gegen group misc issues
#4
Assignees
Labels
Comments
|
I note in the README that gegen does not respect group order; perhaps I should change the function name since you find it misleading? (I'll fix the wrong option bug) |
|
Another problem is that the new IDs seem to have holes: sysuse auto, clear
gegen id = group(turn) in 20/30
tab id(groups 1-4, 6, 8, etc. are missing) |
|
I see. This is not a but per se, though, since all the groups are there and the ID is unique, but I agree that coupled with the order issue it makes the ID rather confusing. |
mcaceresb
added a commit
that referenced
this issue
Jul 27, 2017
Enhancements * `gegen varname = group(varlist)` no longer has holes, as noted in issue #4 * `gegen` and `gcollapse` fall back on `collapse` and `egen` in case there is a collision. Future releases will implement an internal way to resolve collisions. This is not a huge concern, as SpookyHash has no known vulnerabilities (I believe the concern raied in issue #2 was base on a typo; see [here](rurban/smhasher#34)) and the probability of a collision is very low. * `gegen varname = group(varlist)` now has a consistency test (though the group IDs are not the same as `egen`'s, they should map to the `egen` group IDs 1 to 1, which is what the tests now check for). Bug fixes * Additional fixes for issue #1 * Apparentlly the argument Stata passes to plugins have a maximum length. The code now makes sure chuncks are passed when the PATH length will exceed the maximum. The plugin later concatenates the chuncks to set the PATH correctly.
mcaceresb
added a commit
that referenced
this issue
Jul 27, 2017
Bug fixes * `gegen` no longer ignores unavailable options, as noted in issue #4, and now it throws an error.
mcaceresb
added a commit
that referenced
this issue
Jul 27, 2017
gtools-0.6.5 through gtools-0.6.9 Enhancements * Addressed the possible issue noted in issue #3 and the functions now use mata and extended macro functions as applicable. * `gegen varname = group(varlist)` no longer has holes, as noted in issue #4 * `gegen` and `gcollapse` fall back on `collapse` and `egen` in case there is a collision. Future releases will implement an internal way to resolve collisions. This is not a huge concern, as SpookyHash has no known vulnerabilities (I believe the concern raied in issue #2 was base on a typo; see [here](rurban/smhasher#34)) and the probability of a collision is very low. * `gegen varname = group(varlist)` now has a consistency test (though the group IDs are not the same as `egen`'s, they should map to the `egen` group IDs 1 to 1, which is what the tests now check for). * The function now checks numerical variabes to see if they are integers. Working with integers is faster than hashing. * The function is now smarter about generating targets. In prior versions, when the target statistic was a sum the function would force the target type to be `double`. Now if the source already exists and is a float, the function now checks if the resultimg sum would overflow. It will only recast the source as double for collapsing if the sum might overflow, that is, if `_N * min < -10^38` or `10^38 < _N * max` (note +/- 10^38 are the largest/smallest floats stata can represent; see `help data_types`). Bug fixes * `gegen` no longer ignores unavailable options, as noted in issue #4, and now it throws an error. * `gegen varname = tag(varlist)` no longer tags missing values, as noted in issue #5 * Additional fixes for issue #1 * Apparentlly the argument Stata passes to plugins have a maximum length. The code now makes sure chuncks are passed when the PATH length will exceed the maximum. The plugin later concatenates the chuncks to set the PATH correctly. * Fixed issue #1 * The problem was that the wrapper I wrote to print to the Stata console has a maximum buffer size; when it tries to print the new PATH it encounters an error when the string is longer than the allocated size. Since printing this is unnecessary and will only ever be used for debugging, I no longer print the PATH. * Debugging issue #1 on github (in particular, `env_set` on Windows). * Removed old debugging code that had been left uncommented * Improved out-of-memory message (now links to relevant help section).
|
The IDs should no longer have have holes, and extraneous options now give an error. See |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
A few issues with
gegen:The last command fails as the IDs are not the same
gegen id = group(turn) , wrongoptionShould give a warning that an incorrect option was used.
The text was updated successfully, but these errors were encountered: