NEWS.md
thicken preserves missing values in the datetime column and adds them to the added column. The missing values were placed on the wrong position. They were placed on NA position + nr of NAs earlier in the datetime variable, instead of the NA position. Only the first missing value was on the correct position. Bug reported by github user github user levi-nagy.thicken has gained the ties_to_earlier argument. Logical with FALSE as default value. By default when the rounding argument in thicken is set to “up” and the original observation is equal to a value in the higher interval variable, the observation is mapped to the next value in the new variable. (For example 2019-04-14 13:00:00 would be mapped to 2019-04-14 14:00:00 when rounding is “up” and interval is “hour”.) This can be undesired. When this argument is set to TRUE tied observations are mapped to their own value (thus to one value earlier in the new variable). For completeness this argument also works when rounding is “down”. Then, when original and new value are tied, the original value is mapped to the previous value of the higher level interval variable. (For example 2019-04-14 13:00:00 will be mapped to 2019-04-14 12:00:00 when the interval is hour). Feature request by github user stribstrib.
thicken has gained a drop argument. Logical with FALSE as default value. If TRUE the thickened datetime value is dropped from the data frame. Idea by Adam Stone.
thicken converts datetime values to integers to match values to the higher interval. Because of this the function suffers from the YEAR2038 problem, meaning that it will not give output if a POSIXct is higher than 2038-01-19 03:14:07 because of integer overflow. The problem needs to be addressed in the future, for now a meaningful error is raised when a POSIXt with year 2038 or higher is fed to thicken. Problem detected by github users darneiri and Blundys.
An informative error is now thrown in pad, pad_cust, thicken, thicken_cust when a data frame does not have any rows. Requested by Julian During.
The functions thicken and thicken_cust no longer throw a warning when the input datetime variable is unsorted. The functions now silently return the a data frame with the same row order as the input data frame.
Error within padr for break_above error message is corrected. No longer prints the number of millions in millions. Bug found by Sharla Gelfand.
thicken is sped up significantly:
get_interval no longer applied to assess interval validity (its slow on large variables because it converts a POSIX to character). Rather validity is now compared after thickening by checking if results differs from original. Makes function approximately four times faster.
get_interval is sped up significantly:
to convert date to character format is used, instead of as.character. For large vectors it 4 to 5 times faster.
span_date and span_time are new functions and they are wrappers around seq.Date and seq.POSIXt respectively. Because of their default settings (minimal specification of date and datetimes and interval inference) they require very little inputs for straightforward spanning.
The closest_weekday function is introduced. It finds the closest requested weekday around the start of a datetime variable. This function helps to find quickly the start_val for thicken when the interval is “week”.
Two new functions are introduced that help with visualising interval data.
center_interval shifts the datetime variable from either the beginning or the end of the interval, to the center of the interval. This will improve visualisations such as dot plots and bar plots, where the timestamp is still considered to be continuous.
format_interval takes the start_value of an interval and infers the end. It uses strftime on both the start value and the end value, to create a character vector that reflects the full interval.
The _cust suite allows for user-specified spanning to use in thickening and padding.
to create an asymmetric spanning, subset_span subsets a datetime vector to the desired date and time points. These are provided in a list.
span_around takes a datetime variable as input and spans a variable around it of a desired interval. This automates finding the min and the man of x manually, determining which values are needed to create a span of a desired interval, and do the actual spanning.
Both pad and thicken will no longer break when there are missing values in the datetime variable. Rows containing missing values will be retained in the returned data frame. In the case of thicken they will remain on the same position as the input data frame. The added column will have a missing value as well. For pad all the rows with missing values will be moved to the end of the dataframe, since there is no natural position for them in the order of padded rows.
When time variable has NULL as timezone, also posix_to_date used to break (related to #14). This made thicken break when the desired interval is “day” or higher. This is now fixed by don’t regarding the timezone.
get_interval now throws an informative error when the datetime variable has missing values (#33).
pad now throws an informative error when the datetime variable is used in the grouping (#38)
added “ByteCompile: true” to DESCRIPTION.
pad no longer throws a message when the interval is specified (#31).
span around hours and minutes now start at the current hour and minute. This to make span_around sensible.
The interval is no longer limited to be of a single unit, for each of the eight interval sizes. Every time span accepted by seq.Date or seq.POSIXt is now accepted. Since the original implementation was fully around single-unit-intervals, some default behavior had to change. Because of it, this version is not entirely backwards compatible with earlier versions of padr. The following functions are affected:
thicken: the interval argument now has to be specified. In earlier versions it was optional. When it was not specified, the added variable was one interval level higher than that of the input datetime variable. With the widening of the interval definition, there is not longer a natural step up.
get_interval: does no longer only retrieve the interval of a datetime variable, but also its unit (the step size). For instance, the following would have returned “day” in the past, but will now return “2 day”:
date_var <- as.Date(c(‘2017-01-01’, ‘2017-01-03’, ‘2017-01-05’)) get_interval(date_var)
pad: when the interval is not specified, get_interval is applied on the datetime variable. Its outcome might now be different. When get_interval returns a different interval than it used to, pad will do the padding at this different interval. Extending the above example, the have resulted in a data frame with two padded rows:x <- data.frame(date_var, y = 1:3)
Since the interval of date_var used be “day”, there were missing records for 2017-01-02 and 2017-01-04. These records were inserted, with missing values for y. However, now the interval of date_var is “2 day” and on this level there is no need for padding. To get the original result the interval argument should be specified with “day”.
pad
Pad has been reimplemented
The function was slow when applied on many groups becuause it looped over them. Function has been reimplemented so it needs only one join to do the padding for all the groups simultaneously. dplyr functions are used for this new implementation, both for speed and coding clarity.
When applying pad to groups the interval is determined differently. It used to determine the interval separately for each of the groups. With the new interval definition this would often yield undesired results. Now, the interval on the full datetime variable, ignoring the groups. If the user would like to allow for differing intervals over the groups it is advised to use dplyr::do. See also the final example of pad.
Besides its own argument for grouping, pad does now also accepts the grouping from dplyr. Making the following two results equal:
x %>% dplyr::group_by(z) %>% pad x %>% pad(group = ‘z’)
Moreover, both pad and thicken now maintain the grouping of the input data_frame. The return from both functions will have the exact same grouping.
break_above
This new argument to pad is a safety net for situations where the returned dataframe is much larger than the user anticipated. This would happen when the datetime variable is of a lower interval than the user thought it was. Before doing the actual padding, the function estimates the number of rows in the result. If these are above break_above the function will break.
pad has gained a group parameter. This takes a character vector that indicates the column names within which group padding must be done. The returned data frame is complete for the grouping variable(s). Leaving no longer the doubt which record belongs to which group member, especially when start_val and / or end_val was specified.
Issue #8: pad does no longer break when datetime variable contains one value only. Returns x and a warning, if start_val and end_val are NULL and will do proper padding when one or both are specified.
Issue #9: when forgetting to specify at least one column, on which to apply the fill_ function, the fill_ function will now throw a meaningful error.
Issue #10: pad was broken with an error the interval was quarter, month, or year. This was done by check_start_end, even when neither a start_val nor an end_val was specified. It appeared that when concatenating POSIX vectors, as happened in the check_start_end function, the result is enforced to the timezone of the locale (including daylight savings time). This breaks the interval if the original vectors were not of this timezone. Workaround is implemented.