In OlisaNsonwu/diyar: Record Linkage and Epidemiological Case Definitions in 'R'

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)

Version 0.5.1.9000

New features

Changes

Bug fixes

Version 0.5.1

New features

Changes

Bug fixes

• links() - Incorrect results in some situations. Resolved.
• links_af_probabilistic() - Failed in some situations. Resolved.

Version 0.5.0

New features

• New option ("semi") for the batched argument in links(). All matches are compared against the record-set in the next iteration. Therefore, the number of record-pairs increase exponentially as new matches are found. This means fewer record-pairs (memory usage) but a longer run time compared to the "no" option. Conversely, it leads to more record-pairs (memory usage) but a shorter run time compared to the "yes" option.
• New argument (batched) in episodes()
• New argument (split) in episodes(). Split the analysis in N-splits of strata. This leads to fewer record-pairs (and memory usage) but a longer run time.
• New argument (decode) in as.data.frame.pid(), as.data.frame.epid() and as.data.frame.pane()
• New function - episodes_af_shift(). A more vectorised approach to episodes() based on epidm::group_time().
• New function - links_wf_episodes(). Implantation of episodes() using links().

Changes

• Optimised episodes() and links(). Each iteration now uses less time and memory.
• link_id slot in pid objects is now a list.
• links() - records with missing values in a sub_criteria are now skipped at the corresponding iteration.
• Updated argument in links()- recursive. This now takes any of three options [c("linked", "unlinked", "none")] . [c("linked", "unlinked")] collectively were previously [TRUE], while ["none"] was previously [FALSE].
• as.epids() now calls make_episodes().
• The default value for the window argument in partitions() is now NULL
• as.data.frame() and as.data.list() now only creates elements/fields from non-empty fields
• id and gid slots in number_line objects are now integer(0) by default.
• episode_group(), record_group() and range_match_legacy() have been removed.
• ["recurisve"] episodes from episodes() are now presented as ["rolling"] episodes with reference_event = "all_records" i.e
  • Old syntax ~ episodes(..., episode_type == "recursive")
  • New syntax ~ episodes(..., episode_type == "rolling", reference_event = "all_records")

Bug fixes

• When recursive was TRUE, links() ended prematurely and therefore missed some matches. Resolved.
• recurrence_sub_criteria in episodes() was not implemented correctly and lead to incorrect linkage result in some instances. Resolved.
• overlap_method() - logical tests recycled incorrectly. Resolved.
• check_links argument - Option "g" implemented as option "l". Resolved.
• make_pairs_wf_source(). Created incorrect pairs. Resolved.
• case_sub_criteria and recurrence_sub_criteria in episodes() led to incorrect results. Resolved.

Version 0.4.2

New features

• New argument in merge_ids() - shrink and expand.
• New S3 method for class 'd_report' - plot.
• New S3 method for class 'sub_criteria' - format.
• New function - true(). Predefined logical test for use with sub_criteria().
• New function - false(). Predefined logical test for use with sub_criteria().
• New argument in links()- batched. Specify if all record pairs are created or compared at once ("no") or in batches ("yes").
• New argument in links()- repeats_allowed. Specify if record-pairs with duplicate elements should be created.
• New argument in links()- permutations_allowed. Specify if permutations of the same record-pair should be created.
• New argument in links()- ignore_same_source. Specify if record-pairs from different datasets should be created.
• New argument in eval_sub_criteria()- depth. First order of recursion.
• New function - sets() and make_sets(). Create permutations of record-sets.

Changes

• links() - When shrink is TRUE, records in a record-group must meet every listed match criteria and sub_criteria. For example, if pid_cri is 3, then the record must have meet matched another on the the first three match criteria.
• links() - pid@iteration now tracks when a record was dealt with instead of when it was assigned to a record-group. For example, a record can be closed (matched or not matched) at iteration 1 but assigned to a record-group at iteration 5.
• make_pairs() - x.* and y.* values in the output are now swapped.
• sub_criteria can now export any data created by match_func. To do this, match_func must export a list, where the first element is a logical object. See an example below.

library(diyar)
val <- rep(month.abb[1:5], 2); val
match_and_export <- function(x, y){
  output <- list(x == y, 
                 data.frame(x_val = x, y_val = y, is_match = x == y))
  return(output)
}
sub.cri.1 <- sub_criteria(
  val, match_funcs = list(match.export = match_and_export)
)

format(sub.cri.1, show_levels = TRUE)
eval_sub_criteria(sub.cri.1)

• links can now export any data created within a sub_criteria. To do this, the sub_criteria must be created as described above. See an example below

val <- 1:5
diff_one_and_export <- function(x, y){
  diff <- x - y
  is_match <- diff <= 1
  output <- list(is_match, 
                 data.frame(x_val = x, y_val = y, diff = diff,  is_match = is_match))
  return(output)
}
sub.cri.2 <- sub_criteria(
  val, match_funcs = list(diff.export = diff_one_and_export)
)
links(
  criteria = "place_holder", 
  sub_criteria = list("cr1" = sub.cri.2))

Bug fixes

• summary.epid() - Incorrect count for 'by episode type'. Resolved.
• episodes() - Incorrect results in some instances with skip_order. Resolved.
• make_ids() - Did not capture all records in that should be in a record-group when matches are recursive. Resolved.
• make_pairs() - Incorrect record-pairs in some instances. Resolved.
• eval_sub_criteria() - When output of match_func is length one, it's not recycled. Resolved.
• reverse_number_line() - Incorrect results in some instances. Resolved.
• links()- Incorrect iteration (pids slot) for non-matches. Resolved.
• links() and episodes() - Timing for each iteration was incorrect. Resolved.

Version 0.4.1

New features

• New function - overlap_method_names(). Overlap methods for a corresponding overlap method codes.
• Memory usage added to *with_report options for display.

Changes

• "chain" overlap method split into "x_chain_y" and "y_chain_x". "chain" will continue to be supported as a keyword for "x_chain_y" OR "y_chain_x" method
• "across" overlap method split into "x_across_y" and "y_across_x". "across" will continue to be supported as a keyword for "x_across_y" OR "y_across_x" methods
• "inbetween" overlap method split into "x_inbetween_y" and "y_inbetween_x". "inbetween" will continue to be supported as a keyword for "x_inbetween_y" OR "y_inbetween_x" methods
• Optimised overlaps().
• Some overlap method codes have changed. Please review any previously specified codes with overlap_method_names().

Bug fixes

• make_batch_pairs() (internal) created invalid record pairs. Resolved.

Version 0.4.0

New features

• New function - reframe(). Modify the attributes of a sub_criteria object.
• New function - link_records(). Record linkage by creating all record pairs as opposed to batches as with link().
• New function - make_pairs(). Create every combination of records-pairs for a given dataset.
• New function - make_pairs_wf_source(). Create records-pairs from different sources only.
• New function - make_ids(). Convert an edge list to a group identifier.
• New function - merge_ids(). Merge two group identifiers.
• New function - attrs(). Pass a set of attributes to one instance of match_funcs or equal_funcs.

Changes

• Optimised episodes_wf_splits()
• Optimised episodes() and links(). Reduced processing times.
• Three new options for the display argument. "progress_with_report", "stats_with_report" and "none_with_report". Creates a d_report; a status of the analysis over its run time.
• eval_sub_criteria(). Record-pairs are no longer created in the function. Therefore, index_record and sn arguments have been replaced with x_pos and y_pos.
• link_records() and links_wf_probabilistic(). The cmp_threshold argument has been renamed to attr_threshold.
• show_labels argument in schema(). Two new options - "wind_nm" and "length" to replace "length_label".

Bug fixes

• Incorrect wind_id list in episodes(..., data_link = "XX") in . Resolved.
• Incorrect link_id in links(..., recursive = TRUE). Resolved.
• iteration not recorded in some situations with episodes(). Resolved.
• skip_order ends an open episode. Resolved.
• NA in dist_wind_index and dist_epid_index when sn is supplied. Resolved.
• overlap_method_codes() - overlap method codes not recycled properly. Resolved.

Version 0.3.1

New features

• New function - delink(). Unlink identifiers.
• New function - episodes_wf_splits(). Wrapper function of episodes(). Better optimised for handling datasets with many duplicate records.
• New function - combi(). Numeric codes for unique combination of vectors.
• New function - attr_eval(). Recursive evaluation of a function on each attribute of a sub_criteria.

Changes

• Two new case_nm values - Case_CR and Recurrence_CR which are Case and Recurrence without a sub-criteria match.

Bug fixes

• Corrected length arrows in schema.epid.
• Corrected outcome of eval_sub_criteria with 1 result.

Version 0.3.0

New features

• New function - links_wf_probabilistic(). Probabilistic record linkage.
• New function - partitions(). Spilt events into sections in time.
• New function - schema(). Plot schema diagrams for pid, epid, pane and number_line objects.
• New functions - encode() and decode(). Encode and decode slots values to minimise memory usage.
• New argument in episodes() - case_sub_criteria and recurrence_sub_criteria. Additional matching conditions for temporal links.
• New argument in episodes()- case_length_total and recurrence_length_total. Number of temporal links required for a window/episode.
• New argument in links() - recursive. Control if matches can spawn new matches.
• New argument in links() - check_duplicates. Control the checking of logical tests on duplicate values. If FALSE, results are recycled for the duplicates.
• as.data.frame and as.list S3 methods for the pid, number_line, epid, pane objects.
• New option for episode_type in episodes() - "recursive". For recursive episodes where every linked events can be used as a subsequent index event.
• recurrence_from_last renamed to reference_event and given two new options.

Changes

• episodes() and links(). Speed improvements.
• Default time zone for an epid_interval or pane_interval with POSIXct objects is now "GMT".
• number_line_sequence() - splits number_line objects. Also available as a seq method.
• epid_total, pid_total and pane_total slots are populated by default. No need to used group_stats to get these.
• to_df() - Removed. Use as.data.frame() instead.
• to_s4() - Now an internal function. It's no longer exported.
• compress_number_line() - Now an internal function. It's no longer exported. Use episodes() instead.
• sub_criteria() - produces a sub_criteria object. Nested "AND" and "OR" conditions are now possible.
• case_overlap_methods, recurrence_overlap_methods and overlap_methods now take integer codes for different combinations of overlap methods. See overlap_methods$options for the full list. character inputs are still supported.

Bug fixes

• "Single-record" was wrong in links summary output. Resolved.

Version 0.2.0

New features

• Better support for Inf in number_line objects.
• Can now use multiple case_length or recurrence_length for the same event.
  • Can now use multiple overlap_methods for the corresponding case_length and recurrence_length.
• New function links() to replace record_group().
• New function sub_criteria(). The new way of supplying a sub_criteria in links().
• New functions exact_match(), range_match() and range_match_legacy(). Predefined logical tests for use with sub_criteria(). User-defined tests can also be used. See ?sub_criteria.
• New function custom_sort() for nested sorting.
• New function epid_lengths() to show the required case_length or recurrence_length for an analyses. Useful in confirming the required case_length or recurrence_length for episode tracking.
• New function epid_windows(). Shows the period a date will overlap with given a particular case_length or recurrence_length. Useful in confirming the required case_length or recurrence_length for episode tracking.
• New argument - strata in links(). Useful for stratified data linkage. As in stratified episode tracking, a record with a missing strata (NA_character_) is skipped from data linkage.
• New argument - data_links in links(). Unlink record groups that do not include records from certain data sources
• New convenience functions
  • listr(). Format atomic vectors as a written list.
  • combns(). An extension of combn to generate permutations not ordinarily captured by combn.
• New iteration slot for pid and epid objects
• New overlap_method - reverse()

Changes

• number_line() - l and r must have the same length or be 1.
• episodes() - case_nm differentiates between duplicates of "Case" ("Duplicate_C") and "Recurrent" events ("Duplicate_R").
• Strata and episode-level options for most arguments. This gives greater flexibility within the same instance of episodes().
  • Episode-level - The behaviour for each episode is determined by the corresponding option for its index event ("Case").
    • episode_type - simultaneously track both "fixed" and "rolling" episodes.
    • skip_if_b4_lengths - simultaneously track episodes where events before a cut-off range are both skipped and not skipped.
    • episode_unit - simultaneously track episodes by different units of time.
    • case_for_recurrence - simultaneously track "rolling" episodes with and without an additional case window for recurrent events.
    • recurrence_from_last - simultaneously track "rolling" episodes with reference windows calculated from the first and last event of the previous window.
  • Strata-level - The behaviour for each episode is determined by the corresponding option for its strata. Options must be the same in each strata.
    • from_last - simultaneously track episodes in both directions of time - past to present and present to past.
    • episodes_max - simultaneously track different number of episodes within the dataset.
• include_overlap_method - "overlap" and "none" will not be combined with other methods.
  • "overlap" - mutually inclusive with the other methods, so their inclusion is not necessary.
  • "none" - mutually exclusive and prioritised over the other methods (including "none"), so their inclusion is not necessary.
• Events can now have missing cut-off points (NA_real_) or periods (number_line(NA_real_, NA_real_)) case_length and recurrence_length. This ensures that the event does not become an index case however, it can still be part of different episode. For reference, an event with a missing strata (NA_character_) ensures that the event does not become an index case nor part of any episode.

Bug fixes

• fixed_episodes, rolling_episodes and episode_group - include_index_period didn't work in certain situations. Corrected.
• fixed_episodes, rolling_episodes and episode_group - dist_from_wind was wrong in certain situations. Corrected.

Version 0.1.0

New features

• New argument in record_group() - strata. Perform record linkage separately within subsets of a dataset.
• New argument in overlap(), compress_number_line(), fixed_sepisodes(), rolling_episodes() and episode_group() - overlap_methods and methods. Replaces overlap_method and method respectively. Use different sets of methods within the same dataset when grouping episodes or collapsing number_line objects. overlap_method and method only permits 1 method per per dataset.
• New slot in epid objects - win_nm. Shows the type of window each event belongs to i.e. case or recurrence window
• New slot in epid objects - win_id. Unique ID for each window. The ID is the sn of the reference event for each window
  • Format of epid objects updated to reflect this
• New slot in epid objects - dist_from_wind. Shows the duration of each event from its window's reference event
• New slot in epid objects - dist_from_epid. Shows the duration of each event from its episode's reference event
• New argument in episode_group() and rolling_episodes() - recurrence_from_last. Determine if reference events should be the first or last event from the previous window.
• New argument in episode_group() and rolling_episodes() - case_for_recurrence. Determine if recurrent events should have their own case windows or not.
• New argument in episode_group(), fixed_episodes() and rolling_episodes() - data_links. Unlink episodes that do not include records from certain data_source(s).
• episode_group(), fixed_episodes() and rolling_episodes() - case_length and recurrence_length arguments. You can now use a range (number_line object).
• New argument in episode_group(), fixed_episodes() and rolling_episodes() - include_index_period. If TRUE, overlaps with the index event or period are grouped together even if they are outside the cut-off range (case_length or recurrence_length).
• New slot in pid objects - link_id. Shows the record (sn slot) to which every record in the dataset has matched to.
• New function - invert_number_line(). Invert the left and/or right points to the opposite end of the number line
• New accessor functions -left_point(x)<-, right_point(x)<-, start_point(x)<- and end_point(x)<-

Changes

• overlap() renamed to overlaps(). overlap() is now a convenience overlap_method to capture ANY kind of overlap.
• "none" is another convenience overlap_method for NO kind of overlap
• expand_number_line() - new options for point; "left" and "right"
• compress_number_line() - compressed number_line object inherits the direction of the widest number_line among overlapping group of number_line objects
• overlap_methods - have been changed such that each pair of number_line objects can only overlap in one way. E.g.
  • "chain" and "aligns_end" used to be possible but this is now considered a "chain" overlap only
  • "aligns_start" and "aligns_end" use to be possible but this is now considered an "exact" overlap
• number_line_sequence() - Output is now a list.
• number_line_sequence() - now works across multiple number_line objects.
• to_df() - can now change number_line objects to data.frames.
  • to_s4() can do the reverse.
• epid objects are the default outputs for fixed_episodes(), rolling_episodes() and episode_group()
• pid objects are the default outputs for record_group()
• In episode grouping, the case_nm for events that were skipped due to rolls_max or episodes_max is now "Skipped".
• In episode_group() and record_group(), sn can be negative numbers but must still be unique
• Optimised episode_group() and record_group(). Runs just a little bit faster ...
• Relaxed the requirement for x and y to have the same lengths in overlap functions.
  • The behaviour of overlap functions will now be the same as that of standard R logical tests
• episode_group - case_length and recurrence_length arguments. Now accepts negative numbers.
  • negative "lengths" will collapse two periods into one, if the second one is within some days before the end_point() of the first period.
    • if the "lengths" are larger than the number_line_width(), both will be collapsed if the second one is within some days (or any other episode_unit) before the start_point() of the first period.
• cheat sheet updated

Bug fixes

• Recurrence was not checked if the initial case event had no duplicates. Resolved
• case_nm wasn't right for rolling episodes. Resolved

Version 0.0.3

Changes

• #7 episode_group(), fixed_episodes() and rolling_episodes() - optimized to take less time when working with large datasets
• episode_group(), fixed_episodes() and rolling_episodes() - date argument now supports numeric values
• compress_number_line() - the output (gid slot) is now a group identifier just like in epid objects (epid_interval)

Version 0.0.2

New feature

• pid S4 object class for results of record_group(). This will replace the current default (data.frame) in the next major release
• epid S4 object class for results of episode_group(), fixed_episodes() and rolling_episodes(). This will replace the current default (data.frame) in the next release
• to_s4() and to_s4 argument in record_group(), episode_group(), fixed_episodes() and rolling_episodes(). Changes their output from a data.frame (current default) to epid or pid objects
• to_df() changes epid or pid objects to a data.frame
• deduplicate argument from fixed_episodes() and rolling_episodes() added to episode_group()

Changes

• fixed_episodes() and rolling_episodes() are now wrapper functions of episode_group(). Functionality remains the same but now includes all arguments available to episode_group()
• Changed the output of fixed_episodes() and rolling_episodes() from number_line to data.frame, pending the change to epid objects
• pid_cri column returned in record_group is now numeric. 0 indicates no match.
• columns can now be used as criteria multiple times record_group()
• #6 number_line objects can now be used as a criteria in record_group()

Bug fixes

• #3 - Resolved a bug with episode_unit in episode_group()
• #4 - Resolved a bug with bi_direction in episode_group()

Version 0.0.1

Features

• fixed_episodes() and rolling_episodes() - Group records into fixed or rolling episodes of events or period of events.
• episode_group() - A more comprehensive implementation of fixed_episodes() and rolling_episodes(), with additional features such as user defined case assignment.
• record_group() - Multistage deterministic linkage that addresses missing data.
• number_line S4 object.
  • Used to represent a range of numeric values to match using record_group()
  • Used to represent a period in time to be grouped using fixed_episodes(), rolling_episodes() and episode_group()
  • Used as the returned output of fixed_episodes() and rolling_episodes()

OlisaNsonwu/diyar documentation built on July 8, 2024, 9:58 a.m.