Last modified by teodoramisan on 2026/02/13 10:11
From version 80.1
edited by teodoramisan
on 2026/02/13 10:09
Change comment: There is no comment for this version
To version 72.1
edited by teodoramisan
on 2026/02/12 12:07
Change comment: There is no comment for this version

Summary

Details

Page properties
Content
... ... @@ -8,7 +8,7 @@
8 8  
9 9  == Purpose ==
10 10  
11 -This is a Jupyter Lab extension that offers graphical support for TVB and VBI workflows. It is already available in the EBRAINS Lab and it allows users to configure and execute TVB simulations and VBI inference workflows directly from a GUI, while drastically reducing the complexity of configuring them inside a Jupyter Lab notebook.
11 +This is a Jupyter Lab extension that offers graphical support for TVB and VBI workflows. It is already available in the EBRAINS Lab and it allows users to configure and execute TVB simulations and VBI inference workflows directly from a GUI, while drastically reducing the complexity of configuring them inside a Jupyter Lab notebook. Try edit
12 12  
13 13  {{html}}
14 14  <iframe width="1200" height="450" src="https://www.youtube.com/embed/-cjZOsU6PBg" title="YouTube video player" frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share" allowfullscreen></iframe>
... ... @@ -156,27 +156,6 @@
156 156  
157 157  [[image:timeseries-plotly.png||alt="view3.png" height="593" width="1100"]]
158 158  
159 -
160 -=== 3. Visualize posterior samples with a pairplot (VBI inferenece) ===
161 -
162 -This functionality is available for the **SamplePosterior** component. Right click on the component and select the **Open Viewer** functionality.
163 -
164 -This opens a new JupyterLab tab with an editable notebook. After following the instructions in the notebook, you will generate a pairplot of the posterior samples, showing the distribution of each parameter (uncertainty) and the relationships between parameters.
165 -
166 -*Make sure the workflow has been executed so the viewer has data to load.
167 -
168 -[[image:plot-posteriors.png||height="593" width="1100"]]
169 -
170 -
171 -=== 4. Visualize time series from VBI simulations ===
172 -
173 -This functionality is available for the **SimulationRunner** component. Right click on the component and select the **Open Viewer** functionality.
174 -
175 -This opens a new JupyterLab tab with an editable notebook. After following the instructions in the notebook, you will see a simple time-series plot of the simulated model signal over time.
176 -
177 -*Make sure the workflow has been executed so the viewer has data to load.
178 -[[image:plot-timeseries-vbi.png||height="592" width="1100"]]
179 -
180 180  == ==
181 181  
182 182  == Workflow examples ==
... ... @@ -193,14 +193,8 @@
193 193  
194 194  === 2. Full VBI Inference workflow ===
195 195  
196 -Contains all VBI components required to run an inference workflow, from __prior sampling__ and __simulation__ to __posterior training__ and __posterior sampling__.
175 +Contains all VBI components required to run an inference workflow (e.g., inference configuration, model, simulator, train posterior, sample posterior).
197 197  
198 -The workflow starts with **ConfigInference**, which builds the configuration inputs needed by the workflow. It samples parameter values from the prior distribution to generate __theta__ and prepares the feature-extraction configuration (__cfg__) used later in the pipeline.
199 -
200 -Next, **SimulationRunner** executes the selected **VBI model** for a batch of parameter samples (theta) using the chosen backend (//cpp//, //cupy// or //numba//). It selects the requested output signal from the model result and extracts the summary features defined in cfg, producing the __feature matrix__ used for training.
201 -
202 -The resulting features and parameter samples are then passed to **TrainPosterior**, which standardizes the feature matrix with //StandardScaler //and trains a posterior distribution using an SBI method (for example //SNPE//, //SNLE//, or //SNRE//). In the last step, **SamplePosterior** draws parameter samples from the trained posterior distribution, conditioned on the selected observed feature vector.
203 -
204 204  [[image:vbi_workflow.png||height="590" width="1100"]]
205 205  
206 206  === 3. Configuring model parameters using the PhasePlaneWidget: ===
... ... @@ -223,14 +223,5 @@
223 223  
224 224  A workflow simulation which is run remotely, by submitting a job to an HPC site and getting back the results (TimeSeries object).
225 225  
226 -=== 6. Parallel simulations workflow ===
227 -
228 -This example demonstrates **Parameter Space Exploration** (PSE) by running multiple TVB simulations in parallel for different parameter combinations.
229 -
230 -The workflow uses two nested ForEach components to iterate over coupling and conduction speed sets of values. For each (coupling, conduction_speed) pair, a simulation run is executed with the help of the RunParallelProcess component, which runs the workflow body in separate worker processes (using multiprocessing + dill).
231 -
232 -[[image:parallel-simulations.png||height="590" width="1100"]]
233 -
234 -
235 235  {{html}}<iframe width="1280" height="720" src="https://www.youtube.com/embed/M6rZClFgRrM" title="YouTube video player" frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share" allowfullscreen></iframe>{{/html}}
236 236  
parallel-simulations.png
Author
... ... @@ -1,1 +1,0 @@
1 -XWiki.teodoramisan
Size
... ... @@ -1,1 +1,0 @@
1 -325.1 KB
Content
plot-posteriors.png
Author
... ... @@ -1,1 +1,0 @@
1 -XWiki.teodoramisan
Size
... ... @@ -1,1 +1,0 @@
1 -442.0 KB
Content
plot-timeseries-vbi.png
Author
... ... @@ -1,1 +1,0 @@
1 -XWiki.teodoramisan
Size
... ... @@ -1,1 +1,0 @@
1 -579.5 KB
Content
Public

TVB Widgets