9:53 p.m.
I'm not documenting this yet because it needs more testing, but in
current git I've enabled a new operation on string-valued series:
logical product. (Snapshots to follow.)
If sv1 and sv2 are string-valued series, then
series sv3 = sv1 * sv2
now yields another string-valued series with value si_sj at
observations where sv1 has value si and sv2 has value sj.
A simple example is afforded by the R dataset warpbreaks (csv
version attached). This includes two "factor" series, wool (with
values "A" and "B") and tension (with values "L",
"M" and "H").
If you multiply them together you get a series with 6 distinct
values, "A_L" to "B_H".
<hansl>
open warpbreaks.csv
series wt = wool*tension
print -o
</hansl>
Allin
11 a.m.
Am 16.10.20 um 23:53 schrieb Allin Cottrell:
I'm not documenting this yet because it needs more testing, but
in
current git I've enabled a new operation on string-valued series:
logical product. (Snapshots to follow.)
If sv1 and sv2 are string-valued series, then
series sv3 = sv1 * sv2
now yields another string-valued series with value si_sj at observations
where sv1 has value si and sv2 has value sj.
A simple example is afforded by the R dataset warpbreaks (csv version
attached). This includes two "factor" series, wool (with values "A"
and
"B") and tension (with values "L", "M" and "H").
If you multiply them together you get a series with 6 distinct values,
"A_L" to "B_H".
This is pretty useful, Allin! I often face the "problem" that I have two
or more string-valued series which need to be concatenated for creating
a unique identifier before one can set a panel structure.
I also ran the script below with a slightly modified data set where
column "tension_w_nan" includes an NA value. In this case the resulting
series is NA as well which makes somehow sense as multiplying a valid
number with NA yields NA. I guess that's as expected given the current
implementation.
<hansl>
open "../tmp/warpbreaks.csv"
series wt = wool * tension
series wt2 = wool * tension_w_nan
series tw2 = tension_w_nan * wool
print wt tension_w_nan wt2 tw2 -o
</hansl>
However, one comment on the current syntax/ implementation using the
asterisk symbol: In Python/Pandas (we don't have to mimic that behavior
though!) string concatenation is usually done by the summation symbol:
<series sv3 = sv1 + sv2> -- at least that's my experience.
In case one of the RHS variables is NA, let's say sv1[i]=NA, you still
get sv3[i]=NA+sv2[i] which equals the string value of sv[i].
Also, let me go even a step further. A more flexible version may be a
function with three arguments: "sv1", "bridge" and "sv2"
where "bridge"
(not the most ideal parameter name) may be optional allowing for a
user-defined bridging string such as underscore ("_") in the current
implementation.
Artur
11:44 a.m.
On Sat, 17 Oct 2020, Artur Tarassow wrote:
Am 16.10.20 um 23:53 schrieb Allin Cottrell:
> I'm not documenting this yet because it needs more testing, but in current
> git I've enabled a new operation on string-valued series: logical product.
> (Snapshots to follow.)
>
> If sv1 and sv2 are string-valued series, then
>
> series sv3 = sv1 * sv2
>
> now yields another string-valued series with value si_sj at observations
> where sv1 has value si and sv2 has value sj.
>
> A simple example is afforded by the R dataset warpbreaks (csv version
> attached). This includes two "factor" series, wool (with values
"A" and
> "B") and tension (with values "L", "M" and
"H").
> If you multiply them together you get a series with 6 distinct values,
> "A_L" to "B_H".
This is pretty useful, Allin! I often face the "problem" that I have two or
more string-valued series which need to be concatenated for creating a unique
identifier before one can set a panel structure.
I concur: this is a very nice idea, very useful, and handled in a very
elegant way. I only have two remarks: a suggestion and a proposal.
(a) we already use the "^" operator for performing what I see as a very
similar operation on lists. It would seem to me more consistent if we used
"^" instead of "*".
(b) why not extend this syntax to string arrays, or even vectors? Imagine
how cool it would be to do something like
<pseudo-hansl>
s = defarray("a", "b")
ss = s ^ seq(1,3)
</pseudo-hansl>
and have ss be a 6-element string array containing "a_1", "a_2" and so
on.
Also, let me go even a step further. A more flexible version may be a
function with three arguments: "sv1", "bridge" and "sv2"
where "bridge" (not
the most ideal parameter name) may be optional allowing for a user-defined
bridging string such as underscore ("_") in the current implementation.
Or perhaps, the "bridge" character could be a libset variable.
-------------------------------------------------------
Riccardo (Jack) Lucchetti
Dipartimento di Scienze Economiche e Sociali (DiSES)
Università Politecnica delle Marche
(formerly known as Università di Ancona)
r.lucchetti(a)univpm.it
http://www2.econ.univpm.it/servizi/hpp/lucchetti
-------------------------------------------------------
1:08 p.m.
On Sat, 17 Oct 2020, Riccardo (Jack) Lucchetti wrote:
On Sat, 17 Oct 2020, Artur Tarassow wrote:
> Am 16.10.20 um 23:53 schrieb Allin Cottrell:
>> I'm not documenting this yet because it needs more testing, but in current
>> git I've enabled a new operation on string-valued series: logical product.
>> (Snapshots to follow.)
>>
>> If sv1 and sv2 are string-valued series, then
>>
>> series sv3 = sv1 * sv2
>>
>> now yields another string-valued series with value si_sj at observations
>> where sv1 has value si and sv2 has value sj.
>>
>> A simple example is afforded by the R dataset warpbreaks (csv version
>> attached). This includes two "factor" series, wool (with values
"A" and
>> "B") and tension (with values "L", "M" and
"H").
>> If you multiply them together you get a series with 6 distinct values,
>> "A_L" to "B_H".
>
> This is pretty useful, Allin! I often face the "problem" that I have two or
> more string-valued series which need to be concatenated for creating a
> unique identifier before one can set a panel structure.
I concur: this is a very nice idea, very useful, and handled in a very
elegant way. I only have two remarks: a suggestion and a proposal.
(a) we already use the "^" operator for performing what I see as a very
similar operation on lists. It would seem to me more consistent if we used
"^" instead of "*".
Good point. I wasn't quite sure which operator symbol to borrow for
this purpose but consistency with lists makes a good argument for
'^'. I'll make that change.
(b) why not extend this syntax to string arrays, or even vectors?
Imagine how
cool it would be to do something like
<pseudo-hansl>
s = defarray("a", "b")
ss = s ^ seq(1,3)
</pseudo-hansl>
and have ss be a 6-element string array containing "a_1", "a_2" and
so on.
Would be cool, yes. But to my mind the next priority for work on
string-valued data would be arranging for string values to be used
in plots. For example, in a plot with the --dummy option we should
clearly be setting x-tics with strings, not numbers, if the discrete
x variable is string-valued.
> Also, let me go even a step further. A more flexible version may
be a
> function with three arguments: "sv1", "bridge" and
"sv2" where "bridge"
> (not the most ideal parameter name) may be optional allowing for a
> user-defined bridging string such as underscore ("_") in the current
> implementation.
Or perhaps, the "bridge" character could be a libset variable.
Yes, could be. To make the bridge an argument we'd have to implement
this via a function rather than an operator.
And/or, a fairly simple generalization that could help here is to
make strsub() and regsub() apply to string-valued series (and arrays
of strings) as well as plain strings.
One other thought: R uses "." as bridge. At first I thought we
couldn't do that since "." can't be used in a gretl identifier. But
that was a confusion: the result is not supposed to be an
identifier! So in the first instance I'm inclined to switch to dot
as default bridge character, unless anyone sees a good reason not
to.
Allin
7:29 p.m.
On Sat, 17 Oct 2020, Allin Cottrell wrote:
On Sat, 17 Oct 2020, Riccardo (Jack) Lucchetti wrote:
> (a) we already use the "^" operator for performing what I see as a very
> similar operation on lists. It would seem to me more consistent if we used
> "^" instead of "*".
Good point. I wasn't quite sure which operator symbol to borrow for this
purpose but consistency with lists makes a good argument for '^'. I'll make
that change.
That's now in git.
[A] fairly simple generalization [...] is to make strsub() and
regsub() apply to string-valued series (and arrays of strings) as
well as plain strings.
Now in git too, not documented pending more testing. But I'm
satisfied that at least the generalization doesn't break the
original case, where the first argument to strsub and regsub is just
a single string.
One other thought: R uses "." as bridge. At first I thought
we
couldn't do that since "." can't be used in a gretl identifier.
But that was a confusion: the result is not supposed to be an
identifier!
Also in git, the inserted character or "bridge" between string
values in sv1 ^ sv2 is '.' rather than '_'.
Allin
1844
days inactive
1845
days old
4 comments
3 participants
participants (3)
-
Allin Cottrell -
Artur Tarassow -
Riccardo (Jack) Lucchetti