phplyr

Data & Workflow

Statistics, Modeling & Visualization

Guide

Compatibility and limitations

PHP versions

phplyr supports PHP 8.3, 8.4, and 8.5. CI tests all three on Linux.

The pipe operator |> is PHP 8.5 syntax. On PHP 8.3 and 8.4, the same pipelines are method chains, or calls to pipe(). Every function works on all three versions.

Requirements

phplyr requires the mbstring extension. Composer installs these with phplyr:

  • phibble (phpstats/phibble 0.3, 0.4, or 0.5), which provides the tables. phibble requires PHP’s PDO extension.
  • symfony/polyfill-php84, which provides PHP 8.4 functions on PHP 8.3.

Optional extensions

Extension Enables
phit lm(), glm(), lmer(), and glmer(), and native mean(), sd(), var_(), median(), and quantile()
phpolars phibble’s PolarsTable backend
intl Transliteration in make_clean_names() and clean_names() that matches R’s janitor
A PDO driver (SQLite, PostgreSQL, or MySQL) Database tables through Phibble\tbl()

Without phit, the descriptive helpers run in PHP, and the model functions throw a RuntimeException that names the missing function. phit-polyfill defines the model functions in PHP. See Models. Without intl, make_clean_names() transliterates with iconv, and some names can differ from janitor’s.

Backends

A phplyr function calls the methods of phibble’s Table contract, so a pipeline runs on any table that implements it:

  • Phibble\DataFrame holds the data in PHP memory.
  • Phibble\tbl($pdo, 'table') builds SQL for SQLite, PostgreSQL, or MySQL. A function given a PHP closure, and an expression without a SQL translation, runs the query so far, loads the rows, and continues in memory. Databases and expressions lists those expressions.
  • Phibble\Polars\PolarsTable belongs to phibble and runs pipelines in Polars through the phpolars extension. A closure step converts the data to a DataFrame and back.

The test suite runs about 120 pipelines on both DataFrame and PolarsTable and requires the same columns, types, groups, and rows from each.

Not supported

  • Bare column names and tidy evaluation. Columns are strings, and expressions are closures, helpers, or col() expressions.
  • mutate(_keep: 'used') and _keep: 'unused' with closures. These work with expression arguments such as col('x')->mul(2), whose columns phplyr can read. A PHP closure does not expose the columns it reads, so phplyr throws an InvalidArgumentException.
  • cur_data() and cur_data_all(), which dplyr has deprecated. Use pick().
  • Data frame columns and across(.unpack). pack() makes a column of named arrays.
  • The scoped _at, _if, and _all variants, such as mutate_at(). Use across().
  • janitor’s adorn_*() functions and three-way tabyl().

Differences in behavior

  • _by sorts groups. dplyr keeps the order of first appearance for .by. phplyr sorts by key, as it does for group_by().
  • Strings are UTF-8. String functions count UTF-8 characters.
  • Date-times parse to UTC unless a function is given tz:, whatever PHP’s default time zone is, as in lubridate.
  • Models return arrays. lm() and the other model functions return phit’s result array, with R’s coefficient names in columns, in place of a model object.

From phpstats/phplyr@d319dc5.