cmr.ConditionalMomentEstimator

Guide & worked example · Source

cmr.ConditionalMomentEstimator(
    model,
    moment_function,
    optimizer=torch.optim.LBFGS,
    maxiter=100,
    lr=1.0,
    dtype=torch.float32,
    device=None,
    **optimizer_kwargs,
)

Base class for neural conditional moment estimators.

Parameters

Name	Type	Description	Default
model	torch.nn.Module	Structural model mapping treatments/features t to predictions.	required
moment_function	callable	Function moment_function(prediction, y) returning an (n, q) moment matrix. For NPIV this is often prediction - y.	required
optimizer	torch optimizer class	Optimizer used to update model.parameters().	torch.optim.LBFGS
maxiter	int	Number of optimizer iterations. For LBFGS this is passed as max_iter.	100
lr	float	Optimizer learning rate.	1.0

Attributes

Name	Description
device
dtype
history
lr
maxiter
model
moment_function
optimizer_class
optimizer_kwargs
params

Methods

Name	Description
fit	Fit the model to the data.
moments	Evaluate sample moments at the current model parameters.
predict	Predict with the fitted structural model.
summary	Print a summary of the model results.
to

fit

cmr.ConditionalMomentEstimator.fit(*args, **kwargs)

Fit the model to the data.

moments

cmr.ConditionalMomentEstimator.moments(t, y)

Evaluate sample moments at the current model parameters.

predict

cmr.ConditionalMomentEstimator.predict(t)

Predict with the fitted structural model.

summary

cmr.ConditionalMomentEstimator.summary()

Print a summary of the model results.

to

cmr.ConditionalMomentEstimator.to(device)