-
Notifications
You must be signed in to change notification settings - Fork 16
/
todos.txt
62 lines (48 loc) · 1.86 KB
/
todos.txt
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
General
-------
- testing testing testing
- documentation documentation documentation
- speed comparison of opt module
- compare fits to sklearn baseline
- add validate_init() methods to every object.
Losses
------
- Cox
- Gamma
- Hindge
- DWD
- smooth quantile
Penalties to add
----------------
- group lasso default weights
- (long run) group lasso SVD preprocessing
- add multiple overlapping penalties (e.g. sparse group, sparse fused, elastic net, etc)
- infimal sum penalties
Linear regression noise estimates
---------------------------------
- finish adding natural and organic lasso
- implement for multiple responses
- (long run) perhaps add some methods for other types of penalties
- communicate the base estimator to the Inferencer object.
Degrees of freedom
------------------
- add DoF estimators for other penalties: group lasso, generalized Lasso, exclusive lasso, nuclear norm
- We currently estimate DoFs using the raw data even though we fit on the processed (standardized data). Think through if this messes anything up?
- add alternative DoF estimators for SCAD
- addd DoF simulations
Tuning
------
- tuning framework for multiple penalties
- re add pen_val_max for ridge
- max pen val for: generalized ridge, generalized Lasso
- Generalized CV
- 1se rule for multiple parameters (e.g. elastic net)
- add better print progress for tuning in parallel e.g. see https://github.com/scikit-learn/scikit-learn/blob/2beed55847ee70d363bdbfe14ee4401438fba057/sklearn/model_selection/_validation.py#L454
Solvers
-------
- probably make solvers use update_weights() and update_pen_val() instead of update_penalty()
- add numeric proxs for smooth loss functions (so they work with ADMM)
- re-add cvxpy backend
- re-add andersonCD backend
- re-add Linear and Quadartic program solvers for quantile
- bulid coordinate descent framework (e.g. based on https://arxiv.org/abs/1410.1386)