25 KiB
| id | title | sidebar_label |
|---|---|---|
| expressions | Expressions | Expressions |
Overview
You can use expressions in multiple places in OpenRefine to extend data cleanup and transformation. Expressions are available with the following functions:
-
:
- (click “change” after they have been created to bring up an expressions window)
-
:
-
:
- .
In the expressions editor window you have the opportunity to select a supported language. The default is GREL (General Refine Expression Language); OpenRefine also comes with support for Clojure and Jython. Extensions may offer support for more expressions languages.
These languages have some syntax differences but support many of the same variables. For example, the GREL expression value.split(" ")[1] would be written in Jython as return value.split(" ")[1].
This page is a general reference for available functions, variables, and syntax. For examples that use these expressions for common data tasks, look at the Recipes section on the Wiki.
Expressions
There are significant differences between OpenRefine's expressions and the spreadsheet formulas you may be used to using for data manipulation. OpenRefine does not store formulas in cells and display output dynamically: OpenRefine’s transformations are one-time operations that can change column contents or generate new columns. These are applied using variables such as value or cell to perform the same modification to each cell in a column.
Take the following example:
| ID | Friend | Age |
|---|---|---|
| 1. | John Smith | 28 |
| 2. | Jane Doe | 33 |
Were you to apply a transformation to the “friend” column with the expression
value.split(" ")[1]
OpenRefine would work through each row, splitting the “friend” values based on a space character. The value for row 1 is “John Smith” so the output would be “Smith” (as "[1]" selects the second part of the created output); the value for row 2 is “Jane Doe” so the output would be “Doe”. Using variables, a single expression yields different results for different rows. The old information would be discarded; you couldn't get "John" and "Jane" back unless you undid the operation in the History tab.
For another example, if you were to create a new column based on your data using the expression row.starred, it would generate a column of true and false values based on whether your rows were starred at that moment. If you were to then star more rows and unstar some rows, that data would not dynamically update - you would need to run the operation again to have current true/false values.
Note that an expression is typically based on one particular column in the data - the column whose drop-down menu is first selected. Many variables are created to stand for things about the cell in that “base column” of the current row on which the expression is evaluated. There are also variables about rows, which you can use to access cells in other columns.
The expressions editor
When you select a function that accepts expressions, you will see a window overlay the screen with what we call the expressions editor.
The expressions editor offers you a field for entering your formula and shows you a preview of its transformation on your first few rows of cells.
There is a dropdown menu from which you can choose an expression language. The default at first is GREL; if you begin working with another language, that selection will persist across OpenRefine. Jython and Clojure are also offered with the installation package, and you may be able to add more language support with third-party extensions and customizations.
There are also tabs for:
- History, which shows you formulas you’ve recently used from across all your projects
- Starred, which shows you formulas from your History that you’ve starred for reuse
- Help, a quick reference to GREL functions.
Starring formulas you’ve used in the past can be helpful for repetitive tasks you’re performing in batches.
You can also choose how formula errors are handled: replicate the original cell value, output an error message into the cell, or ouput a blank cell.
Regular expressions
OpenRefine offers several fields that support the use of regular expressions (regex), such as in a or a operation. GREL and other expressions can also use regular expression markup to extend their functionality.
If this is your first time working with regex, you may wish to read this tutorial specific to the Java syntax that OpenRefine supports. We also recommend this testing and learning tool.
GREL-supported regex
To write a regular expression inside a GREL expression, wrap it between a pair of forward slashes (/) much like the way you would in Javascript. For example, in
value.replace(/\s+/, " ")
the regular expression is \s+, and the syntax used in the expression wraps it with forward slashes (/\s+/). Though the regular expression syntax in OpenRefine follows that of Java (normally in Java, you would write regex as a string and escape it like "\s+"), a regular expression within a GREL expression is similar to Javascript.
Do not use slashes to wrap regular expressions outside of a GREL expression.
On the GREL functions page, functions that support regex will indicate that with a “p” for “pattern.” The GREL functions that support regex are:
Jython-supported regex
You can also use regex with Jython expressions, instead of GREL, for example with a :
python import re g = re.search(ur"\u2014 (.*),\s*BWV", value) return g.group(1)
Clojure-supported regex
Clojure uses the same regex engine as Java, and can be invoked with re-find, re-matches, etc. You can use the #"pattern" reader macro as described in the Clojure documentation. For example, to get the nth element of a returned sequence, you can use the nth function:
clojure (nth (re-find #"\u2014 (.*),\s*BWV" value) 1)
Variables
Most OpenRefine variables have attributes: aspects of the variables that can be called separately. We call these attributes “member fields” because they belong to certain variables. For example, you can query a record to find out how many rows it contains with row.record.rowCount: rowCount is a member field specific to the record variable, which is a member field of row. Member fields can be called using a dot separator, or with square brackets (row["record"]). The square bracket syntax is also used for variables that can call columns by name, for example, cells["Postal Code"].
| Variable | Meaning |
|---|---|
value |
The value of the cell in the current column of the current row (can be null) |
row |
The current row |
row.record |
One or more rows grouped together to form a record |
cells |
The cells of the current row, with fields that correspond to the column names (or row.cells) |
cell |
The cell in the current column of the current row, containing value and other attributes |
cell.recon |
The cell's reconciliation information returned from a reconciliation service or provider |
rowIndex |
The index value of the current row (the first row is 0) |
columnName |
The name of the current cell's column, as a string |
Row
The row variable itself is best used to access its member fields, which you can do using either a dot operator or square brackets: row.index or row["index"].
| Field | Meaning |
|---|---|
row.index |
The index value of the current row (the first row is 0) |
row.cells |
The cells of the row, returned as an array |
row.columnNames |
An array of the column names of the project. This will report all columns, even those with null cell values in that particular row. Call a column by number with row.columnNames[3] |
row.starred |
A boolean indicating if the row is starred |
row.flagged |
A boolean indicating if the row is flagged |
row.record |
The record object containing the current row |
For array objects such as row.columnNames you can preview the array using the expressions window, and output it as a string using toString(row.columnNames) or with something like:
forEach(row.columnNames,v,v).join("; ")
Cells
The cells object is used to call information from the columns in your project. For example, cells.Foo returns a cell object representing the cell in the column named “Foo” of the current row. If the column name has spaces, use square brackets, e.g., cells["Postal Code"]. To get the corresponding column's value inside the cells variable, use .value at the end, for example, cells["Postal Code"].value. There is no cells.value - it can only be used with member fields.
Cell
A cell object contains all the data of a cell and is stored as a single object.
You can use cell on its own in the expressions editor to copy all the contents of a column to another column, including reconciliation information. Although the preview in the expressions editor will only show a small representation (“[object Cell]”), it will actually copy all the cell's data. Try this with → .
| Field | Meaning | Member fields |
|---|---|---|
cell |
An object containing the entire contents of the cell | .value, .recon, .errorMessage |
cell.value |
The value in the cell, which can be a string, a number, a boolean, null, or an error | |
cell.recon |
An object encapsulating reconciliation results for that cell | See the reconciliation section |
cell.errorMessage |
Returns the message of an EvalError instead of the error object itself (use value to return the error object) | .value |
Reconciliation
Several of the fields here provide the data used in reconciliation facets. You must type cell.recon; recon on its own will not work.
| Field | Meaning | Member fields |
|---|---|---|
cell.recon.judgment |
A string: either “matched”, "new”, "none” | |
cell.recon.judgmentAction |
A string: either "single” or “similar” (or “unknown”) | |
cell.recon.judgmentHistory |
A number, the epoch timestamp (in milliseconds) of your judgment | |
cell.recon.matched |
A boolean, true if judgment is “matched” | |
cell.recon.match |
The recon candidate that has been matched against this cell (or null) | .id, .name, .type |
cell.recon.best |
The highest scoring recon candidate from the reconciliation service (or null) | .id, .name, .type, .score |
cell.recon.features |
An array of reconciliation features to help you assess the accuracy of your matches | .typeMatch, .nameMatch, .nameLevenshtein, .nameWordDistance |
cell.recon.features.typeMatch |
A boolean, true if your chosen type is “matched” and false if not (or “(no type)” if unreconciled) | |
cell.recon.features.nameMatch |
A boolean, true if the cell and candidate strings are identical and false if not (or “(unreconciled)”) | |
cell.recon.features.nameLevenshtein |
A number representing the Levenshtein distance: larger if the difference is greater between value and candidate | |
cell.recon.features.nameWordDistance |
A number based on the word similarity | |
cell.recon.candidates |
An array of the top 3 candidates (default) | .id, .name, .type, .score |
The cell.recon.candidates and cell.recon.best objects have a few deeper fields: id, name, type, and score. type is an array of type identifiers for a list of candidates, or a single string for the best candidate.
Arrays such as cell.recon.candidates and cell.recon.candidates.type can be joined into lists and stored as strings with something like:
forEach(cell.recon.candidates,v,v.name).join("; ")
Record
A row.record object encapsulates one or more rows that are grouped together, when your project is in records mode. You must call it as row.record; record will not return values.
| Field | Meaning |
|---|---|
row.record.index |
The index of the current record (starting at 0) |
row.record.cells |
The cells of the row |
row.record.fromRowIndex |
The row index of the first row in the record |
row.record.toRowIndex |
The row index of the last row in the record + 1 (i.e. the next row) |
row.record.rowCount |
A count of the number of rows in the record |
GREL (General Refine Expression Language)
Basics
GREL is designed to resemble Javascript. Formulas use variables and depend on data types to do things like string manipulation or mathematical calculations:
| Example | Output |
|---|---|
value + " (approved)" |
Concatenate two strings; whatever is in the cell gets converted to a string first |
value + 2.239 |
Add 2.239 to the existing value (if a number); append text "2.239" to the end of the string otherwise |
value.trim().length() |
Trim leading and trailing whitespace of the cell value and then output the length of the result |
value.substring(7, 10) |
Output the substring of the value from character index 7, 8, and 9 (excluding character index 10) |
value.substring(13) |
Output the substring from index 13 to the end of the string |
Note that the operator for string concatenation is + (not “&” as is used in Excel).
Evaluating conditions uses symbols such as <, >, *, /, etc. To check whether two objects are equal, use two equal signs (value=="true").
Syntax
In OpenRefine expression language function can use either of these two forms:
- functionName(arg0, arg1, ...)
- arg0.functionName(arg1, ...)
The second form is a shorthand to make expressions easier to read. It simply pulls the first argument out and appends it to the front of the function, with a dot:
| Dot notation | Full notation |
|---|---|
value.trim().length() |
length(trim(value)) |
value.substring(7, 10) |
substring(value, 7, 10) |
So, in the dot shorthand, the functions occur from left to right in the order of calling, rather than in the reverse order with parentheses. This allows you to string together multiple functions in a readable order.
The dot notation can also be used to access the member fields of variables. For referring to column names that contain spaces (anything not a continuous string), use square brackets instead of dot notation:
| Example | Description |
|---|---|
FirstName.cells |
Access the cell in the column named “FirstName” of the current row |
cells["First Name"] |
Access the cell in the column called “First Name” of the current row |
Square brackets can also be used to get substrings and sub-arrays, and single items from arrays:
| Example | Description |
|---|---|
value[1,3] |
A substring of value, starting from character 1 up to but excluding character 3 |
"internationalization"[1,-2] |
Will return “nternationalizati” (negative indexes are counted from the end) |
row.columnNames[5] |
Will return the name of the fifth column |
Any function that outputs an array can use square brackets to select only one part of the array to output as a string (remember that the index of the items in an array starts with 0).
For example, partition() would normally output an array of three items: the part before your chosen fragment, the fragment you've identified, and the part after. Selecting only the third part with "internationalization".partition("nation")[2] will output “alization” (and so will [-1], indicating the final item in the array).
GREL controls
GREL offers controls to support branching and looping (that is, “if” and “for” functions), but unlike functions, their arguments don't all get evaluated before they get run. A control can decide which part of the code to execute and can affect the environment bindings. Functions, on the other hand, can't do either. Each control decides which of their arguments to evaluate to value, and how.
Please note that the GREL control names are case-sensitive: for example, the isError() control can't be called with iserror().
if(e, eTrue, eFalse)
Expression e is evaluated to a value. If that value is true, then expression eTrue is evaluated and the result is the value of the whole if() expression. Otherwise, expression eFalse is evaluated and that result is the value.
Examples:
| Example expression | Result |
|---|---|
if("internationalization".length() > 10, "big string", "small string") |
“big string” |
if(mod(37, 2) == 0, "even", "odd") |
“odd” |
Nested if (switch case) example:
if(value == 'Place', 'http://www.example.com/Location',
if(value == 'Person', 'http://www.example.com/Agent',
if(value == 'Book', 'http://www.example.com/Publication',
null)))
with(e1, variable v, e2)
Evaluates expression e1 and binds its value to variable v. Then evaluates expression e2 and returns that result.
| Example expression | Result |
|---|---|
with("european union".split(" "), a, a.length()) |
2 |
with("european union".split(" "), a, forEach(a, v, v.length())) |
[ 8, 5 ] |
with("european union".split(" "), a, forEach(a, v, v.length()).sum() / a.length()) |
6.5 |
filter(e1, v, e test)
Evaluates expression e1 to an array. Then for each array element, binds its value to variable v, evaluates expression test - which should return a boolean. If the boolean is true, pushes v onto the result array.
| Expression | Result |
|---|---|
filter([ 3, 4, 8, 7, 9 ], v, mod(v, 2) == 1) |
[ 3, 7, 9 ] |
forEach(e1, v, e2)
Evaluates expression e1 to an array. Then for each array element, binds its value to variable v, evaluates expression e2, and pushes the result onto the result array.
| Expression | Result |
|---|---|
forEach([ 3, 4, 8, 7, 9 ], v, mod(v, 2)) |
[ 1, 0, 0, 1, 1 ] |
forEachIndex(e1, i, v, e2)
Evaluates expression e1 to an array. Then for each array element, binds its index to variable i and its value to variable v, evaluates expression e2, and pushes the result onto the result array.
| Expression | Result |
|---|---|
forEachIndex([ "anne", "ben", "cindy" ], i, v, (i + 1) + ". " + v).join(", ") |
1. anne, 2. ben, 3. cindy |
forRange(n from, n to, n step, v, e)
Iterates over the variable v starting at from, incrementing by the value of step each time while less than to. At each iteration, evaluates expression e, and pushes the result onto the result array.
forNonBlank(e, v, eNonBlank, eBlank)
Evaluates expression e. If it is non-blank, forNonBlank() binds its value to variable v, evaluates expression eNonBlank and returns the result. Otherwise (if e evaluates to blank), forNonBlank() evaluates expression eBlank and returns that result instead.
Unlike other GREL functions beginning with “for,” forNonBlank() is not iterative. forNonBlank() essentially offers a shorter syntax to achieving the same outcome by using the isNonBlank() function within an “if” statement.
isBlank(e), isNonBlank(e), isNull(e), isNotNull(e), isNumeric(e), isError(e)
Evaluates the expression e, and returns a boolean based on the named evaluation.
Examples:
| Expression | Result |
|---|---|
isBlank("abc") |
false |
isNonBlank("abc") |
true |
isNull("abc") |
false |
isNotNull("abc") |
true |
isNumeric(2) |
true |
isError(1) |
false |
isError("abc") |
false |
isError(1 / 0) |
true |
Remember that these are controls and not functions: you can’t use dot notation (for example, the format e.isX() will not work).
Constants
| Name | Meaning |
|---|---|
| true | The boolean constant true |
| false | The boolean constant false |
| PI | From Java's Math.PI, the value of pi (that is, 3.1415...) |
Jython
Jython 2.7.2 comes bundled with the default installation of OpenRefine 3.4.1. You can add libraries and code by following this tutorial. A large number of Python files (.py or .pyc) are compatible.
Python code that depends on C bindings will not work in OpenRefine, which uses Java / Jython only. Since Jython is essentially Java, you can also import Java libraries and utilize those.
You will need to restart OpenRefine, so that new Jython or Python libraries are initialized during startup.
OpenRefine now has most of the Jsoup.org library built into GREL functions for parsing and working with HTML and XML elements.
Syntax
Expressions in Jython must have a return statement:
return value[1:-1]
return rowIndex%2
Fields have to be accessed using the bracket operator rather than dot notation:
return cells["col1"]["value"]
For example, to access the edit distance between a reconciled value and an original cell value using recon variables:
return cell["recon"]["features"]["nameLevenshtein"]
To return the lower case of value (if the value is not null):
if value is not None:
return value.lower()
else:
return None
Tutorials
Full documentation on the Jython language can be found on its official site: http://www.jython.org.
Clojure
Clojure 1.10.1 comes bundled with the default installation of OpenRefine 3.4.1. At this time, not all variables can be used with Clojure expressions: only value, row, rowIndex, cell, and cells are available.
For example, functions can take the form
(.. value (toUpperCase) )
Or can look like
(-> value (str/split #" ") last )
which functions like value.split(" ") in GREL.
For help with syntax, see the Clojure website's guide to syntax.
User-contributed Clojure recipes can be found on our wiki at https://github.com/OpenRefine/OpenRefine/wiki/Recipes#11-clojure.
Full documentation on the Clojure language can be found on its official site: https://clojure.org/.