new DCM structure #314

david-hofmann · 2023-12-01T02:56:08Z

New:

integration with the system_from_graph framework
parameters that should be fitted to data can now be flagged with tunable
states that will be used by an observer model (such as the bold signal) need to be flagged as output with a string label such as hemodynamic_observer
removed the necessity to combine neural mass model with hemodynamic response into a superblock.
generalized codebase of spectralDCM such that blocks with arbitrary number of states can be used (before it was limited to the combination of linear neural mass + balloon model)
fixed the use of parameter scoping in some of the neural mass blocks

Added for convolution option

Same format as the four half cosine model

Cleaning up to prepare for more fMRI simulations

DO NOT PR THIS BRANCH! It's very hacky at the moment to get things running quickly. It shouldn't be included until we have some in-depth conversations about how to combine hemodynamic observer blocks with neural mass blocks.

Added for convolution option

Same format as the four half cosine model

Cleaning up to prepare for more fMRI simulations

DO NOT PR THIS BRANCH! It's very hacky at the moment to get things running quickly. It shouldn't be included until we have some in-depth conversations about how to combine hemodynamic observer blocks with neural mass blocks.

Just committing so that I have a backup as I start messing with some other things - this is still NOT ready for merging so needs to remain quarantined. For reference: This includes some modular implementations to extend the fitting procedures to blocks beyond LinearNeuralMass. As it is, there's hard-coded state number issues with the old implementation, and hierarchy issues with the `system_from_graph` implementation. So we'll need to talk through some design choices after the deadline is met for other stuff.

…ased on Haris's System from Graph framework

…lox.jl into bold-extensions

…ted if a float is passed. If a graph weight is a symbolic expression use that directly instead.

@parameters

…ad of adding an @parameters in progress_scope, otherwise scoping is not working properly

… created within blox is automatically tunable. Note that weight parameters in comparison are currently not tunable by default

…now grouped by region

…se that are non-tunable which are also needed when computing the jacobian

…h with the ordering in the balloon model, otherwise we run into issues when computing the gradient

…. Gradient of bold signal is correct, jacobian of hemodynamic response is erroneous

…mic_observers to have the correct indices. However, there is still a very subtle deviation of the results with previous version. Free Energy is 709 instead of 704. Need to figure where that comes from. Should be the last issue!

… instance in Striatum and other subcortical blox.

… instance in Striatum and other subcortical blox. And in compileparameterlist I added an Int alongside the Float64 since some blox pass integers

…meters

…meters is wrong, the correct use is to call compileparameterlist which will then produce symbolics if numbers are passed or directly use the symbol if that is passed instead.

agchesebro and others added 30 commits September 25, 2023 14:44

Four half cosine parameterization

e169929

Added for convolution option

Double gamma added for convolution

519f1da

Same format as the four half cosine model

Citation details

a8a2cbe

Minor testing updates + exports

0c1bb5f

Cleaning up to prepare for more fMRI simulations

Updates to run DCM on the new system_from_graph

f781918

DO NOT PR THIS BRANCH! It's very hacky at the moment to get things running quickly. It shouldn't be included until we have some in-depth conversations about how to combine hemodynamic observer blocks with neural mass blocks.

Four half cosine parameterization

07eff35

Added for convolution option

Double gamma added for convolution

439c14e

Same format as the four half cosine model

Citation details

d382fa0

Minor testing updates + exports

a8d30c0

Cleaning up to prepare for more fMRI simulations

Updates to run DCM on the new system_from_graph

8fe5ec4

DO NOT PR THIS BRANCH! It's very hacky at the moment to get things running quickly. It shouldn't be included until we have some in-depth conversations about how to combine hemodynamic observer blocks with neural mass blocks.

WIP: David's integration into the new system_from_graph structure.

2e5f581

Indexing and finding hemodynamic observer states

8162d26

integrated my changes with first round of Anthony's changes - model b…

3dc36c1

…ased on Haris's System from Graph framework

Merge branch 'bold-extensions' of https://github.com/Neuroblox/Neurob…

9445402

…lox.jl into bold-extensions

changed connections such that new connection parameters are only crea…

9f8be59

…ted if a float is passed. If a graph weight is a symbolic expression use that directly instead.

just removed some white spaces in cortical_blox.jl

d37e923

add the namespace and more importantly use compileparameterlist inste…

57e0270

…ad of adding an @parameters in progress_scope, otherwise scoping is not working properly

add some namespace parameters

5b614cb

add tunable to compileparameterlist, this way every parameter that is…

fe874bd

… created within blox is automatically tunable. Note that weight parameters in comparison are currently not tunable by default

augmented get_hemodynamic_observers such that indices and states are …

6cd8307

…now grouped by region

added function that completes the list of tunable parameters with tho…

7c949ff

…se that are non-tunable which are also needed when computing the jacobian

had to switch ordering of variables in bold function to ensure a matc…

41aca52

…h with the ordering in the balloon model, otherwise we run into issues when computing the gradient

removed the division by 100 of the neuronal signal in the balloon model

d9decb7

WIP: integration almost finished. However, fitting values still wrong…

813b792

…. Gradient of bold signal is correct, jacobian of hemodynamic response is erroneous

merged with master branch

7167526

fixed issue of reduce over empty set when there are no weights as for…

ea55407

… instance in Striatum and other subcortical blox.

fixed issue of reduce over empty set when there are no weights as for…

cb71273

… instance in Striatum and other subcortical blox. And in compileparameterlist I added an Int alongside the Float64 since some blox pass integers

it is important to note that the use of progress_scope with the @para…

c3b23bb

…meters is wrong, the correct use is to call compileparameterlist which will then produce symbolics if numbers are passed or directly use the symbol if that is passed instead.

david-hofmann added 2 commits December 1, 2023 03:32

some cleanup to wrap up

1f06d6f

synced with current master and removed issue with paths

b27cf3f

david-hofmann merged commit f1fe424 into master Dec 5, 2023
1 check passed

david-hofmann deleted the bold-extensions branch December 5, 2023 15:32

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

new DCM structure #314

new DCM structure #314

david-hofmann commented Dec 1, 2023 •

edited

Loading

new DCM structure #314

new DCM structure #314

Conversation

david-hofmann commented Dec 1, 2023 • edited Loading

david-hofmann commented Dec 1, 2023 •

edited

Loading