macrostat.models package

Submodules

macrostat.models.model module

Generic model class as a wrapper to specific implementations

class macrostat.models.model.Model(parameters: dict, hyper_parameters: dict, name: str = 'model')

Bases: object

A class representing a macroeconomic model.

This class provides a wrapper for users to write their underlying model behavior while maintaining a uniformly accessible interface. Specifically, the user is expected to adapt the model.simulate() function to their needs, respecting only that the return of that function is a pandas dataframe.

name

Name of the model, such as “model”. Used for file and database names

Type:

str

parameters

Dictionary of all parameters

Type:

dict

hyper_parameters

Dictionary of all hyperparameters

Type:

dict

output

None, or the latest simulation run for given parameters

Type:

pd.DataFrame

Example

A general workflow for a model might look like

>>> model = Model(parameters, hyper_parameters)
>>> output = model.simulate()
>>> model.save()

classmethod load(path=None)

Class method to load a model instance from a pickled file.

Parameters:

• path – path to the targeted Sampler

• optional – path to the targeted Sampler

Notes

Note

This implementation is dependent on your pickling version

save(path=None)

Save the model object as a pickled file

Parameters:

• path – path where the model will be stored. If it is None then the model’s name will be used and the file stored in the working directory.

• optional – path where the model will be stored. If it is None then the model’s name will be used and the file stored in the working directory.

Notes

Note

This implementation is dependent on your pickling version

simulate(*args, **kwargs) → DataFrame

Simulate a model run using the stored parameters

This function is designed to be overwritten by the user’s specific implementation of their model. Note that it is expected for the user to set the ‘’self.output’’ attribute to the output generated.

Returns:

output – Output of the model. Generically it should have a “time”-like index and variables across the columns

Return type:

pd.DataFrame

macrostat.models.torchmodel module

Generic model class for models implemented using PyTorch

class macrostat.models.torchmodel.TorchModel(parameters: dict = None, hyper_parameters: dict = None, name: str = 'torchmodel')

Bases: Model, Module

Generic model class for models implemented using PyTorch

This class provides a wrapper for users to write their underlying model behavior while maintaining a uniformly accessible interface. Specifically, the user is expected to adapt the model.forward() function to their needs. The use of PyTorch allows for automatic differentiation which has computational advantages compared to finite differences.

name

Name of the model, such as “model”. Used for file and database names

Type:

str

parameters

Dictionary of all parameters

Type:

dict

hyper_parameters

Dictionary of all hyperparameters

Type:

dict

output

None, or the latest simulation run for given parameters

Type:

pd.DataFrame

Example

A general workflow for a model might look like

>>> model = Model(parameters, hyper_parameters)
>>> output = model.simulate()
>>> model.save()

diffmax(x1, x2)

Smooth approximation to the minimum B: https://mathoverflow.net/questions/35191/a-differentiable-approximation-to-the-minimum-function

Requires:

self.hyper_parameters[‘max_constant’] as a large number

diffmax_v(x)

Smooth approximation to the maximum for a tensor. See diffmax

Requires:

self.hyper_parameters[‘max_constant’] as a large number

diffmin(x1, x2)

Smooth approximation to the minimum B: https://mathoverflow.net/questions/35191/a-differentiable-approximation-to-the-minimum-function

Requires:

self.hyper_parameters[‘min_constant’] as a large number

diffmin_v(x)

Smooth approximation to the minimum. See diffmin

Requires:

self.hyper_parameters[‘min_constant’] as a large number

diffwhere(condition, x1, x2)

Where condition that is differentiable with respect to the condition.

Requires:

self.hyper_parameters[‘diffwhere’] = True self.hyper_parameters[‘sigmoid_constant’] as a large number

Note: For non-NaN/inf, where(x > eps, z, y) is (x - eps > 0) * (z - y) + y, so we can use the sigmoid function to approximate the where function.

Parameters:

• condition (torch.Tensor) – Condition to be evaluated expressed as x - eps

• x1 (torch.Tensor) – Value to be returned if condition is True

• x2 (torch.Tensor) – Value to be returned if condition is False

forward(*args, **kwargs)

Run the model forward through time, e.g. the loop over timesteps goes here.

This method is called by the simulation method and by the pytorch autograd system. Generally, it shouldn’t be called directly by the user.

initialize_simulation()

Initialize the model’s state variables before running the simulation For instance, if one wants to load in a model state to start from.

simulate() → DataFrame

Simulate a model run using the stored parameters

This function is designed to be overwritten by the user’s specific implementation of their model. Note that it is expected for the user to set the ‘’self.output’’ attribute to the output generated.

The function will run ‘’self.initialize_simulation’’ to set up the model and then run the forward pass of the model. Furthermore, in the pure simulation case, we omit the gradient calculation.

Returns:

output – Output of the model. Generically it should have a “time”-like index and variables across the columns

Return type:

pd.DataFrame

tanhmask(x)

Convert a variable into 0 (x<0) and 1 (x>0)

Module contents