Connect to a sqlite database.

Use src_sqlite to connect to an existing sqlite database, and tbl to connect to tables within that database. If you are running a local sqliteql database, leave all parameters set as their defaults to connect. If you're connecting to a remote database, ask your database administrator for the values of these variables. src_memdb is an easy way to use an in-memory SQLite database that is scoped to the current session.

src_sqlite(path, create = FALSE)

# S3 method for src_sqlite tbl(src, from, ...)


Path to SQLite database


if FALSE, path must already exist. If TRUE, will create a new SQlite3 database at path if path does not exist and connect to the existing database if path does exist.


a sqlite src created with src_sqlite.


Either a string giving the name of table in database, or sql described a derived table or compound join.


Included for compatibility with the generic, but otherwise ignored.


To see exactly what SQL is being sent to the database, you see show_query and explain.


Typically you will create a grouped data table is to call the group_by method on a mysql tbl: this will take care of capturing the unevalated expressions for you.

For best performance, the database should have an index on the variables that you are grouping by. Use explain to check that the database is using the indexes that you expect.


All data manipulation on SQL tbls are lazy: they will not actually run the query or retrieve the data unless you ask for it: they all return a new tbl_sql object. Use compute to run the query and save the results in a temporary in the database, or use collect to retrieve the results to R.

Note that do is not lazy since it must pull the data into R. It returns a tbl_df or grouped_df, with one column for each grouping variable, and one list column that contains the results of the operation. do never simplifies its output.

Query principles

This section attempts to lay out the principles governing the generation of SQL queries from the manipulation verbs. The basic principle is that a sequence of operations should return the same value (modulo class) regardless of where the data is stored.

  • arrange(arrange(df, x), y) should be equivalent to arrange(df, y, x)

  • select(select(df, a:x), n:o) should be equivalent to select(df, n:o)

  • mutate(mutate(df, x2 = x * 2), y2 = y * 2) should be equivalent to mutate(df, x2 = x * 2, y2 = y * 2)

  • filter(filter(df, x == 1), y == 2) should be equivalent to filter(df, x == 1, y == 2)

  • summarise should return the summarised output with one level of grouping peeled off.

  • src_sqlite
  • tbl.src_sqlite
library(dplyr) # Connection basics --------------------------------------------------------- # To connect to a database first create a src: my_db <- src_sqlite(path = tempfile(), create = TRUE) # Then reference a tbl within that src my_tbl <- tbl(my_db, "my_table") # Here we'll use the Lahman database: to create your own local copy, # run lahman_sqlite() if (requireNamespace("RSQLite") && has_lahman("sqlite")) { lahman_s <- lahman_sqlite() # Methods ------------------------------------------------------------------- batting <- tbl(lahman_s, "Batting") dim(batting) colnames(batting) head(batting) # Data manipulation verbs --------------------------------------------------- filter(batting, yearID > 2005, G > 130) select(batting, playerID:lgID) arrange(batting, playerID, desc(yearID)) summarise(batting, G = mean(G), n = n()) mutate(batting, rbi2 = 1.0 * R / AB) # note that all operations are lazy: they don't do anything until you # request the data, either by `print()`ing it (which shows the first ten # rows), by looking at the `head()`, or `collect()` the results locally. system.time(recent <- filter(batting, yearID > 2010)) system.time(collect(recent)) # Group by operations ------------------------------------------------------- # To perform operations by group, create a grouped object with group_by players <- group_by(batting, playerID) group_size(players) # sqlite doesn't support windowed functions, which means that only # grouped summaries are really useful: summarise(players, mean_g = mean(G), best_ab = max(AB)) # When you group by multiple level, each summarise peels off one level per_year <- group_by(batting, playerID, yearID) stints <- summarise(per_year, stints = max(stint)) filter(ungroup(stints), stints > 3) summarise(stints, max(stints)) # Joins --------------------------------------------------------------------- player_info <- select(tbl(lahman_s, "Master"), playerID, birthYear) hof <- select(filter(tbl(lahman_s, "HallOfFame"), inducted == "Y"), playerID, votedBy, category) # Match players and their hall of fame data inner_join(player_info, hof) # Keep all players, match hof data where available left_join(player_info, hof) # Find only players in hof semi_join(player_info, hof) # Find players not in hof anti_join(player_info, hof) # Arbitrary SQL ------------------------------------------------------------- # You can also provide sql as is, using the sql function: batting2008 <- tbl(lahman_s, sql("SELECT * FROM Batting WHERE YearID = 2008")) batting2008 }
Documentation reproduced from package dplyr, version 0.5.0, License: MIT + file LICENSE

Community examples

Looks like there are no examples yet.