Add trim option to marginal rate computation

[benjello]

No description provided.

[benjello]

@adrienpacifico: you can use this to trim the large marginal rate you may hit because of nonlinearities like seuil de non versement and kinks.
@cbenz you can merge this with no impact for other openfisca users

[adrienpacifico]

@benjello
Thanx a lot. I did not try your trim function.
To deal with that I'm currently using a smoothing function that gives the average value between n-1 and n+1 element of the array if the variation is too large.

What should be the typical value of your trim parameter ?

[benjello]

trim is a list of two elements. The marginal rates that are higher than the max of these two parameters or less than the min are converted to nan and thus not appear in the plots.

[adrienpacifico]

Thanx for you answer.
It solves big arterfacts that are higher than a given value (due to threshold, kinks, etc). But it doesn't solve my fat lines issues :-) (as reported in #349).

I need to find a good way to deal with that, other than reducing the number of observations :-).
(désolé je me sers un peu de cet espace comme si c'était un canard en plastique).

P.s : you might want to put a docstring for the function such that ones know that trim is supposed to be a two element array/list/tupple/whatever else.

[benjello]

Your problem is due to rounding. Reduce the number of points.

Le sam. 10 oct. 2015 16:44, adrienpacifico [email protected] a
écrit :

Thanx for you answer.
It solves big arterfacts that are higher than a given value (due to
threshold, kinks, etc). But it doesn't solve my fat lines issues :-) (as
reported in #349 #349
).

I need to find a good way to deal with that, other than reducing the
number of observations :-).
(désolé je me sers un peu de cet espace comme si c'était un canard en
plastique).

P.s : you might want to put a docstring for the function such that ones
know that trim is supposed to be a two element array/list/tupple/whatever
else.

—
Reply to this email directly or view it on GitHub
#350 (comment)
.

[adrienpacifico]

Yes I know I've been trying to do something a bit more fancy today, it's
not working perfectly but it's still something to improve.
Le 10 oct. 2015 18:21, "Mahdi Ben Jelloul" [email protected] a
écrit :

Your problem is due to rounding. Reduce the number of points.

Le sam. 10 oct. 2015 16:44, adrienpacifico [email protected] a
écrit :

Thanx for you answer.
It solves big arterfacts that are higher than a given value (due to
threshold, kinks, etc). But it doesn't solve my fat lines issues :-) (as
reported in #349 <#349

).

I need to find a good way to deal with that, other than reducing the
number of observations :-).
(désolé je me sers un peu de cet espace comme si c'était un canard en
plastique).

P.s : you might want to put a docstring for the function such that ones
know that trim is supposed to be a two element array/list/tupple/whatever
else.

—
Reply to this email directly or view it on GitHub
<
#350 (comment)

.

—
Reply to this email directly or view it on GitHub
#350 (comment)
.

[benjello]

@cbenz, you can merge this without trouble