NuclearNormMatrixCompletion

Guide & worked example · Source

NuclearNormMatrixCompletion(
    lambda_L=None,
    lambda_fraction=0.25,
    fit_unit_effects=True,
    fit_time_effects=True,
    maxiter=500,
    effect_iters=2,
    tol=1e-06,
    device=None,
)

Nuclear-norm penalized matrix completion with additive fixed effects.

Solves

min_{L,u,v} |Omega|^{-1} ||P_Omega(Y - L - u1' - 1v')||_F^2
           + lambda_L ||L||_*

by alternating least-squares updates for the additive effects with the SVT projection update from MCPanel. The implementation is dense and pure torch, so it runs on CPU or GPU and is intended as a clean scalable baseline.

Attributes

Name	Description
device
effect_iters
fit_time_effects
fit_unit_effects
history
lambda_L
lambda_fraction
maxiter
params
tol

Methods

Name	Description
fit
predict
summary	Print a summary of the model results.
to	Move all model parameters to specified device.

fit

NuclearNormMatrixCompletion.fit(Y, mask=None)

predict

NuclearNormMatrixCompletion.predict()

summary

NuclearNormMatrixCompletion.summary()

Print a summary of the model results.

to

NuclearNormMatrixCompletion.to(device)

Move all model parameters to specified device.

Parameters

Name	Type	Description	Default
device	torch.device or str	Target device (‘cuda’, ‘cpu’, etc.)	required

Returns

Name	Type	Description
	BaseEstimator	Self for method chaining