Bayesian networks | Morphometric analyzer - NeuroSuites

Bayesian networks - BayeSuites

This project has received funding from the European Union’s Horizon 2020 Framework Programme for Research and Innovation under Specific Grant Agreement No. 785907 (HBP SGA2)

1. Upload dataset/Bayesian network

Select the dataset and features:

(Load discrete dataset example)

(Load continuous dataset example)

No data available. If you want to upload your dataset please go here and then come back here.

Upload only structure (graph):

Formats supported: .csv or .gzip Apache Parquet file
(first column is the id, first row are nodes names).

Note: For large networks (thousands of nodes) it is recommended to upload
the network as a .gzip Apache Parquet file.
Very large networks can take up to a couple of minutes to load.

Structure 1

eFIygMtkLd5ZHLZt1BvcCSgI37WGGTiLbQd2LUnSbTnM4T9z3XRKzu1b0Rxxs99X

Structure 2 (to compare)

eFIygMtkLd5ZHLZt1BvcCSgI37WGGTiLbQd2LUnSbTnM4T9z3XRKzu1b0Rxxs99X

Upload only continuous parameters:

This must be uploaded after uploading a graph file or a Bayesian network file

Formats supported: .json

(Template file)

eFIygMtkLd5ZHLZt1BvcCSgI37WGGTiLbQd2LUnSbTnM4T9z3XRKzu1b0Rxxs99X

Upload a Bayesian network file (structure + discrete parameters):

Formats supported: .bif

(Bayesian networks example files from the bnlearn website).

(Load example BN discrete file) (Load example BN continous file)

Loading

Class features (optional):

eFIygMtkLd5ZHLZt1BvcCSgI37WGGTiLbQd2LUnSbTnM4T9z3XRKzu1b0Rxxs99X

(Template file)

Use this to upload external info about groups of nodes:

eFIygMtkLd5ZHLZt1BvcCSgI37WGGTiLbQd2LUnSbTnM4T9z3XRKzu1b0Rxxs99X

Loading dataset...

Estimated time to finish:

Data uploaded

Additional parameters uploaded

2. Learn the structure of the Bayesian network

Select the structure learning algorithm:

Learning algorithm:

Pearson correlation algorithm parameters

Note: this method does not detect edges directionalities so they have been assigned arbitrarily in order to be able to learn the Gaussian parameters. Edges weights have been squared to be able to draw the edges thickness in a meaningful way.

Backend:

Mutual information algorithm parameters

Note: this method does not detect edges directionalities so they have been assigned arbitrarily in order to be able to learn the Gaussian parameters.

Backend:

Linear regression algorithm parameters

Note: this method does not detect edges directionalities so they have been assigned arbitrarily in order to be able to learn the Gaussian parameters. Edges weights have been squared to be able to draw the edges thickness in a meaningful way.

Backend:

Graphical Lasso algorithm parameters

Note: this method does not detect edges directionalities so they have been assigned arbitrarily in order to be able to learn the Gaussian parameters. Edges weights have been squared to be able to draw the edges thickness in a meaningful way.

Backend:

Genie3 algorithm parameters

Note: this method does not detect edges directionalities so they have been assigned arbitrarily in order to be able to learn the Gaussian parameters. Edges weights have been squared to be able to draw the edges thickness in a meaningful way. The algorith is run in only one of our cores for our server limitations.

GENIE3 implementation

Backend:

PC algorithm parameters

Backend:

Grow-Shrink algorithm parameters

Backend:

Incremental Association algorithm parameters

Backend:

Fast Incremental Association algorithm parameters

Backend:

Interleaved Incremental Association algorithm parameters

Backend:

Hill climbing algorithm parameters

Backend:

Hill climbing with tabu search algorithm parameters

Backend:

Chow-Liu tree algorithm parameters

Backend:

Hiton Parents and Children algorithm parameters

Backend:

sparseBn algorithm parameters

Backend:

FGES-Merge algorithm parameters

Bernaola et al (2019). Details can be found here.

Mode:

Backend:

Max-Min Hill-Climbing algorithm parameters

Backend:

Max-Min Parents and Children algorithm parameters

Backend:

Naive Bayes algorithm parameters

Backend:

Tree Augmented Naive Bayes algorithm parameters

Backend:

Multi-dimensional Bayesian network classifier algorithm parameters

Backend:

Learning structure...

Estimated time to finish:

Structure learned
Check the graph below.

3. Learn the parameters of the Bayesian network

Select the parameter learning algorithm:

Learning algorithm:

Maximum likelihood estimation (MLE) of a Gaussian distribution

Backend:

Discrete maximum likelihood estimation algorithm parameters

Backend:

Discrete Bayesian estimation algorithm parameters

Backend:

Learning parameters...

Estimated time to finish:

Parameters learned
Click on the nodes to check them.

Visualize Bayesian network graph

color_lens Green nodes are class features, blue nodes are predictor features, red nodes have fixed evidence.

Double click to reset the colors. When a node is clicked, its name is copied into the clipboard.

zoom_in

Find one node:

filter_list

Select multiple nodes:

info

Check d-separation:

group_work

List of groups:

Category:

compare

Select structure:

equalizer

Nodes by evidences effect:

KL divergence:

Mean:

Standard deviation:

bubble_chart

Groups by evidences effect (KL divergence):

Category:

Filter edges by weight:

Loading


Evidence:

Loading parameters

nodes ( selected) edges ( selected)

Structures metrics

Export the Bayesian network

Export Bayesian network as:

Loading