new predict

[josef...@gmail.com]

Linear Regression - WLS

In [12]:

mod_wls = WLS(y, X, weights=1./w)

res_wls = mod_wls.fit()

pred_res = get_prediction(res_wls)

pf = pred_res.summary_frame()

pf.head()

Out[12]:

	mean	mean_se	mean_ci_lower	mean_ci_upper	obs_ci_lower	obs_ci_upper
0	5.110110	0.213037	4.681771	5.538449	3.611625	6.608595
1	5.275544	0.205344	4.862672	5.688417	3.781406	6.769682
2	5.440978	0.197792	5.043292	5.838665	3.950966	6.930991
3	5.606413	0.190395	5.223598	5.989227	4.120300	7.092525
4	5.771847	0.183173	5.403553	6.140141	4.289408	7.254286

In [3]:

[josef...@gmail.com]

switching to plain text

I was trying out a notebook to see if the new `get_prediction` feels
like containing too many hoops that we have to jump through. We have
many plans for extensions, but it might become overkill for simple use
cases.

http://nbviewer.ipython.org/gist/josef-pkt/4c8d1f0eadf03375d8aa

PR is at https://github.com/statsmodels/statsmodels/pull/2151

I'm running currently low in my ability to come up with or decide on
good names, so the interface will be "experimental" in 0.7.

mean -> conditional_expectation
mean_se -> standard_error_of_conditional_expectation
....

only sometimes we reuse the code and then it might be something different.


----
The plan is to extend the prediction facilities in statsmodels 0.8.
The current version in PR is focused on the confidence interval and
setting up the structure.

predict as method stays simple so it can be used in a loop for example
for cross-validation without extra overhead. All additional predict
features will be added as a result instance of a new get_prediction
method.

The discussion for the design for enhancements is spread out over
several issues on github.

Josef

[josef...@gmail.com]

It's missing a nice plotting methods for fill_between confidence intervalls.

My 5 minute attempts are not pretty

see first plot in updated notebook
http://nbviewer.ipython.org/gist/josef-pkt/4c8d1f0eadf03375d8aa


Josef