NAME

PDL::IO::DBI - Create PDL from database (optimized for speed and large data)

SYNOPSIS

  use PDL;
  use PDL::IO::DBI ':all';

  # simple usage - using DSN + SQL query
  my $sql = "select ymd, open, high, low, close from quote where symbol = 'AAPL' AND ymd >= 20140404 order by ymd";
  my $pdl = rdbi2D("dbi:SQLite:dbname=Quotes.db", $sql);

  use DBI;

  # using DBI handle + SQL query with binded values
  my $dbh = DBI->connect("dbi:Pg:dbname=QDB;host=localhost", 'username', 'password');
  my $sql = "select ymd, open, high, low, close from quote where symbol = ? AND ymd >= ? order by ymd";
  # rdbi2D
  my $pdl = rdbi2D($dbh, $sql, ['AAPL', 20140104]);                     # 2D piddle
  # rdbi1D
  my ($y, $o, $h, $l, $c) = rdbi1D($dbh, $sql, ['AAPL', 20140104]);     # 5x 1D piddle (for each column)

  # using DBI handle + SQL query with binded values + extra options
  my $dbh = DBI->connect("dbi:Pg:dbname=QDB;host=localhost", 'username', 'password');
  my $sql = "select ymd, open, high, low, close from quote where symbol = ? AND ymd >= ? order by ymd";
  my $pdl = rdbi2D($dbh, $sql, ['AAPL', 20140104], { type=>float, fetch_chunk=>100000, reshape_inc=>100000 });

DESCRIPTION

For creating a piddle from database data one can use the following simple approach:

  use PDL;
  use DBI;
  my $dbh = DBI->connect($dsn);
  my $pdl = pdl($dbh->selectall_arrayref($sql_query));

However this approach does not scale well for large data (e.g. SQL queries resulting in millions of rows).

This module is optimized for creating piddles populated with very large database data. It currently supports only reading data from database not updating/inserting to DB.

The goal of this module is to be as fast as possible. It is designed to silently converts anything into a number (wrong or undefined values are converted into 0).

FUNCTIONS

By default, PDL::IO::DBI doesn't import any function. You can import individual functions like this:

 use PDL::IO::DBI 'rdbi2D';

Or import all available functions:

 use PDL::IO::DBI ':all';

rdbi1D

Queries the database and stores the data into 1D piddles.

 $sql_query = "SELECT high, low, avg FROM data where year > 2010";
 my ($high, $low, $avg) = rdbi1D($dbh_or_dsn, $sql_query);
 #or
 my ($high, $low, $avg) = rdbi1D($dbh_or_dsn, $sql_query, \@sql_query_params);
 #or
 my ($high, $low, $avg) = rdbi1D($dbh_or_dsn, $sql_query, \@sql_query_params, \%options);
 #or
 my ($high, $low, $avg) = rdbi1D($dbh_or_dsn, $sql_query, \%options);

Example:

  my ($id, $high, $low) = rdbi1D($dbh, 'SELECT id, high, low FROM sales ORDER by id');

  # column types:
  #   id   .. INTEGER
  #   high .. NUMERIC
  #   low  .. NUMERIC

  print $id->info, "\n";
  PDL: Long D [100000]          # == 1D piddle, 100 000 rows from DB

  print $high->info, "\n";
  PDL: Double D [100000]        # == 1D piddle, 100 000 rows from DB

  print $low->info, "\n";
  PDL: Double D [100000]        # == 1D piddle, 100 000 rows from DB

  # column names (lowercase) are stored in loaded piddles in $pdl->hdr->{col_name}
  print $id->hdr->{col_name},   "\n";  # prints: id
  print $high->hdr->{col_name}, "\n";  # prints: high
  print $low->hdr->{col_name},  "\n";  # prints: low

Parameters:

dbh_or_dsn

DBI handle of database connection or data source name.

sql_query

SQL query.

sql_query_params

Optional bind values that can be used for queries with placeholders.

Items supported in options hash:

type

Defines the type of output piddles: double, float, longlong, long, short, byte. Default value is auto which means that the type of the output piddles is auto detected. BEWARE: type `longlong` can be used only on perls with 64bitint support.

You can set one type for all columns/piddles:

  my ($high, $low, $avg) = rdbi1D($dbh_or_dsn, $sql_query, {type => double});

or separately for each column/piddle:

  my ($high, $low, $avg) = rdbi1D($dbh_or_dsn, $sql_query, {type => [long, double, double]});
fetch_chunk

We do not try to load all query results into memory at once, we load them in chunks defined by this parameter. Default value is 8000 (rows). If reuse_sth is true, rdbi1D will return one chunk per call, and the number of rows in a chunk will never exceed fetch_chunk.

reshape_inc

As we do not try to load all query results into memory at once; we also do not know at the beginning how many rows there will be. Therefore we do not know how big piddle to allocate, we have to incrementally (re)allocate the piddle by increments defined by this parameter. Default value is 80000 (unless reuse_sth is used).

If you know how many rows there will be you can improve performance by setting this parameter to expected row count.

If you are using reuse_sth, reshape_inc is by default equal to fetch_chunk to avoid reallocations, but you could set it to a different value if you wanted to.

null2bad

Values 0 (default) or 1 - convert NULLs to BAD values (there is a performance cost when turned on).

reuse_sth

Whether to reuse the statement handle used to fetch the rows.

When reuse_sth is false, all rows matching the select statement are fetched at once, and the statement handle is never reused. Every new call to rdbi1D will rerun the select statement and fetch the same rows again.

When reuse_sth is not false, it must be a reference (either to undef, or to a statement handle). In this case, the operation mode changes: rdbi1D will try to fetch fetch_chunk rows from the database, and will return early. It will reuse the statement handle passed in via reuse_sth. If a reference to undef is passed, rdbi1D will initialize the statement handle itself. The idea is that you call rdbi1D repeatedly to obtain subsets of the total number of rows in the database matching the select statement. This can be useful if the logic to handle subsets is already present in your code, and you don't need all rows in memory at once.

As an example, suppose you are calculating a minimum value. (You would probably do this in the database directly, but it makes for a simple example.) You don't need to have all matching rows in memory at once. Fetching chunk by chunk will do just fine:

  my $N = 500_000;
  my $minimum;
  my $sth;
  for (;;) {
    my ($values) = rdbi1D($dbh, "SELECT value FROM table", {reuse_sth => \$sth, fetch_chunk => $N});
    last unless $sth;
    if (!defined($minimum) || $values->minimum->sclr < $minimum) { $minimum = $values->minimum->sclr }
  }

You can avoid the allocation of a single large PDL in this way. This wouldn't help you much if the database was small. But if it was so large the resulting PDL didn't fit in memory, working in chunks allows you to process all of the data. Note that reshape_inc will be set to the same value as fetch_chunk to avoid a reallocation to the chunk size, unless you explicitly set reshape_inc to another value.

Note that rdbi1D sets the reused statement handle to undef if there are no more chunks, i.e., when the database query returns no rows. You can use this to your advantage to terminate the loop fetching the chunks, without having to count the rows yourself.

debug

Values 0 (default) or 1 - turn on/off debug messages

rdbi2D

Queries the database and stores the data into a 2D piddle.

  my $pdl = rdbi2D($dbh_or_dsn, $sql_query);
  #or
  my $pdl = rdbi2D($dbh_or_dsn, $sql_query, \@sql_query_params);
  #or
  my $pdl = rdbi2D($dbh_or_dsn, $sql_query, \@sql_query_params, \%options);
  #or
  my $pdl = rdbi2D($dbh_or_dsn, $sql_query, \%options);

Example:

  my $pdl = rdbi2D($dbh, 'SELECT id, high, low FROM sales ORDER by id');

  # column types:
  #   id   .. INTEGER
  #   high .. NUMERIC
  #   low  .. NUMERIC

  print $pdl->info, "\n";
  PDL: Double D [100000, 3]     # == 2D piddle, 100 000 rows from DB

Parameters and items supported in options hash are the same as by "rdbi1D". reuse_sth is not supported yet for "rdbi2D".

Handling DATE, DATETIME, TIMESTAMP database types

By default DATETIME values are converted to double value representing epoch seconds e.g.

 # 1970-01-01T00:00:01.001     >>          1.001
 # 2000-12-31T12:12:12.5       >>  978264732.5
 # BEWARE: timestamp is truncated to milliseconds
 # 2000-12-31T12:12:12.999001  >>  978264732.999
 # 2000-12-31T12:12:12.999999  >>  978264732.999

If you specify an output type longlong for DATETIME column then the DATETIME values are converted to longlong representing epoch microseconds e.g.

 # 1970-01-01T00:00:01.001        >>          1001000
 # 2000-12-31T12:12:12.5          >>  978264732500000
 # 2000-12-31T12:12:12.999999     >>  978264732999999
 # BEWARE: timestamp is truncated to microseconds
 # 2000-12-31T12:12:12.999999001  >>  978264732999999
 # 2000-12-31T12:12:12.999999999  >>  978264732999999

If you have PDL::DateTime installed then rcsv1D automaticcally converts DATETIME columns to PDL::DateTime piddles:

 # autodetection - same as: type=>'auto'
 my ($datetime_piddle, $pr) = rdbi1D("select mydate, myprice from sales");

 # or you can explicitely use type 'datetime'
 my ($datetime_piddle, $pr) = rdbi1D("select mydate, myprice from sales", {type=>['datetime', double]});

SEE ALSO

PDL, DBI

LICENSE

This program is free software; you can redistribute it and/or modify it under the same terms as Perl itself.

COPYRIGHT

2014+ KMX <kmx@cpan.org>