forked from eminSerin/NBS-Predict
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathCHANGELOG.txt
40 lines (36 loc) · 2.35 KB
/
CHANGELOG.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
## [v1.0.0-beta1] - 13.04.2021
- NBS-Predict is now in v1.0.0-beta1.
- Add user MANUAL and data analysis tutorial.
- Fix history.mat file issue.
- Add option to set a specific random seed.
- Fix several minor issues in the GUI.
- Add functions for alternative algorithms (elastic net, lasso, p-value thresholding, top 5% and connectome-based predictive modeling (Shen et al., 2017)).
- Add simulation function to evaluate the performance of NBS-Predict and alternative algorithms on simulated network data.
- Add plot functions to plot simulation results.
## [v1.0.0-alpha3] - 06.09.2020
- NBS-Predict is now in v1.0.0-alpha3.
- Fix several bugs in the GUI.
- Fix bug in the contrast network generation algorithm.
- Fix minor bugs in assert statements.
- Minor tweaks in BrainNetViewer to visualize 3D Brain Image better.
- Add functionality to generate synthetic data for regression problems.
- Replace maximum percentile with p-value for pre-filtering.
- Changed several functions with regards to change in pre-filtering.
- Fix minor issue in function generating parameter combinations.
- divSelect, divSelectWide, simulatedAnnealing search algorithms were removed.
- Bayesian Optimization is now usable.
- Update hyperparameter space for decision tree estimator.
- Drop weight distribution plot as it might be misleading for interpretation of feature importance.
- Import circularGraph toolbox written by Paul Kassebaum to NBS-Predict to display circular network figures.
## [v1.0.0-alpha2] - 09.02.2020
- NBS-Predict is now in v1.0.0-alpha2.
- Fix several bugs in the GUI.
- Fix bug in correlation metric.
- Change default parameters for searching algorithms. Now they have more liberal parameters (e.g., reduced number of iterations).
- Add function to control confounds.
- Add a menu in GUI to choose a scaling method.
- SVM Regression is now built-in function instead of LIBSVM.
- Add Linear Regression, Logistic Regression and Linear Discriminant Analysis
- Add MinMaxScaler, StandardScaler and MaxAbsScaler
- LIBSVM is now removed. Instead, builtin SVM functions are now available to use. They give the same prediction performance as LIBSVM, but are significantly faster.
- Add preprocessing step in shrinkMat function, that features where information is less than 10% are removed. In this way, very sparse features are now removed. It might be increased to 20% in future updates.