path: root/config-model/src/test/derived/music/defs/vespa.config.search.rank-profiles.def

# Copyright Yahoo. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
namespace=vespa.config.search
## name of this rank profile. maps to table index for internal use.
rankprofile[].name string

## the name of a generic property available to the feature execution framework and feature plugins
rankprofile[].fef.property[].name string

## the value of a generic property available to feature plugins
rankprofile[].fef.property[].value string

## the catalog name overrides apply to
rankprofile[].catalog[].name string

## Boost value for AND queries in this catalog.
rankprofile[].catalog[].andboost int default=0

## Boost value for OR queries in this catalog.
rankprofile[].catalog[].orboost int default=0

## Boost value for ANY queries in this catalog.
rankprofile[].catalog[].anyboost int default=0

## Boost value for NEAR queries in catalog.
rankprofile[].catalog[].nearboost int default=0

## Boost value for ORDEREDNEAR queries in this catalog.
rankprofile[].catalog[].orderednearboost int default=0

## Boost value for phrase queries in this catalog.
rankprofile[].catalog[].phraseboost int default=0

## Boost value for all queries in catalog.
rankprofile[].catalog[].rankboost int default=0

## If true, the context boost is the max value of
## the individual contextboosts.
## When false, the context boost when a term is in
## several contexts is the sum of the individual contextboosts.
rankprofile[].catalog[].bestcontextboostonly bool default=false


## If true, then use extnumoccboost only when calculating rank values.
## Also, do not normalize the extnumoccboost value with
## global term frequency. Default value is false.
rankprofile[].catalog[].extnumoccboostonly bool default=false

## If yes, then use extnumoccboost only when calculating rank values.
## Also, do not normalize the extnumoccboost value with
## global term frequency. Default value is no.
rankprofile[].catalog[].numoccandextnumoccboostonly bool default=false

## If yes, then use bitvectors when possible.
## Default value is false.
rankprofile[].catalog[].preferbitvector bool default=false

## Load extnumoccboost for this catalog from the named file.
## extnumoccboost specifies boost values due to the number of
## occurences of a term that are external to the document. If
## "NULL" is given as file name, then all extnumoccboost values
## will be set to 0.
rankprofile[].catalog[].extnumoccboost.table string default="/home/vespa/conf/vespa/search/ranktables/constant-0000"

## Load numoccboost for this catalog from the named file.
## numoccboost specifies boost values due to the number of occurences in
## a document. If "NULL" is given as file name, then all numoccboost
## values will be set to 0.
rankprofile[].catalog[].numoccboost.table string default="/home/vespa/conf/vespa/search/ranktables/constant-0000"

## Load firstoccboost for catalog from the file named.
## firstoccboost specifies boost values due to the position of the
## first occurence in a document. If "NULL" is given as file name,
## then all firstoccboost values will be set to 0.
rankprofile[].catalog[].firstoccboost.table string default="/home/vespa/conf/vespa/search/ranktables/constant-0000"


## Load firstoccproximityboost for this catalog from the file named.
## firstoccproximity boost specifies boost values due to the correlation between
## positions of the first occurence in a document for two and two words.
##
## If "NULL" is given as file name, then all
## firstoccproximityboost values will be set to 0. If otherwise set,
## should be the name of a file to load into the table. The file
## should have 256 lines each containing a single integer.
##
## There are 256 elements in the table, handling forward distances from 1.
## The corresponding firstoccrevproximityboost table is used
## to handle closeness in reverse order.
##
## The last array index specifies the proximity table set. During
## evaluation, the bigram proximity weight supplied by the query segmenter
## specifies which proximity table set to use, with a fallback to set 0
## when no information is available.
rankprofile[].catalog[].firstoccproximityboost[].table string default="/home/vespa/conf/vespa/search/ranktables/constant-0000"

## Load firstoccrevproximityboost table for this catalog from the named file.
## Specifies boost values due to the correlation between positions
## of the first occurence in a document for two and two words when
## the second word in the query comes first in the document.
## See also firstoccproximityboost above.
rankprofile[].catalog[].firstoccrevproximityboost[].table string default="/home/vespa/conf/vespa/search/ranktables/constant-0000"

## Load proximityboost for this catalog from the named file.
## proximity boost specifies boost values due to the correlation between
## positions of the occurences in a document for two and two words.
## See also firstoccproximityboost above.
rankprofile[].catalog[].proximityboost[].table string default="/home/vespa/conf/vespa/search/ranktables/constant-0000"

## Load revproximityboost for this catalog from the named file.
## revproximity boost specifies boost values due to the correlation between
## positions of the occurences in a document for two and two words.
## See also firstoccproximityboost above.
rankprofile[].catalog[].revproximityboost[].table string default="/home/vespa/conf/vespa/search/ranktables/constant-0000"

## Load divtable for this catalog from the named file.
## Rank values for a query term are divided by the entry
## in divtable indexed by log2 of term frequence.
## The file should contain ?? lines each with a single integer.
rankprofile[].catalog[].divtable string default=""

## The name of a context in this catalog to specify boosts for.
rankprofile[].catalog[].context[].name string

## Boost occurrences in this context with the given value.
## XXX -1 uses default (???) from somewhere(TM).
rankprofile[].catalog[].context[].contextboost int default=0

## Boost pair of occurrences in this context with
## the given value when evaluating 2 words from same catalog in
## parallell.
## XXX -1 uses default (???) from somewhere(TM).
rankprofile[].catalog[].context[].commoncontextboost.pair int default=0

## Boost triple of occurrences in this context with
## the given value when evaluating 3 words from same catalog in
## parallell.
## XXX -1 uses default (???) from somewhere(TM).
rankprofile[].catalog[].context[].commoncontextboost.triple int default=0

## Boost quad of occurrences in this context with
## the given value when evaluating 4 words from same catalog in
## parallell.
## XXX -1 uses default (???) from somewhere(TM).
rankprofile[].catalog[].context[].commoncontextboost.quad int default=0


## The name of the attribute
rankprofile[].attribute[].name string

## Boost value for queries that hit in this attribute
rankprofile[].attribute[].attributecontextboost int default=0

## Load weightboost for this attribute from the named file.
## weightboost specifies boost values due to the weight (weighted set)
## or number of occurences (single, array) in an attribute.
## If "NULL" is given as file name, then all weightboost values will be set to 0.
rankprofile[].attribute[].weightboost.table string default="/home/vespa/conf/vespa/search/ranktables/constant-0000"


## Load static rank values from the given staticrank docattr vector.
## Must be specified in index.cf as a staticrankfile.
rankprofile[].staticrankfile string default=""

## Multiply static rank values with given value when calculating total
## rank value.
rankprofile[].staticcoefficient int default=1

## If false then use only static ranking when sorting result hits.
## Default is true.
rankprofile[].dynamicranking bool default=true

## If dynamic ranking is turned off, then ascending will sort the
## result hits with lowest static rank values first, while
## descending will sort with highest static rank values
## first. Default is descending. This keyword has no effect if
## dynamic ranking is on.
rankprofile[].staticranksortorder string default="descending"

## Load static rank mapping from the file named table. The static
## rank mapping maps each 8-bit static rank value into a 32-bit static
## rank value. This option may only be used with 8-bit static rank files.
rankprofile[].staticrankmap string default=""

## If set to "true", total rank will be reduced when dynamic rank is less than
## 25% of static rank, to suppress irrelevant hits from popular sites.
## If set to "false", total rank is not reduced.
rankprofile[].clampstaticrank bool default=false

## Load document datetime values used for freshness boost calculation from
## this file. The values must be coded as minutes since
## 1900-01-01T00:00Z. The value 0 has the special meaning
## "no datetime value exists".
rankprofile[].freshnessboost.file string default=""

## Load freshnessboost lookup-table values from the file named
## table instead of using built-in default values. The file must
## contain 32 white-space separated non-negative integers.
rankprofile[].freshnessboost.table string default="/home/vespa/conf/vespa/search/ranktables/constant-0000"

## When calculating the freshness boost value multiply difference between
## current datetime and document datetime with timeoffset before taking
## the base-2 logarithm. Default value is 1. Max value is 31.
rankprofile[].freshnessboost.timeoffset int default=1

## If a document has datetime value 0, then use defaultboostvalue
## as freshness boost value instead of doing table lookup. The default
## default value is 0 (no boost).
rankprofile[].freshnessboost.defaultboostvalue int default=0

## Multiply freshness boost value with coefficient when calculating
## total freshness boost value. If coefficient 0 is used, no freshness
## boost value will be computed or added. Default value is 0.
rankprofile[].freshnessboost.coefficient int default=0

## boost table files for distance ranking, 1 dimension.
## The tables have 465 elements each, where slots 0..15 represents
## distances 0..15 while the remaining slots represents distance
## (16 + (slot & 15)) << ((slot >> 4) - 1).  Linear interpolation is
## used for distances "between" table slots.
##
## If "NULL" is given as the file name then all 1D distance boost values
## for that table will be set to 0.
rankprofile[].distance1dboosttable[].table string

## boost table files for distance ranking, 2 dimensions.
## The tables have 977 elements each, where slots 0..15 represents
## square of distance being 0..15 while the remaining slots represents
## square of distance distance being
## (16 + (slot & 15)) << ((slot >> 4) - 1). Linear interpolation is
## used for distances "between" table slots.
##
## If "NULL" is given as the file name then all 2D distance boost values
## for that table will be set to 0.
rankprofile[].distance2dboosttable[].table string

## The lowest possible size of a ranked result. This is the lower ramp
## of the percentage specified in the binsize variable. The default is
## specified in fsearchrc.
rankprofile[].binlow int default=-1

## The high limit of the ranked result bin. If the percentage of the
## resultset specified in binsize is higher than this limit, this will be
## the max size. The default is specified in fsearchrc.
rankprofile[].binhigh int default=-1

## The size of the ranked results as a percentage of the total result
## set size. The percentage can be ramped off with the binlow and binhigh
## variables. The default is specified in fsearchrc.
rankprofile[].binsize double default=-1

## Minimum value for maximum value of number of 'posocc' entries for a word.
## The default is specified in fsearchrc.
rankprofile[].posbinlow int default=-1

## Maximum value for maximum value of number of 'posocc' entries for a word.
## The default is specified in fsearchrc.
rankprofile[].posbinhigh int default=-1

## The maximum value for number of 'posocc' entries for a word, specified
## as a percentage of the number of documents in the index.  If more
## entries are needed for evaluation, posocc entries are not used for that
## word and evaluation will be performed without full proximity support.
## The percentage can be ramped off with the posbinlow and posbinhigh
## variables. The default is specified in fsearchrc.
rankprofile[].posbinsize int default=-1

## Boost value that is added to the relevance score of hits from superior
## searches (searches where recall is sacrificed for better
## precision). The rank cutoff feature will not be affected by this
## feature (rank cutoff is applied before the superior boost).
rankprofile[].superiorboost int default=0

## Name of rank profile to be used when running superior searches
## (searches where recall is sacrificed for better precision). If not
## set, the current ranking profile will be used.
##
## If a profile for a superior search has this set then a superior^2
## search exist (with more recall sacrificed than for superior searches)
## and search behavior is slightly changed:
##
## If the search node has been asked to perform a superior search then an
## internal double fallthrough with (superior, superior^2) search is
## performed.  If the search node has been asked to perform an internal
## double fallthrough then a triple fallthrough with (normal, superior,
## superior^2) is performed.
rankprofile[].superiorname string default=""

## After all other rank calculations, the rank value is tuned according
## to the tunefactor and tunebias values. The rank value is modified
## as follows: new_rank = old_rank * tunefactor + tunebias.
rankprofile[].tunefactor double default=1.0

## After all other rank calculations, the rank value is tuned according
## to the tunefactor and tunebias values. The rank value is modified
## as follows: new_rank = old_rank * tunefactor + tunebias.
rankprofile[].tunebias int default=0

## A lower limit for the rankvalue of the results returned from the
## search node.  If rankcutoff.advanced is set to "true", determines
## the constant value used in the internal advanced rank cutoff
## calculations.  This roughly reflects the expected rank contribution
## of one good term.
## The rankcutoff.val value and the rankcutoff.advanced parameter
## may be used if you only want hits with a minimum relevancy to show
## up in the resultset.
## A value below zero means no rankcutoff is done.
rankprofile[].rankcutoff.val int default=-1

## When rankcutoff.val is in use, this flag controls whether to use
## an internal calculation is used for determining the rank cutoff
## value. If "false", use rankcutoff.val as a direct lower limit.
rankprofile[].rankcutoff.advanced bool default=false

## If set to "ON", use of posocc files is enabled, except when
## "forceemptyposoccs" is set in fsearchrc or posocc files doesn't exist.
## If set to "OFF", use of posocc files is disabled.
## If "NOTSET" the fsearchrc "proximity" parameter is used instead.
rankprofile[].proximity.full.enable enum { OFF, ON, NOTSET } default=NOTSET

## If set to "ON", use of firstoccproximity is enabled.
## If set to "OFF", use of firstoccproximity is disabled.
## When NOTSET use the firstoccproximity value in fsearchrc configuration.
rankprofile[].proximity.firstocc.enable enum { OFF, ON, NOTSET } default=NOTSET

## If set to "ON", use of proximity (cf. proximity and firstoccproximity)
## will affect phrases in addition to single words.
## If set to "OFF", proximity is never used for phrases.
## When NOTSET use the phraseproximity value in fsearchrc configuration.
rankprofile[].proximity.phrase.enable enum { OFF, ON, NOTSET } default=NOTSET

## Selects behavior when proximity can be used for two words but not three
## words while firstoccproximity can be used for three words.
## If set to "ON", then use proximity for two words.
## If set to "OFF", then use firstoccproximity for three words.
## When NOTSET use the proximitypairbeforefirstoccproximitytriple value
## in fsearchrc configuration.
rankprofile[].proximity.pairbeforefirstocctriple.enable enum { OFF, ON, NOTSET } default=NOTSET

## Selects behavior when proximity can be used for three words but not four
## words while firstoccproximity can be used for four words.
## If set to "ON", then use proximity for three words.
## If set to "OFF", then use firstoccproximity for four words.
## When NOTSET use the proximitytriplebeforefirstoccproximityquad value
rankprofile[].proximity.triplebeforefirstoccquad.enable enum { OFF, ON, NOTSET } default=NOTSET