Changelog

Removed non-API calls to R in the C++ code.

Fixed noRemap additional issue by adding Rf_ prefix in the c++ code.

Fix: emr_track.mv did not move track attributes.
“exists” virtual track function without parameters now checks for existence of any value.

Fix: do not truncate expression names in emr_extract when tidy=TRUE.

Fixed compilation warnings on M1 Mac.
Added emr_time2posix, emr_posix2time, emr_time2char and emr_char2time functions.

Fixed clang warnings.
Fix: filter name to include operator when name is automatically generated

Allow filters names with a dot when creating an automatic virtual track name.
Fixed an error when multiple processes are calling emr_track.var.set on the same track.

Fixed memory misalignment errors.

Added emr_filters.info that returns filters information given a filter expression (e.g. f1 | f2).
Bug fix: vtrack on logical track to categorical track with no parameters returned an wrong results.
Fixed clang-UBSAN error when creating an empty track.
Added “modification_time” to emr_track.info.

Removed C++11 specification from Makevars.

fixed some CRAN warnings.

fix: emr_vtrack.create with func="exists" returned NA instead of 0 when the track didn’t have any value at the iterator interval.

Fix: deal with complex filters in emr_vtrack.name and emr_vtrack.create_from_name.

New functions: emr_vtrack.name and emr_vtrack.create_from_name allow creating virtual tracks with standard names.

Added some examples to the documentation.
Fix: crash when calling emr_track.create with expr=''

Added entries - a key/value store for storing simple strings and numbers in a naryn database. See emr_entries.ls, emr_entries.get, emr_entries.set, emr_entries.rm and emr_entries.reload.
Added string_to_var and var_to_string functions to generate valid variable names from strings.
emr_track.rm can now accept multiple tracks as input.
Do not allow importing patients that do not appear at patients.dob.
Fixed issue #83
Fix: emr_track.rm returned an error when given character(0).

Added include in order to compile with gcc13

Fixed c++ clang-UBSAN and gcc-UBSAN errors.
Fixed rchk warnings.

Fixed: hang when all iterator intervals were outside of stime and etime.

Fixed an additional clang warning.

removed emr_traceback function
Fixed C++ compilation warnings and possible memory leaks.

Removed global variables from the package - EMR_ROOTS, EMR_GROOT, EMR_UROOT, EMR_FILTERS and EMR_VTRACKS are now a part of a dedicated environment called .naryn. You can access them using .naryn::EMR_ROOTS etc.

Optimized the performance of the emr_track.exists function.

Check that attributes are in ASCII format.

Allow duplicated tracks when running emr_track.attr.set in batch mode.

Added batch mode for emr_track.attr.set and emr_track.attr.rm.
Added include_missing parameter emr_track.attr.export allowing missing tracks and attributes.

Added support for filters which are based on virtual tracks.

Added support for operator based filtering, see ‘operator’ argument in emr_filter.create
Added support for track attributes to logical tracks.
Bug fix: dbs order was changed to lexicographical order on emr_db.connect.
Bug fix: stray logical tracks on non-global dbs were shown.
Bug fix: emr_time functions failed when there was an NA in an integer vector.
Bug fix: emr_track.attr.get returned attribute name instead of value.
emr_track.var.get now returns NULL when variable doesn’t exist.
Make sure logical tracks cannot be created on non-global db.
Added db_id parameter to emr_track.ls
Removed cpp11 package dependency

Added support for multiple databases.
New functions: emr_filter.name and emr_filter.create_from_name allow creating filters with standard names.
Bug fix: allow calling emr_track.logical.create when values are a list of length 1.
Bug fix: wrong results when creating virtual track where the source is a logical track without parameters.

Added support for logical tracks.
emr_filter.create and emr_vtrack.create now return silently the name of the filter/vtrack.
Allow addition of entries with the same id/time/values to emr_track.addto.
emr_dist and emr_cor now have a dataframe parameter that returns the results in a tidy format.
Bug fix: wrong results in emr_dist when right=FALSE and values were a fraction.
Added convenience time functions (emr_time, year, months etc).
Support periodic iterators: emr_monthly_iterator, emr_yearly_iterator
Use the devtools ecosystem: roxygen documentation and testthat for tests.
Added a NEWS.md file to track changes to the package.

New concept: track attributes
New functions: emr_track.attr.get, emr_track.attr.set, emr_track.attr.rm, emr_track.attr.export.
Support filtering by track attributes in: emr_track.ls, emr_track.global.ls, emr_track.user.ls.

Fixed an occasional crash in functions receiving a track expression.
Redirect all messages and progress to stderr instead of stdout.

Fixed a resource leak effecting nearly all functions but manifesting mainly in emr_ids_coverage and emr_ids_vals_coverage by “protection stack overflow” error.

Bug fix in emr_track.mv: track variables are not moved.

Bug fix in emr_track.mv: error when moving a file between different file systems.

New function: emr_track.mv

Bug fix: occasional memory corruption in track expressions using OR operator in the filter.

Bug fix: occasional crash in emr_track.rm.
Bug fix: Ctrl+C doesn’t stop emr_track.import.

Bug fix: sub-second creation/deletion/modification of tracks in concurrent sessions might corrupt the database.

emr_track.import, emr_track.addto: run-time and memory consumption improvements.

Allow creation of empty tracks.

Support concurrent sessions adding / deleting / modifying tracks. As long as the tracks files are not manipulated outside of naryn, sessions should always resync the changes at the beginning of a new transaction and hence access up-to-date version of tracks and up-to-date list of existing tracks. Manipulating track files outside of naryn requires emr_db.reload.
emr_db.init: split load.on.demand parameter to global.load.on.demand and user.load.on.demand.
emr_db.init: new ‘do.reload’ parameter.
Run-time improvement in emr_db.init when load.on.demand==FALSE: previously - rebuild list of tracks. Currently - read track list from ‘.tracks’.
Major run-time and memory consumption improvement in emr_db.reload: previously - rebuild list of tracks and reload all tracks into memory. Currently - rebuild list of tracks and reload outdated tracks (if load.on.demand==FALSE) or kick them out of memory (if load.on.demand==TRUE).
emr_db.reload: while rebuilding the list DB gets locked, concurrent transactions will need to wait till the task completes.
Currently - each transaction starts with fetching a track list from ‘.tracks’, if ‘.tracks’ had been modified between the transactions. Outdated tracks are unloaded from the memory (if load.on.demand==TRUE) or reloaded (if load.on.demand==FALSE).
Previously - new transaction continues using old track list and outdated tracks up until emr_db.reload is called.
Bug fix: ‘.ids’ file is reloaded in each transaction even if the updated version of it is already in the memory.
emr_time2hour, emr_time2dayofmonth, emr_time2month, emr_time2year, emr_date2time do not require anymore to open a database.

Removed MINID/MAXID constants from global environment.
IDs for Beat iterator and Time Interval take their ids from ‘dob’ track.
Bug fix in emr_track.import, emr_track.addto: crash when source data frame has ‘ref’ but no ‘value’ column.
Bug fix in Time Intervals iterator: in some cases ids outside of subset might be produced.
DB cache file format has changed. The file is now called ‘.tracks’. The first call to emr_db.init will recreate this file after a full DB reload.

Removed MINTIME/MAXTIME constants from global environment.
All functions receiving stime/etime parameters and an iterator: defaults can no longer be used for stime/etime when the iterator type is Ids, Beat or Extended Beat.
DB cache file format has changed, the first call to emr_db.init will recreate this file after a full DB reload.

Bug fix in emr_track.info: returned path is NULL.

More robust handling of .naryn-cache file when several users simultaneously add / delete tracks. Might resolve various glitches when emr_db.reload() is needed.

emr_vtrack.create / emr_vtrack.attr.src: allow ‘src’ to be a data frame. Syntax: src=list(data.frame, T/F), where T/F indicates whether the data is treated as categorical (otherwise: quantitative).
Bug fix in emr_filter.attr.src: error when ‘src’ is a data frame

Bug fix: various functions may suddenly return an empty data frame

Bug fix: crash if the period of beat / extended beat iterator is set to 0

Bug fix: crash in certain scenarios while using ‘filter’ parameter. For example: filter=“(track1 & track2) & (track3 & track4)”

Bug fix in emr_filter.attr.src: error if ‘src’ is ID-Time table

Bug fix in emr_filter.create: named filters become inconsistent if src of the existing filter is changed from global track to user track or vice versa
Bug fix in emr_vtrack.create: virtual tracks become inconsistent if src of the existing filter is changed from global track to user track or vice versa
Bug fix in emr_filter.ls: if filters exist with src based on user space tracks, only these filters are returned
Bug fix in emr_vtrack.ls: if virtual tracks exist with src based on user space tracks, only these virtual tracks are returned
Run time optimizations in emr_filter.create and emr_vtrack.create

Bug fix in emr_db.subset: error if src==NULL
Bug fix in virtual track, func=“stddev”: crash when stddev is computed over a single value

Bug fix in emr_track.percentile: lower bound is returned even if lower=FALSE

Bug fix: MINTIME/MAXTIME may be wrongly set. As a result performance might be compromised and invalid results might be produced when defaults are used for stime,etime. Please call emr_db.reload() to set the correct values!

Replaced “categorial” with “categorical” in all the parameters and documentation

Bug fix: “child process ended unexpectedly” errors, crashes and hang ups whilst multitasking when running out of memory

New iterator type: “Id-Time” Intervals (dataframe with ‘id’, ‘stime’, ‘etime’ columns)
New filter type: “Id-Time” Intervals

Set/retrieve a single attribute of a virtual track via: emr_vtrack.attr.src, emr_vtrack.attr.func, emr_vtrack.attr.params, emr_vtrack.attr.keepref, emr_vtrack.attr.time.shift, emr_vtrack.attr.id.map, emr_vtrack.attr.filter
Set/retrieve a single attribute of a named filter via: emr_filter.attr.src, emr_filter.attr.keepref, emr_filter.attr.time.shift, emr_filter.attr.val, emr_filter.attr.expiration

Filters can be applied now to the virtual tracks
emr_vtrack.create: new ‘filter’ parameter

emr_vtrack.create: new “sample.time” function

Switched from custom random seed control (options(grnd.seed=…)) to R’s standard (set.seed)
emr_db.subset: ‘seed’ parameter removed
New function: emr_db.subset.info

Issue a warning if beat / extended beat iterator is used without a filter and it produces more than options(emr_warning.itr.no.filter.size = …) points (default: 100000). To suppress this type of warning use options(emr_warning.itr.no.filter.size = -1)

emr_dist, emr_cor: new “right” parameter (similar to “cut” function)

emr_vtrack.create: new “sample” function

Bug in multitasking: buffer overflow and possible memory corruption when the child process generates an error

emr_vtrack.create: time.shift column can now be used within ‘id.map’ data frame uniquely remapping times and not only ids

Multitasking stability issues

Bug fix in emr_screen: in multitasking mode returns wrong results
Bug fix: multitasking + debug mode might lead to hanging

Bug fix in track expression evaluation engine: virtual track with func=“size” returns NaN instead of 0 when no values fall into the iterator interval

New function: emr_annotate

New functions: emr_track.var.get, emr_track.var.set, emr_track.var.rm, emr_track.var.ls
emr_track.import, emr_track.create: introduce restrictions for track naming

Bug fix in track expression evaluation engine: if two or more virtual tracks differ only by values (supplied in params) and are used in one track expression, the value of the first one is used for the rest

Bug fix: in certain circumstances an error “Database was not loaded. Please call emr_db.init.” is printed

Bug fix in emr_filter.create: warning is issued when ID-Time table is used as source
Bug fix in emr_ids_coverage: when src is an ID-Time table and the filter is used “..emr_tmp_filter does not exist (9)” error is generated
Bug fix in emr_ids_vals_coverage: when src is an ID-Time table “..emr_tmp_filter does not exist (9)” error is generated

New function: emr_track.percentile
emr_summary: renamed two members of the result member from “Total intervals”/“NaN intervals” to “Total values”/“NaN values”

New functions: emr_track.exists, emr_vtrack.exists, emr_filter.exists
emr_filter.create, emr_vtrack.create - run-time optimizations

New function: emr_track.ids
emr_ids_coverage - new additional parameters: stime, etime, filter
emr_ids_vals_coverage - run-time optimizations

emr_filter.create: support vector of values for “val” parameter

Redesigned emr_track.import function - supports data frame, does not allow anymore to add records to the existing track
New function: emr_track.addto
New “space” argument to emr_track.create function

New filter on values for all virtual track functions applied to categorical tracks. The filter is given via “param” in emr_vtrack.create
Bug fix: virtual track function “exists” might not work on non-integer values where double(val) != float(val)

Bug fix: various functions might hang when emr_debug and emr_multitasking options are both TRUE

New virtual track functions for simple linear regression model: lm.slope, lm.intercept

Bug fix: occasional defunc processes AND/OR hanging in multitasking mode

Mild run-time optimizations due to switching to C++11

Bug fix: MAXTIME and MAXID might be incorrectly set which might have grave effect on iterators, multitasking, etc. Please call emr_db.reload() on all your trackdbs
New option: emr_debug=T/F

Bug fix in multitasking mode: child process does not die after generating an error

Run-time optimization in a filter when NOT is applied to a list of time intervals

Run-time optimization in a filter when NOT is applied to a list of ids

emr_db.init: new “load.on.demand” parameter
emr_db.init: very fast run-time when “load.on.demand” parameter is used

Multitasking for all functions that use track expression evaluation engine
emr_extract and emr_screen: new “sort” parameter
New R options controlling multitasking behavior: emr_multitasking, emr_min.processes, emr_max.processes
Run-time optimizations in track expression evaluation engine in single-task mode
Bug-fix: wrong iteration on dense tracks when subsets are used

Bug-fix: in certain cases incorrect results when NOT is used in the filter
Run-time optimization when NOT is used in the filter

emr_extract: receive track expressions as a vector
emr_extract: new “tidy” parameter
emr_extract: “colnames” parameter renamed to “names”
Run-time optimizations

Run-time optimization on long filters
Run-time optimization when NOT operator is used in a filter

Bug fix in filters: incorrect result when OR and NOT are used together

Bug fix in filters: expiration and val produce incorrect results when used together
Bug fix in emr_db.init: occasional crash due to too many files open

New functions: emr_ids_coverage and emr_ids_vals_coverage
Bug fix: MINTIME, MAXTIME, MINID, MAXID constants are not updated after emr_track.create, emr_track.import and emr_track.rm are called

New function: emr_cor. Returns correlation statistics for each pair of a track expression vector

New concept: subsets. Use subsets to define train / test sets. Please refer user manual for further details
New functions: emr_db.subset, emr_db.subset.ids
New option: emr_rnd.seed. Currently used by emr_quantiles

New global EMR_TIME variable that stores the times of the current iterator
Return an empty data frame instead of NULL in emr_extract and emr_screen when result set is empty
Added user manual (still incomplete but covers already tracks, virtual tracks, track expressions and iterators).

Run-time optimizations

New iterator type: beat iterator aligned to the time that appears in a data frame or a track (aka birthday). Syntax:
iterator=list(period, data.frame(id=…, time=…))
iterator=list(period, trackname)

Note: each id must appear only once in the data frame / track.

New concept: global tracks and user tracks. emr_track.import creates tracks in global tracks directory, emr_track.create does it in user tracks directory.
New functions: emr_track.create, emr_track.rm, emr_track.global.ls, emr_track.user.ls
emr_db.init: new user.dir parameter
emr_db.init: renamed root parameter to global.dir
emr_track.info: added file path to the output.
emr_track.import: removed reload_db parameter. Imported track will automatically be efficiently reloaded.

emr_filter.create: new “val” parameter (slicing)
emr_filter.create: new “expiration” parameter. Please refer reference manual for usage.
Run-time optimizations

New concept: named filters. No more parameters (time shift, keepref) are allowed in track expression filters. Please use named filters instead.
New functions: emr_filter.create, emr_filter.rm, emr_filter.ls, emr_filter.info
New virtual track functions: earliest.time, latest.time, closest.earlier.time, closest.later.time
emr_time2…: return NaN if the time is NaN
emr_time2…: return numeric instead of integer
4+5 = ability to use emr_time2… in track expressions
Bug fix in emr_track.import: do not allow creation of a track if identically named virtual track already exists

emr_vtrack.create: new id.map parameter
emr_vtrack.create: renamed time_shift parameter to time.shift

emr_dist: allow implicit breaks (use breaks=NULL) if the track expression constitutes of a categorical track or a virtual track that is based on a categorical track

Run-time optimizations when track expression equals to a track or a virtual track

New logical parameter in emr_track.import: “categorical”
Track format had been changed
New function: emr_track.unique - returns sorted unique values of a track. Hint: use this functions to set breaks when categorical track is used
emr_vtrack.create: “val” parameter had been removed
emr_vtrack.create: new functions “value”, “size”, “frequent”, “exists”
emr_vtrack.create: functions “earliest”, “latest” and “closest” return -1 in case of ambiguity for categorical tracks
Interpret data frame with id, time columns as patient-time even if ref column is missing
Bug fix: emr_db.init continues failing after it failed once (for example: due to invalid track format)

New functions: emr_track.info, emr_summary
Enable patient ids to be used as iterators
Enable patient ids to be used in filters
Patient ids format for 2 & 3: data frame with the first column named “id”
Enable time intervals to be used as iterators
Enable time intervals to be used in filters
Time intervals format for 5 & 6: data frame with the first columns named “stime” and “etime”
Run-time optimization when OR is used in iterator filter
Report error id-time list used for iterator / filter contains two records that differ only by reference when keepref == FALSE

Bug fix in filter when time after shift precedes the epoch
Bug fix in beat iterator: invalid progress report

Enable beat iterator (iterator=integer) produce references if keepref=T
Support filters when beat iterator is used
1+2 means that one can now run complex iterators. For example iterator=track1 | track2, keepref = R can be simulated with help of a beat iterator and a filter as: iterator=1, keepref=R, filter=“track1(R) | track2(R)”
Support filters when patient-time list is used as an iterator
Enable patient-time lists to be used in the filter. The syntax is identical to tracks, e.g.: filter=“mylist1(F, c(10,20)) | track & !mylist2”
Prefetch track files into shared memory: initial emr_db.init is expected to be slower, however subsequent access to the tracks is faster

Use shared memory to store tracks in RAM

Added regression tests
Bug fix in emr_date2time: crash if year is out of range
Bug fix in emr_vtrack.create: functions percentile.upper.min, percentile.lower.min, percentile.upper.max, percentile.lower.max generate an error

Added track filters. Examples of valid filters:
“track1”, “track1(T)”, “track1(F, 2)”, “track1(F, c(10, 20))”, “track1 & track2”, “track1 | track2”, “!track1”, “!track1 & (track2(F, c(10, 20)) | !track3(T))”
Bug fix: track format changes depending on the OS.

New function: emr_dist

Increase the maximal number of simultaneously used virtual tracks to 10K

Add “val” parameter to emr_vtrack.create

Allow addition of data into the existing tracks emr_track.import (use “add” parameter)

New functions: emr_quantiles, emr_track.import
Support new functions in virtual tracks: percentile.upper, percentile.lower, percentile.max.upper, percentile.max.lower, percentile.min.upper, percentile.min.lower
Track format changed, old track format is not supported anymore
New keepref parameter in emr_vtrack.create
Changes in vtrack evaluation

First version

naryn 2.6.302024-09-27

naryn 2.6.29

naryn 2.6.28

naryn 2.6.27

naryn 2.6.262024-01-15

naryn 2.6.252023-11-30

naryn 2.6.24

naryn 2.6.232023-08-17

naryn 2.6.22

naryn 2.6.212023-06-02

naryn 2.6.20

naryn 2.6.19

naryn 2.6.18

naryn 2.6.17

naryn 2.6.16

naryn 2.6.15

naryn 2.6.142023-01-09

naryn 2.6.132022-12-14

naryn 2.6.122022-11-22

naryn 2.6.11

naryn 2.6.10

naryn 2.6.92022-10-13

naryn 2.6.8

naryn 2.6.7

naryn 2.6.6

naryn 2.6.5

naryn 2.6.4

naryn 2.6.3

naryn 2.6.2

naryn 2.6.1

naryn 2.6.0

naryn 2.5.10

naryn 2.5.9

naryn 2.5.8

naryn 2.5.7

naryn 2.5.6

naryn 2.5.5

naryn 2.5.4

naryn 2.5.3

naryn 2.5.2

naryn 2.5.1

naryn 2.5.0

naryn 2.4.4

naryn 2.4.3

naryn 2.4.2

naryn 2.4.1

naryn 2.4.0

naryn 2.3.12

naryn 2.3.11

naryn 2.3.10

naryn 2.3.9

naryn 2.3.8

naryn 2.3.7

naryn 2.3.6

naryn 2.3.5

naryn 2.3.4

naryn 2.3.3

naryn 2.3.2

naryn 2.3.1

naryn 2.3.0

naryn 2.2.11

naryn 2.2.10

naryn 2.2.9

naryn 2.2.8

naryn 2.2.7

naryn 2.2.6

naryn 2.2.5

naryn 2.2.4

naryn 2.2.3

naryn 2.2.2

naryn 2.2.1

naryn 2.2.0

naryn 2.1.11

naryn 2.1.10

naryn 2.1.9

naryn 2.1.8

naryn 2.1.7

naryn 2.1.6

naryn 2.1.5