Create a dm object from a database

Building a dm from a subset of tables

A dm can also be constructed from individual tables or views. This is useful for when you want to work with a subset of a database’s tables, perhaps from different schemas.

Below, we use the $ notation to extract two tables from the financial database. Then we create our dm by passing the tables in as arguments. Note that the tables arguments have to all be from the same source, in this case my_db.

dbListTables(my_db)

library(dbplyr)
loans <- tbl(my_db, "loans")
accounts <- tbl(my_db, "accounts")

my_manual_dm <- dm(loans, accounts)
my_manual_dm

Defining Primary and Foreign Keys

Primary keys and foreign keys are how relational database tables are linked with each other. A primary key is a column or column tuple that has a unique value for each row within a table. A foreign key is a column or column tuple containing the primary key for a row in another table. Foreign keys act as cross references between tables. They specify the relationships that gives us the relational database. For more information on keys and a crash course on databases, see vignette("howto-dm-theory").

In many cases, dm_from_con() already returns a dm with all keys set. If not, dm allows us to define primary and foreign keys ourselves. For this, we use learn_keys = FALSE to obtain a dm object with only the tables.

library(dm)

fin_dm <- dm_from_con(my_db, learn_keys = FALSE)
fin_dm

The model diagram (https://relational.fel.cvut.cz/assets/img/datasets-generated/financial.svg) provided by our test database loosely illustrates the intended relationships between tables. In the diagram, we can see that the loans table should be linked to the accounts table. Below, we create those links in 3 steps:

1. Add a primary key id to the accounts table
2. Add a primary key id to the loans table
3. Add a foreign key account_id to the loans table referencing the accounts table

Then we assign colors to the tables and draw the structure of the dm.

Note that when the foreign key is created, the primary key in the referenced table does not need to be specified, but the primary key must already be defined. And, as mentioned above, primary and foreign key constraints on the database are currently only imported for Postgres, SQL Server databases and MariaDB, and only when dm_from_con() is used. This process of key definition needs to be done manually for other databases.

my_dm_keys <-
  my_manual_dm %>%
  dm_add_pk(accounts, id) %>%
  dm_add_pk(loans, id) %>%
  dm_add_fk(loans, account_id, accounts) %>%
  dm_set_colors(green = loans, orange = accounts)

my_dm_keys %>%
  dm_draw()

Once you have instantiated a dm object, you can continue to add tables to it. For tables from the original source for the dm, use dm()

trans <- tbl(my_db, "trans")

my_dm_keys %>%
  dm(trans)

Serializing a dm object

A dm object is always linked to a database connection. This connection is lost when the dm object is saved to disk, e.g., when saving the workspace in R or in Posit Workbench, or when using knitr chunks:

unserialize(serialize(my_dm_keys, NULL))

The connection is tightly coupled with the tables in the dm object and cannot be replaced. A practical solution is to define, for each dm object your project uses, a function that recreates it using a new database connection:

my_db_fun <- function() {
  dbConnect(
    MariaDB(),
    username = "guest",
    password = "ctu-relational",
    dbname = "Financial_ijs",
    host = "relational.fel.cvut.cz"
  )
}

my_dm_fun <- function(my_db = my_db_fun()) {
  loans <- tbl(my_db, "loans")
  accounts <- tbl(my_db, "accounts")
  dm(loans, accounts) %>%
    dm_add_pk(accounts, id) %>%
    dm_add_pk(loans, id) %>%
    dm_add_fk(loans, account_id, accounts) %>%
    dm_set_colors(green = loans, orange = accounts)
}

To avoid reconnecting and/or recreating every time you need a dm object, you can use memoise::memoise() to memoize the connection and/or dm functions.

Transient nature of operations

Like other R objects, a dm is immutable and all operations performed on it are transient unless stored in a new variable.

my_dm_keys

my_dm_trans <-
  my_dm_keys %>%
  dm(trans)

my_dm_trans

And, like {dbplyr}, results are never written to a database unless explicitly requested.

my_dm_keys %>%
  dm_flatten_to_tbl(loans)

my_dm_keys %>%
  dm_flatten_to_tbl(loans) %>%
  sql_render()

Performing operations on tables by “zooming”

As the dm is a collection of tables, if we wish to perform operations on an individual table, we set it as the context for those operations using dm_zoom_to(). See vignette("tech-dm-zoom") for more detail on zooming.

dm operations are transient unless persistence is explicitly requested. To make our chain of manipulations on the selected table permanent, we assign the result of dm_insert_zoomed() to a new object, my_dm_total. This is a new dm object, derived from my_dm_keys, with a new lazy table total_loans linked to the accounts table.

my_dm_total <-
  my_dm_keys %>%
  dm_zoom_to(loans) %>%
  group_by(account_id) %>%
  summarize(total_amount = sum(amount, na.rm = TRUE)) %>%
  ungroup() %>%
  dm_insert_zoomed("total_loans")

Context is set to the table “loans” using dm_zoom_to(loans). You can learn more about zooming in the tutorial vignette("tech-dm-zoom"). We then use {dplyr} functions on the zoomed table to generate a new summary table.

summarize() returns a temporary table with one row for each group created by the preceding group_by() function. The columns in the temporary table are constrained to the columns passed as arguments to the group_by() function and the column(s) created by the summarize() function.

dm_insert_zoomed("total_loans") adds the temporary table created by summarize() to the data model under a new name, total_loans. Because the grouping variable account_id is a primary key, the new derived table is automatically linked to the accounts table.

my_dm_total %>%
  dm_set_colors(violet = total_loans) %>%
  dm_draw()

The resulting table total_loans can be accessed like any other table in the dm object.

my_dm_total$total_loans

It is a lazy table powered by the {dbplyr} package: the results are not materialized; instead, an SQL query is built and executed each time the data is requested.

my_dm_total$total_loans %>%
  sql_render()

Use compute() on a zoomed table to materialize it to a temporary table and avoid recomputing. See vignette("howto-dm-copy") for more details.

Downloading data

When it becomes necessary to move data locally for analysis or reporting, the {dm} method collect() is used. Operations on dm objects for databases are limited to report only the first ten results. collect() forces the evaluation of all SQL queries and the generation of the complete set of results. The resulting tables are transferred from the RDBMS and stored as local tibbles.

my_dm_local <-
  my_dm_total %>%
  collect()

my_dm_local$total_loans

Use this method with caution. If you are not sure of the size of the dataset you will be downloading, you can call dm_nrow() on your dm for the row count of your data model’s tables.

my_dm_total %>%
  dm_nrow()

Persisting results

It is just as simple to move a local relational model into an RDBMS as is using collect() to download it. The method used is copy_dm_to() and it takes as arguments a database connection and a dm object. In the example below, a local SQLite database is used to demonstrate it, but {dm} is designed to work with any RDBMS supported by {DBI}.

destination_db <- DBI::dbConnect(RSQLite::SQLite())

deployed_dm <- copy_dm_to(destination_db, my_dm_local)

deployed_dm
my_dm_local

In the output, you can observe that the src for deployed_dm is the SQLite database, while for my_dm_local the source is the local R environment.

Persisting tables are covered in more detail in vignette("howto-dm-copy").

When done, do not forget to disconnect:

DBI::dbDisconnect(destination_db)
DBI::dbDisconnect(my_db)

Conclusion

In this tutorial, we have demonstrated how simple it is to load a database into a dm object and begin working with it. Currently, loading a dm from most RDBMS requires you to manually set key relations, but {dm} provides methods to make this straightforward. It is planned that future versions of dm will support automatic key creation for more RDBMS.

The next step is to read vignette("howto-dm-copy"), where copying your tables to and from an RDBMS is covered. vignette("howto-dm-rows") discusses manipulation of individual rows in a database.

Further reading

vignette("howto-dm-df") – Is your data in local data frames? This article covers creating a data model from your local data frames, including building the relationships in your data model, verifying your model, and leveraging the power of dplyr to operate on your data model.

vignette("howto-dm-theory") – Do you know all about data frames but very little about relational data models? This quick introduction will walk you through the key similarities and differences, and show you how to move from individual data frames to a relational data model.