ai.h2o.automl.colmeta

Class ColMeta

java.lang.Object

water.Iced

ai.h2o.automl.colmeta.ColMeta

All Implemented Interfaces:

java.io.Externalizable, java.io.Serializable, java.lang.Cloneable, water.Freezable

public class ColMeta
extends water.Iced

Column Meta Data Holds usual rollup stats and additional interesting bits of information.

See Also:

Serialized Form

Field Summary

Fields

Modifier and Type	Field and Description
double	_avgUniquesPerChunk
int	_cardinality
double[]	_chunkBoundaries
boolean	_chunksMonotonicallyIncreasing
double[]	_dist
double	_fourthMoment
hex.tree.DHistogram	_histo Meta data collected on the first pass over this column.
int	_idx
boolean	_ignored
boolean	_isCategorical
boolean	_isClass
boolean	_isDate
boolean	_isNonIidId
boolean	_isNumeric
boolean	_isRowBasedId
double	_kurtosis
double	_median
long	_MRTaskMillis
java.lang.String	_name
byte	_nameType
long	_numUniques
double	_percentNA
boolean	_response
double	_sigma
double	_skew
boolean	_stratify
double	_thirdMoment
water.fvec.Vec	_v
double	_variance
double	_vif
double[]	_weightMult
static java.lang.String[]	METAVALUES
static double	SQLNAN

Constructor Summary

Constructors

Constructor and Description
ColMeta(water.fvec.Vec v, java.lang.String colname, int idx, boolean response)
ColMeta(water.fvec.Vec v, java.lang.String colname, int idx, boolean response, boolean ignored) Build metrics against the response column for this vec.

Method Summary

Modifier and Type	Method and Description
void	fillColMeta(java.util.HashMap<java.lang.String,java.lang.Object> cm, int idFrame)
boolean	isClassification()
static java.util.HashMap<java.lang.String,java.lang.Object>	makeEmptyColMeta()
java.lang.String	selectBasicTransform()

Methods inherited from class water.Iced

asBytes, clone, copyOver, frozenType, read, readExternal, readJSON, reloadFromBytes, toJsonString, write, writeExternal, writeJSON

Methods inherited from class java.lang.Object

equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Field Detail

METAVALUES

public static final java.lang.String[] METAVALUES

_v

public final water.fvec.Vec _v

_nameType

public byte _nameType

_name

public final java.lang.String _name

_idx

public final int _idx

_ignored

public boolean _ignored

_response

public boolean _response

_percentNA

public double _percentNA

_variance

public double _variance

_sigma

public double _sigma

_stratify

public boolean _stratify

_dist

public double[] _dist

_weightMult

public double[] _weightMult

_isNumeric

public boolean _isNumeric

_isCategorical

public boolean _isCategorical

SQLNAN

public static final double SQLNAN

See Also:

Constant Field Values

_isClass

public boolean _isClass

_histo

public hex.tree.DHistogram _histo

Meta data collected on the first pass over this column. This is done by computing a very coarse histogram (~500 bins) for numeric vectors, and building a table of occurrences for enum/string vectors (see NB below). Additionally, the time it takes to MRTask over the column is collected and stored. This will be helpful in understanding the time it takes to build DTree instances in the tree-based models. NB: For any table'ing done on a vec, there's a limit of 10K uniques before a more efficient (i.e., MDowle-style radix sort) scheme should be engaged.

_MRTaskMillis

public long _MRTaskMillis

_thirdMoment

public double _thirdMoment

_fourthMoment

public double _fourthMoment

_kurtosis

public double _kurtosis

_skew

public double _skew

_cardinality

public int _cardinality

_vif

public double _vif

_numUniques

public long _numUniques

_avgUniquesPerChunk

public double _avgUniquesPerChunk

_chunksMonotonicallyIncreasing

public boolean _chunksMonotonicallyIncreasing

_chunkBoundaries

public double[] _chunkBoundaries

_median

public double _median

_isRowBasedId

public boolean _isRowBasedId

_isNonIidId

public boolean _isNonIidId

_isDate

public boolean _isDate

Constructor Detail

ColMeta

public ColMeta(water.fvec.Vec v,
               java.lang.String colname,
               int idx,
               boolean response,
               boolean ignored)

Build metrics against the response column for this vec. Four scenarios: this vec response vec 1. Categorical, Categorical - build int[this_vec.domain()][counts_per_response_cat] 2. Categorical, Numerical - build double[this_vec.domain()][mean,sd] 3. Numerical, Categorical - build double[response_vec.domain()][mean,sd] 4. Numerical, Numerical - not defined, could do a stupid thing with quantile

ColMeta

public ColMeta(water.fvec.Vec v,
               java.lang.String colname,
               int idx,
               boolean response)

Method Detail

isClassification

public boolean isClassification()

makeEmptyColMeta

public static java.util.HashMap<java.lang.String,java.lang.Object> makeEmptyColMeta()

selectBasicTransform

public java.lang.String selectBasicTransform()

fillColMeta

public void fillColMeta(java.util.HashMap<java.lang.String,java.lang.Object> cm,
                        int idFrame)