FeatureGenerator.h File Reference

FeatureGenerator : creating features from raw signals. More...

#include <InfraMed/InfraMed/InfraMed.h>
#include <Logger/Logger/Logger.h>
#include <MedProcessTools/MedProcessTools/RepProcess.h>
#include <MedProcessTools/MedProcessTools/MedFeatures.h>
#include <SerializableObject/SerializableObject/SerializableObject.h>
#include <MedProcessTools/MedProcessTools/MedModelExceptions.h>
#include <MedTime/MedTime/MedTime.h>
#include <MedAlgo/MedAlgo/MedAlgo.h>
#include <MedAlgo/MedAlgo/MedLM.h>
#include <cfloat>
#include <regex>

Go to the source code of this file.

Data Structures
class	FeatureGenerator
class	BasicFeatGenerator
	A Basic Stats Generator for calcing simple statics on time window. More...
class	AgeGenerator
	Age Generator. More...
class	SingletonGenerator
	Singleton. More...
class	GenderGenerator
	Gender. More...
struct	BinnedLmEstimatesParams
	BinnedLinearModels : parameters. More...
class	BinnedLmEstimates
	BinnedLinearModels : Apply a set of liner models to generate features. More...
class	RangeFeatGenerator
	RangeFeatGenerator : Generate features for a time range with value signal (for example drug) More...
class	ModelFeatGenerator
	Use a model to generate predictions to be used as features. More...
class	TimeFeatGenerator
class	AttrFeatGenerator
	Attribute Feature Generator: creating features from samples attributes. More...
class	CategoryDependencyGenerator
	Creates multipal features based on categorical values and statistical dependency strength by Age,Gender groups. More...

Macros
#define	DEFAULT_FEAT_GNRTR_NTHREADS   8

Enumerations
enum	FeatureGeneratorTypes {   FTR_GEN_NOT_SET , FTR_GEN_BASIC , FTR_GEN_AGE , FTR_GEN_SINGLETON ,   FTR_GEN_GENDER , FTR_GEN_BINNED_LM , FTR_GEN_SMOKING , FTR_GEN_KP_SMOKING ,   FTR_GEN_UNIFIED_SMOKING , FTR_GEN_RANGE , FTR_GEN_DRG_INTAKE , FTR_GEN_ALCOHOL ,   FTR_GEN_MODEL , FTR_GEN_TIME , FTR_GEN_ATTR , FTR_GEN_CATEGORY_DEPEND ,   FTR_GEN_EMBEDDING , FTR_GEN_EXTRACT_TBL , FTR_GEN_ELIXHAUSER , FTR_GEN_DIABETES_FINDER ,   FTR_GEN_LAST }
enum	BasicFeatureTypes {   FTR_LAST_VALUE = 0 , FTR_FIRST_VALUE = 1 , FTR_LAST2_VALUE = 2 , FTR_AVG_VALUE = 3 ,   FTR_MAX_VALUE = 4 , FTR_MIN_VALUE = 5 , FTR_STD_VALUE = 6 , FTR_LAST_DELTA_VALUE = 7 ,   FTR_LAST_DAYS = 8 , FTR_LAST2_DAYS = 9 , FTR_SLOPE_VALUE = 10 , FTR_WIN_DELTA_VALUE = 11 ,   FTR_CATEGORY_SET = 12 , FTR_CATEGORY_SET_COUNT = 13 , FTR_CATEGORY_SET_SUM = 14 , FTR_NSAMPLES = 15 ,   FTR_EXISTS = 16 , FTR_CATEGORY_SET_FIRST = 17 , FTR_MAX_DIFF = 18 , FTR_FIRST_DAYS = 19 ,   FTR_RANGE_WIDTH = 20 , FTR_CATEGORY_SET_FIRST_TIME = 21 , FTR_SUM_VALUE =22 , FTR_LAST_NTH_VALUE = 23 ,   FTR_CATEGORY_SET_LAST_NTH = 24 , FTR_TIME_SINCE_LAST_CHANGE = 25 , FTR_LAST }
enum	TimeRangeTypes { TIME_RANGE_CURRENT = 0 , TIME_RANGE_BEFORE = 1 , TIME_RANGE_LAST }
enum	BinnedLMSamplingStrategy { BINNED_LM_TAKE_ALL = 0 , BINNED_LM_STOP_AT_FIRST = 1 , BINNED_LM_STOP_AT_LAST = 2 , BINNED_LM_LAST }
	BinnedLinearModels : which time-points to take.
enum	RangeFeatureTypes {   FTR_RANGE_CURRENT = 0 , FTR_RANGE_LATEST = 1 , FTR_RANGE_MAX = 2 , FTR_RANGE_MIN = 3 ,   FTR_RANGE_EVER = 4 , FTR_RANGE_TIME_DIFF = 5 , FTR_RANGE_RECURRENCE_COUNT = 6 , FTR_RANGE_TIME_COVERED = 7 ,   FTR_RANGE_LAST_NTH_TIME_LENGTH = 8 , FTR_RANGE_TIME_DIFF_START = 9 , FTR_RANGE_TIME_INSIDE = 10 , FTR_RANGE_LAST }
enum	TimeFeatTypes {   FTR_TIME_YEAR = 0 , FTR_TIME_MONTH = 1 , FTR_TIME_DAY_IN_MONTH = 2 , FTR_TIME_DAY_IN_WEEK = 3 ,   FTR_TIME_HOUR = 4 , FTR_TIME_MINUTE = 5 , FTR_TIME_DATE = 6 , FTR_TIME_LAST }
	Time Feature Generator: creating sample-time features (e.g. More...
enum class	category_stat_test { chi_square = 1 , mcnemar = 2 }

Functions
FeatureGeneratorTypes	ftr_generator_name_to_type (const string &generator_name)
void	get_window_in_sig_time (int _win_from, int _win_to, int _time_unit_win, int _time_unit_sig, int _win_time, int &_min_time, int &_max_time, bool boundOutcomeTime=false, int outcome_time=-1)
	gets a [-_win_to, -_win_from] window in win time unit, and returns [_min_time, _max_time] window in signal time units relative to _win_time boundOutcomeTime is used to future time windows when looking to the future to limit the time window till the outcomeTime
TimeRangeTypes	time_range_name_to_type (const string &name)
	Conversion between time-range type and name.
string	time_range_type_to_name (TimeRangeTypes type)
void	get_updated_time_window (UniversalSigVec &time_range_usv, TimeRangeTypes type, int time_unit_range_sig, int time_unit_win, int time_unit_sig, int time, int win_from, int &updated_win_from, int win_to, int &updated_win_to, bool delta_win, int d_win_from, int &updated_d_win_from, int d_win_to, int &updated_d_win_to)
void	get_updated_time_window (TimeRangeTypes type, int range_from, int range_to, int time, int _win_from, int _win_to, int &updated_win_from, int &updated_win_to)

Detailed Description

FeatureGenerator : creating features from raw signals.

Enumeration Type Documentation

BasicFeatureTypes

enum BasicFeatureTypes

Enumerator
FTR_LAST_VALUE	"last" - Last Value in Window
FTR_FIRST_VALUE	"first" - First Value in Window
FTR_LAST2_VALUE	"last2" - One before last value in Window
FTR_AVG_VALUE	"avg" - Mean value in Window
FTR_MAX_VALUE	"max" - Max value in Window
FTR_MIN_VALUE	"min" - Min value in Window
FTR_STD_VALUE	"std" - Standart Dev. value in Window
FTR_LAST_DELTA_VALUE	"last_delta" - Last delta. last-previous_last value
FTR_LAST_DAYS	"last_time" - time diffrence from prediction time to last time has signal in range of values
FTR_LAST2_DAYS	"last2_time" - time diffrence from prediction time to one previous last time has signal in range of values
FTR_SLOPE_VALUE	"slope" - calculating the slope over the points in the window
FTR_WIN_DELTA_VALUE	"win_delta" - diffrence in value in two time windows (only if both exists, otherwise missing_value). value in [win_from,win_to] minus value in [d_win_from, d_win_to]
FTR_CATEGORY_SET	"category_set" - boolean 0/1 if the signal has the value in the given lut (which initialized by the "sets" that can be specific single definition or name of set definition. the lookup is hierarchical)
FTR_CATEGORY_SET_COUNT	"category_set_count" - counts the number of appearnces of sets in the time window
FTR_CATEGORY_SET_SUM	"category_set_sum" - sums the values of appearnces of sets in the time window
FTR_NSAMPLES	"nsamples" - counts the number of times the signal apear in the time window
FTR_EXISTS	"exists" - boolean 0/1 if the signal apears in the time window
FTR_CATEGORY_SET_FIRST	"category_set_first" - boolean 0/1 if the signal apears in the time window and did not appear ever before the window
FTR_MAX_DIFF	"max_diff" maximum diff in window
FTR_FIRST_DAYS	"first_time" time diffrence from prediction time to first time with signal
FTR_RANGE_WIDTH	"range_width" maximal value - minimal value in a given window time frame
FTR_CATEGORY_SET_FIRST_TIME	"category_set_first_time" - first time of category set found in the time window
FTR_SUM_VALUE	"sum" - sum of values in window
FTR_LAST_NTH_VALUE	"last_nth" : (set also N_th parameter to use), get the last N_th in window, 0 is last, 1 is last2, etc.
FTR_CATEGORY_SET_LAST_NTH	"category_set_last_nth" : (set also N_th parameter to use), check is the last N_th in window is in the given set
FTR_TIME_SINCE_LAST_CHANGE	"time_since_last_change" : go over states signal, take last time since the value changed

FeatureGeneratorTypes

enum FeatureGeneratorTypes

Enumerator
FTR_GEN_BASIC	"basic" - creates basic statistic on time windows - BasicFeatGenerator
FTR_GEN_AGE	"age" - creating age feature - AgeGenerator
FTR_GEN_SINGLETON	"singleton" - take the value of a time-less signale - SingletonGenerator
FTR_GEN_GENDER	"gender" - creating gender feature - GenderGenerator (special case of signleton)
FTR_GEN_BINNED_LM	"binnedLm" or "binnedLM" - creating linear model for esitmating feature in time points - BinnedLmEstimates
FTR_GEN_SMOKING	"smoking" - creating smoking feature - SmokingGenerator
FTR_GEN_KP_SMOKING	"kp_smoking" - creating smoking feature - KpSmokingGenerator
FTR_GEN_UNIFIED_SMOKING	"unified_smoking" - creating smoking feature - UnifiedSmokingGenerator
FTR_GEN_RANGE	"range" - creating RangeFeatGenerator
FTR_GEN_DRG_INTAKE	"drugIntake" - creating drugs feature coverage of prescription time - DrugIntakeGenerator
FTR_GEN_ALCOHOL	"alcohol" - creating alcohol feature - AlcoholGenerator
FTR_GEN_MODEL	"model" - creating ModelFeatGenerator
FTR_GEN_TIME	"time" - creating sample-time features (e.g. differentiate between times of day, season of year, days of the week, etc.). Creates TimeFeatGenerator
FTR_GEN_ATTR	"attr" - creating features from samples attributes. Creates AttrFeatGenerator
FTR_GEN_CATEGORY_DEPEND	"category_depend" - creates features from categorical signal that have statistical strength in samples - CategoryDependencyGenerator
FTR_GEN_EMBEDDING	"embedding" - allows applying a pre trained embedding model to incorporate features into matrix. Creates EmbeddingGenerator
FTR_GEN_EXTRACT_TBL	"extract_tbl" - extract values from table with keys and rules to join with each patient. Creates FeatureGenExtractTable
FTR_GEN_ELIXHAUSER	Calculate Current Elixhauser given latest DRG and Diagnosis information. Creates ElixhauserGenerator.
FTR_GEN_DIABETES_FINDER	"diabetes_finder" - Diabetes Finder feature. Creates DiabetesFinderGenerator

RangeFeatureTypes

enum RangeFeatureTypes

Enumerator
FTR_RANGE_CURRENT	"current" - finds the value of the time range signal that intersect with win_from. signal start_time is before this time and signal end_time is after this time point
FTR_RANGE_LATEST	"latest" - finds the last value of the time range signal, that there is intersection of time signal range with the defined time window
FTR_RANGE_MAX	"max" - finds the maximal value of the time range signal, that there is intersection of time signal range with the defined time window
FTR_RANGE_MIN	"min" - finds the minimal value of the time range signal, that there is intersection of time signal range with the defined time window
FTR_RANGE_EVER	"ever" - boolean 0/1 - finds if there is intersection between signal time window and the defined time window with specific lut value. uses set.
FTR_RANGE_TIME_DIFF	"time_diff" - returns time diffrences between first intersection(if check_first is True) between signal time window and the defined time window with specific lut value. uses set. if check_first is false returns the time diffrences between last intersection between signal time window and the defined time window. prediction time minus the last intersecting signal end time window. if the last intersction if time ranges has no match to sets value and check_first is false will return -win_to value, otherwise missing value "recurrence_count" - count the number of time the event occur shortly after a previous event, there is an intersection of the time signal range with the defined time window previous event does not need to intersect the time window.
FTR_RANGE_TIME_COVERED	"time_covered" : give a time window, sum up all the times in ranges that intersect the time window
FTR_RANGE_LAST_NTH_TIME_LENGTH	"last_nth_time_len" : gives the length (in win_time_unit) of the last_n range in the window. If in middle of range, cuts to current time
FTR_RANGE_TIME_INSIDE	< "time_inside" : checks if the prediction time point is currently INSIDE a range, if not returns 0, if it is , then how long since the start.

TimeFeatTypes

enum TimeFeatTypes

Time Feature Generator: creating sample-time features (e.g.

differentiate between times of day, season of year, days of the week, etc.)

Enumerator
FTR_TIME_YEAR	Year (as is)
FTR_TIME_MONTH	Month of year (0-11)
FTR_TIME_DAY_IN_MONTH	Day of the month (0-30)
FTR_TIME_DAY_IN_WEEK	Day of the week (0-6)
FTR_TIME_HOUR	Hour of the day (0-23)
FTR_TIME_MINUTE	Minute of the hout (0-59)
FTR_TIME_DATE	Completete date (as is)

TimeRangeTypes

enum TimeRangeTypes

Enumerator
TIME_RANGE_CURRENT	"current" - consider only the current time-range
TIME_RANGE_BEFORE	"before" - consider anything before the current time-range