Package uk.ac.starlink.table.join
Class RowMatcher
java.lang.Object
uk.ac.starlink.table.join.RowMatcher
Performs matching on the rows of one or more tables.
The specifics of what constitutes a matched row, and some additional
intelligence about how to determine this, are supplied by an
associated
MatchEngine
object, but the generic parts of
the matching algorithms are done here.
Note that since the LinkSets and other objects handled by this class may be very large when large tables are being matched, the algorithms in this class are coded carefully to use as little memory as possible. Techniques include removing items from one collection as they are added to another. This means that in many cases input values may be modified by the methods.
Some of the computationally intensive work done by this abstract class is defined as abstract methods to be implemented by concrete subclasses.
- Since:
- 13 Jan 2004
- Author:
- Mark Taylor (Starlink)
-
Field Summary
FieldsModifier and TypeFieldDescriptionstatic final int
Actual value for default parallelism (also limited by machine).static final int
Maximum suggested value for parallelism. -
Method Summary
Modifier and TypeMethodDescriptionConstructs a new empty LinkSet for use by this matcher.static RowMatcher
createMatcher
(MatchEngine engine, uk.ac.starlink.table.StarTable[] tables, uk.ac.starlink.table.RowRunner runner) Creates a RowMatcher instance.findGroupMatches
(MultiJoinType[] joinTypes) Returns a list of RowLink objects corresponding to a match performed with this matcher's tables using its match engine.findInternalMatches
(boolean includeSingles) Returns a list of RowLink objects corresponding to all the internal matches in this matcher's sole table using its match engine.findMultiPairMatches
(int index0, boolean bestOnly, MultiJoinType[] joinTypes) Returns a set of RowLink objects each of which represents matches between one of the rows of a reference table and any of the other tables which can provide matches.findPairMatches
(PairMode pairMode) Returns a set of RowLink objects corresponding to a pairwise match between this matcher's two tables performed with its match engine.Returns the current progress indicator for this matcher.void
setIndicator
(ProgressIndicator indicator) Sets the progress indicator for this matcher.
-
Field Details
-
DFLT_PARALLELISM_LIMIT
public static final int DFLT_PARALLELISM_LIMITMaximum suggested value for parallelism. This value is limited mainly because not all the steps of matching have been (can be?) parallelised, and so in accordance with Amdahl's Law there are diminishing returns as the number of processors increases. Some steps in fact slow down as more threads are added, because of additional data structure combination work required. The current value of 6 is somewhat, though not completely, arbitrary, having been set following some experimentation.- See Also:
-
DFLT_PARALLELISM
public static final int DFLT_PARALLELISMActual value for default parallelism (also limited by machine).
-
-
Method Details
-
setIndicator
Sets the progress indicator for this matcher.- Parameters:
indicator
- new indicator
-
getIndicator
Returns the current progress indicator for this matcher.- Returns:
- indicator
-
createLinkSet
Constructs a new empty LinkSet for use by this matcher. The current implementation returns one based on a HashSet, but future implementations may provide the option of LinkSet implementations backed by disk.- Returns:
- new LinkSet
-
findPairMatches
Returns a set of RowLink objects corresponding to a pairwise match between this matcher's two tables performed with its match engine. Each element in the returned list corresponds to a matched pair with one entry from each of the input tables.- Parameters:
pairMode
- matching mode to determine which rows appear in the result- Returns:
- links representing matched rows
- Throws:
IOException
InterruptedException
-
findMultiPairMatches
public LinkSet findMultiPairMatches(int index0, boolean bestOnly, MultiJoinType[] joinTypes) throws IOException, InterruptedException Returns a set of RowLink objects each of which represents matches between one of the rows of a reference table and any of the other tables which can provide matches. Elements of the result set will be instances ofPairsRowLink
.- Parameters:
index0
- index of the reference table in the list of tables owned by this row matcherbestOnly
- true if only the best match between the reference table and any other table should be retainedjoinTypes
- inclusion criteria for output table rows- Returns:
- set of PairsRowLink objects representing multi-pair matches
- Throws:
IOException
InterruptedException
-
findGroupMatches
Returns a list of RowLink objects corresponding to a match performed with this matcher's tables using its match engine. Each element in the returned list corresponds to a matched group of input rows, with no more than one entry from each table. Each input table row appears in no more than one RowLink in the returned list. Any number of tables can be matched.- Parameters:
joinTypes
- inclusion criteria for output table rows- Returns:
- list of
RowLink
s corresponding to the selected rows - Throws:
IOException
InterruptedException
-
findInternalMatches
Returns a list of RowLink objects corresponding to all the internal matches in this matcher's sole table using its match engine.- Parameters:
includeSingles
- whether to include unmatched (singleton) row links in the returned link set- Returns:
- a set of
RowLink
objects giving all the groups of matched objects in this matcher's sole table - Throws:
IOException
InterruptedException
-
createMatcher
public static RowMatcher createMatcher(MatchEngine engine, uk.ac.starlink.table.StarTable[] tables, uk.ac.starlink.table.RowRunner runner) Creates a RowMatcher instance.- Parameters:
engine
- matching enginetables
- the array of tables on which matches are to be donerunner
- RowRunner to control multithreading, or null to fall back to sequential implementation- Returns:
- new RowMatcher
-