SDTL Function Library
SDTL version 0.4; Revised Nov. 4, 2020
The function library is stored in a JSON file that describes how to translate functions in various source languages into their SDTL equivalents.
Functions are frequently used in statistical languages to perform common calculations. Most functions require at least one parameter or argument used in the calculation, which may be specified as a variable name or a constant. Functions are usually specified as a function name followed by a list of parameters surrounded by parentheses, like “log(x)”. Parameters may be identified by the order in which they appear, by preceding the value with a name (e.g. “cut(groups=4)”), or a combination of position and names.
SDTL uses functions to describe arithmetic operations (addition, subtraction, etc.) and logical comparisons (“x>y”).
SDTL functions are grouped into four types shown in Table 1. “Horizontal” functions return a value for each row in a dataframe by using constants or variables found on the same row. “Vertical” functions are used in the SDTL Aggregate command, which adds new variables to every row in a group without changing the number of rows in the dataframe. The SDTL Collapse command also aggregates across rows, but it returns only one row per group. Although every function in SDTL has a unique name, source languages sometimes use the same function name in more than one way. For example, “mean()” may be used to compute the average of several variables on each row (horizontal) or to compute the mean of a single variable across groups of rows (vertical or collapse).
Table 1. SDTL function types
horizontal |
Horizontal functions perform actions on one or more variables on each row. |
vertical |
Vertical functions create a new variable by acting upon multiple rows in an SDTL Aggregate command. A new variable is added to every row in the dataframe within groups specified by a groupByVariables property. |
collapse |
Collapse functions create a new variable by acting upon multiple rows in an SDTL Collapse command. A new dataframe is created with one row per group as defined by a groupByVariables property. |
logical |
Logical functions are used in conditions. |
The Function Library maps the SDTL version of a function to the same function in various source languages. Table 2 shows the JSON properties used to describe the SDTL version of a function. The “pseudocode” property is used to translate an SDTL function into natural language. For example, the SDTL modulo function is described as “Remainder of EXP1 divided by EXP2”. “EXP1” and “EXP2” are names of parameters that will be replaced by variable names or values. For example, if “mod(varX,3)” is found in a source script, the natural language translation will be: “Remainder of varX divided by 3.” “ReturnType” is the data type of the result of the function.
Each parameter in the SDTL version of a command is described by three properties. “Param” is the name of the property in SDTL. For convenience, we use EXP1, EXP2, etc. for parameter names. The “type” of an SDTL parameter may be a data type, like numeric or string, or it may be the type of an element in SDTL. SDTL types VariableListExpression and ValueListExpression play an important role in the Function Library, which is discussed below.
Table 2. SDTL properties in the Function Library
SDTLname |
Name of the function in SDTL |
|
Pseudocode |
A template for representing this function in natural language |
|
definition |
A description of the computation performed by the function |
|
notes |
Notes about using the function |
|
n_arguments |
Number of parameters |
|
returnType |
Data type returned by this function: numeric, string, boolean |
|
parameters |
Parameters used in the function. Each parameter is described by these properties: |
|
param |
parameter name: EXP1, EXP2, … |
|
isRequired |
Whether this parameter must be present (true) or is optional (false). |
|
defaultValue |
The default value is used when a value of this parameter is null. |
|
type |
This field is used when a specific SDTL $type must be used. For example, if the parameter is a list of variable names, the type must be ValueListExpression. |
The SDTL properties of a function are followed by arrays of source language functions in each of the supported source languages: SPSS, Stata, SAS, R, and Python. Every SDTL function may correspond to one or more functions in each source language. Properties defining a source language function are shown in Table 3.
Table 3. Properties of Source Language function definitions in the Function Library |
||
“function_name” |
Name of the function in the source language. |
|
“syntax” |
Syntax used in the source language. |
|
“URL” |
URL of a description of the function in a source language. |
|
“parameters” |
A list of parameters used in the function. Each parameter has the following properties: |
|
isRequired |
This parameter is required in the source language: true or false. |
|
defaultValue |
Value to use when no value is given. Note that the default value for a source language may be different from the default in SDTL. |
|
param |
Name of the parameter in SDTL: EXP1, EXP2, … in the order specified in SDTL. |
|
name |
Name of the parameter in the source language. Omitted or set to null for parameters identified by position only. |
|
position |
Position of the parameter in the source language. May be omitted or set to null for parameters identified by name only. When a parameter is given before the function name, eg., “param.function()”, it is assigned position 0. |
|
index_offset |
Used with Python to indicate when the parameter must be converted from 0-indexing to 1-indexing. |
The name of the function in the source language is given in the “function_name” property.
The “syntax” and “URL” properties are included as aids to programmers. We use “syntax” to provide a template for translating the source language function into its SDTL equivalent, such as “RND(EXP1, EXP2)”.
Parameters used in the source language versions of functions are described with four properties: “required”, “param”, “name”, “position”. “Param” is the SDTL name for the parameter (EXP1, EXP2, etc.), which is used to match it to the SDTL version of the command. The “required” property describes whether the parameter is required for conversion to SDTL. When a parameter takes a default value if it is omitted, the “required” property is set to the default value. Parameters are always given in the order specified in SDTL (EXP1, EXP2, EXP3…), and the “position” property is used to show the order in the source language.
Function arguments in source languages may be identified by the order in which they appear, by name, or both. In Python, parameters to functions may be identified by position or name, but references by position may not follow references by name. For example, we can generate a random number between .5 and .7 using the numpy library of Python with either “numpy.random.uniform(.5,.7)” or “numpy.random.uniform(low=.5,high=.7)”.
Some functions operate on a list of variables or values, which are described in SDTL with VariableListExpression or ValueListExpression. For example, the Stata command
egen varAvg =rowmean(varA - var99)
sets varAvg equal to the average of all variables from varA to var99. The number of variables included in the average depends upon the number and ordering of variables in the data set. It would be impossible to specify this function in SDTL if every variable was considered a parameter. In SDTL a list of variables of any length can be described by a VariableListExpression, which allows us to treat variable lists as a single parameter in the Function Library.
Table 4 illustrates in tabular form the description of a complicated function in the R language, which corresponds to the SDTL “str_replace” function.
Note that the order of the parameters in R is different from the order in SDTL. We can see this in the “syntax” property, which begins “sub(EXP2, EXP3, EXP1, …”. Parameters are given in SDTL order in the Function Library, but the “position” properties show that the order in R is different.
The “required” property shows that EXP4 and EXP5 have default values that are used in SDTL when these parameters are omitted in the R script.
Parameters EXP4 and EXP5 are named properties in R, “ignore.case=” and “fixed=” respectively.
The syntax field shows that the R function has two parameters (“perl=” and “useBytes=”) that are not currently implemented in SDTL.
Note that the value “true” for “required” in parameter EXP5 is a default value to be used if the “fixed=” parameter is omitted in the R script. We use “yes” and “no” as values for “required”, because “true” and “false” are used as default values for boolean variables.
Table 4. Example Source Language function definitions
“function _name” |
“sub” |
||||
“syntax” |
“sub(EXP2, EXP3, EXP1, ignore.case = EXP4, perl = FALSE, fixed = EXP5, useBytes = FALSE)” |
||||
“URL” |
“https://www.rdocumentation.org/packages/base/versions/3.6.2/topics/grep” |
||||
“parameters” |
|||||
“isRequired” |
“true” |
“true” |
“true” |
“false” |
“false” |
“defaultValue” |
“false” |
“true” |
|||
“param” |
“EXP1” |
“EXP2” |
“EXP3” |
“EXP4” |
“EXP5” |
“name” |
null |
null |
null |
“ignore.case” |
“fixed” |
“position” |
3 |
1 |
2 |
4 |
6 |
Python differs from the other languages in the way that it refers to elements in sequences. Python is “0 indexed,” which means that the index of the first element in a sequence is 0. The other languages use 1 for the first element in a sequence. In addition, ranges of elements in Python, such as “4:7”, are inclusive on the left and exclusive on the right, where other languages are inclusive on both sides. The SDTL function “substr_by_position” illustrates this difference. Substring is a common function used to extract part of a larger string. In R “str_sub(“ABCDEFGH”, 4,7)” will return the part of the string “ABCDEFGH” beginning at position 4 and ending at position 7, i.e. “DEFG”. The equivalent function in Python “str.slice(start=4, stop=7)” returns “EFG”. In Python the “A” in “ABCDEFGH” is at position 0, which puts “E” in position 4, and “stop=7” tells Python to exclude the letter at position 7. We signal this difference in Python by adding an index_offset property that may be used with any parameter in the Python section of the Function Library.
Example: Random Number Functions in SDTL, SPSS, and Python
{
"horizontal": [
{
"SDTLname": "random_variable_uniform",
"Pseudocode": "random number between EXP1 and EXP2 with a uniform distribution",
"n_arguments": "2",
"returnType": "numeric",
"parameters": [
{
"param": "EXP1",
"isRequired": "true",
"defaultValue": 0.0,
"type": "num"
},
{
"param": "EXP2",
"isRequired": "true",
"defaultValue": 1.0,
"type": "num"
}
],
"SPSS": [
{
"function_name": "RAND_NUM",
"syntax": "RAND_NUM(EXP1, EXP2)",
"URL": null,
"parameters": [
{
"isRequired": true,
"defaultValue": null,
"param": "EXP1",
"name": null,
"position": "1"
},
{
"isRequired": true,
"defaultValue": null,
"param": "EXP2",
"name": null,
"position": "2"
}
]
},
{
"function_name": "RV.UNIFORM",
"syntax": "RV.UNIFORM(EXP1, EXP2)",
"URL": null,
"parameters": [
{
"isRequired": true,
"defaultValue": null,
"param": "EXP1",
"name": null,
"position": "1"
},
{
"isRequired": true,
"defaultValue": null,
"param": "EXP2",
"name": null,
"position": "2"
}
]
}
],
"Python": [
{
"function_name": "numpy.random.uniform",
"syntax": "numpy.random.uniform(low=EXP1, high=EXP2)",
"URL": "https://numpy.org/devdocs/reference/random/generated/numpy.random.uniform.html#numpy.random.uniform",
"parameters": [
{
"isRequired": false,
"defaultValue": 0.0,
"param": "EXP1",
"name": "low",
"position": "1"
},
{
"isRequired": false,
"defaultValue": 1.0,
"param": "EXP2",
"name": "high",
"position": "2"
}
]
}
]
}
],
"vertical": [
"..."
],
"collapse": [
"..."
],
"logical": [
"..."
]
}