_CONSTANT,const,
6,OBFUSCATE,test3,"{""one"": 0, ""three"": 1, ""two"": 2}"
7,SCALE_DEFAULT,test1,"[1.1,3.3]"
8,SCALE_DEFAULT,test5,"[0.1,1.0]"
9,SCALE_DEFAULT,test6,"[0.0,5.0]"
10,SCALE_DEFAULT,test7,"[-1.0,2.5]"
11,SCALE_DEFAULT,foo.bar.test,"[1,4]"
12,SCALE_DEFAULT,foo.bar.test2,"[1,4]"
13,SORT,"['test4', 'test3']",
14,RENAME,test1,test1_renamed
15,RENAME,test3,test3_regex_renamed
16,RENAME,test4,test4_regex_renamed
17,RENAME,foo.bar.test,foo_1
18,RENAME,foo.bar.test2,foo_2

This tool was made possible with Pandas, PyArrow, jsonschema, and of course Python.