Run#

class whobpyt.run.model_fitting.ModelFitting(model: AbstractNeuralModel, cost: AbstractLoss, device=device(type='cpu'))[source]#

This Model_fitting class is able to fit resting state data or evoked potential data for which the input training data is empty or some stimulus to one or more NMM nodes, and the label is an associated empirical neuroimaging recording.

Studies which consider different kinds of input, such as if SC or some other variable is associated with an empirical recording, must use a different fitting class.

Attributes:

model: AbstractNMM: Whole Brain Model to Simulate
cost: AbstractLoss: A particular objective function which the model will be optimized for.
trainingStats: TrainingStats: Information about objective function loss and parameter values over training windows/epochs
lastRec: Recording: The last simulation of fitting(), evaluation(), or simulation()
devicetorch.device: Whether the fitting is to run on CPU or GPU

__init__(model: AbstractNeuralModel, cost: AbstractLoss, device=device(type='cpu'))[source]#

Parameters:

model: AbstractNMM: Whole Brain Model to Simulate
cost: AbstractLoss: A particular objective function which the model will be optimized for.
devicetorch.device: Whether the fitting is to run on CPU or GPU

evaluate(u, empRec: list, TPperWindow: int, base_window_num: int = 0, transient_num: int = 10)[source]#

Parameters:

uint or Tensor: external or stimulus
empRec: list of Recording: This is the ML “Training Labels”
TPperWindow: int: Number of Empirical Time Points per window. model.forward does one window at a time.
base_window_numint: length of num_windows for resting
transient_numint: The number of initial time points to exclude from some metrics
———–

save(filename)[source]#

Parameters:

filename: String: filename to use when saving object to file

train(u, empRecs: list, num_epochs: int, TPperWindow: int, warmupWindow: int = 0, learningrate: float = 0.05, lr_2ndLevel: float = 0.05, lr_scheduler: bool = False)[source]#

Parameters:

u: type: This stimulus is the ML “Training Input”
empRec: list of Recording: This is the ML “Training Labels”
num_epochs: int: the number of times to go through the entire training data set
TPperWindow: int: Number of Empirical Time Points per window. model.forward does one window at a time.
learningrate: float: rate of gradient descent
lr_2ndLevel: float: learning rate for priors of model parameters, and possibly others
lr_scheduler: bool: Whether to use the learning rate scheduler