add Lee (2001) local spatial Pearson #51
Conversation
ljwolf
requested review from
knaaptime,
sjsrey,
weikang9009,
renanxcortes and
jGaboardi
|
also maybe want to keep/return the "spatial smoothing scalars" from lee |
| class Spatial_Pearson(BaseEstimator): | ||
| def __init__(self, connectivity=None, permutations=999): | ||
| self.connectivity = connectivity | ||
| self.permutations = permutations |
There was a problem hiding this comment.
Should there be type caster here just in case the user give permutations as something other than an integer?
self.permutations = int(permutations)| class Local_Spatial_Pearson(BaseEstimator): | ||
| def __init__(self, connectivity = None, permutations=999): | ||
| self.connectivity = connectivity | ||
| self.permutations = permutations |
|
The significance testing is not correct; not sure how I'm screwing up the form of the null statistic, but it's definitely not right. Will look into. |
|
I think I've fixed up that code for the local inference. Stil need docstrings & unittests. Before I do that, can I get a ruling on the sklearn dependency from maintainers? If I rewrite this, I'd use array checking logic from spreg, or I'd write my own. |
|
Oh also on the API; this uses init().fit()-style stuff, which is different from the existing pattern. |
|
@ljwolf Are there any real drawbacks to having |
I think sklearn dependency is +1 |
|
docstrings and tests are added. |
|
@ljwolf is this WIP stil, or ready to merge? |
|
It's ready if the sklearn dependency is accounted for? |
This adds the Lee local spatial pearson estimators, alongside a wrapped version of the R code you can use to get them from spdep. This still needs