simpleR/TODO at master · IPL-UV/simpleR · GitHub

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
MODELS
  - Neural networks are not too well represented. Add (1) rbfnets, and (2) Feedforward Backpropagation nets in http://www.mathworks.com/matlabcentral/fileexchange/38310-deep-learning-toolbox
  - ELM needs to be checked
  - TGP doesnt work well!
  - RKS needs tuning the sigma param!

CROSSVAL
  - cross validate the number of trees in RF and bag and boost!
  - cross validate all methods the same way?

COMBINATION OF MODELS
  - Ensemble methods

LARGESCALE
  - include the option of divide-and-conquer strategy
  - modify the kernel test functions to run in batches...

ANALYSIS OF RESULTS
  - Improve assessment.m with other specific measures for regression (nMSE, loglik, Average Posterior Predictive KL Divergence)

FEATURE RANKING
  - trees analysis
  - rf feature ranking
  - GPs: ARD and SA
  - All: permutation!
  - sensitivity analysis in kernel methods

CONFIDENCE INTERVALS
  - confidence intervals?

WARNINGS
  - if a multioutput problem is detected, ask and run MO models or one by one SO model
  - Warnings on expected memory, cpu cost, before training
  - Warnings on illposing, run PCA before that?

MISC
   - Control verbosity
   - compute cputime and memory for a complete comparison
   - 1D problems do not work!
   - Cleanup folders, remove duplicated functions, remove unused functions ---> check dependencies
   - Data normalization and scaling!