Commit 23c0356

committed

Update the ridge_regression format

1 parent e1f22dc commit 23c0356Copy full SHA for 23c0356

File tree

3 files changed

+15

-31

lines changed

sysidentpy
- model_structure_selection
  - forward_regression_orthogonal_least_squares.py
- parameter_estimation
  - estimators.py
  - tests
    - test_estimators.py

3 files changed

+15

-31

lines changed

`‎sysidentpy/model_structure_selection/forward_regression_orthogonal_least_squares.py`

Lines changed: 1 addition & 5 deletions

Original file line number	Diff line number	Diff line change
`@@ -89,15 +89,13 @@ class FROLS(Estimators, BaseMSS):`
`89`	`89`	`eps : float, default=np.finfo(np.float64).eps`
`90`	`90`	`Normalization factor of the normalized filters.`
`91`	`91`	`ridge_param : float, default=np.finfo(np.float64).eps`
`92`		`- Regularization parameter used in ridge regression`
	`92`	`+ Regularization parameter used in ridge regression.`
`93`	`93`	`gama : float, default=0.2`
`94`	`94`	`The leakage factor of the Leaky LMS method.`
`95`	`95`	`weight : float, default=0.02`
`96`	`96`	`Weight factor to control the proportions of the error norms`
`97`	`97`	`and offers an extra degree of freedom within the adaptation`
`98`	`98`	`of the LMS mixed norm method.`
`99`		`- ridge_param : float, default=0.01`
`100`		`- Constant parameter used in ridge regression (K = ridge_param*I).`
`101`	`99`	`model_type: str, default="NARMAX"`
`102`	`100`	`The user can choose "NARMAX", "NAR" and "NFIR" models`
`103`	`101`
`@@ -171,7 +169,6 @@ def __init__(`
`171`	`169`	`ridge_param: np.float64 = np.finfo(np.float64).eps, # default is machine eps`
`172`	`170`	`gama: float = 0.2,`
`173`	`171`	`weight: float = 0.02,`
`174`		`- ridge_param: float = 0.01,`
`175`	`172`	`basis_function: Union[Polynomial, Fourier] = Polynomial(),`
`176`	`173`	`model_type: str = "NARMAX",`
`177`	`174`	`):`
`@@ -200,7 +197,6 @@ def __init__(`
`200`	`197`	`ridge_param=ridge_param, # ridge regression parameter`
`201`	`198`	`gama=gama,`
`202`	`199`	`weight=weight,`
`203`		`- ridge_param=ridge_param,`
`204`	`200`	`basis_function=basis_function,`
`205`	`201`	`)`
`206`	`202`	`self.ensemble = None`

`‎sysidentpy/parameter_estimation/estimators.py`

Lines changed: 14 additions & 10 deletions

Original file line number	Diff line number	Diff line change
`@@ -23,22 +23,20 @@ def __init__(`
`23`	`23`	`offset_covariance=0.2,`
`24`	`24`	`mu=0.01,`
`25`	`25`	`eps=np.finfo(np.float64).eps,`
`26`		`- ridge_param=np.finfo(np.float64).eps,# for regularized ridge regression`
	`26`	`+ ridge_param=np.finfo(np.float64).eps,`
`27`	`27`	`gama=0.2,`
`28`	`28`	`weight=0.02,`
`29`		`- ridge_param: float=0.01,`
`30`	`29`	`basis_function=None,`
`31`	`30`	`):`
`32`	`31`	`self.eps = eps`
`33`		`- self.ridge_param = ridge_param# for regularized ridge regression`
	`32`	`+ self.ridge_param = ridge_param`
`34`	`33`	`self.mu = mu`
`35`	`34`	`self.offset_covariance = offset_covariance`
`36`	`35`	`self.max_lag = max_lag`
`37`	`36`	`self.lam = lam`
`38`	`37`	`self.delta = delta`
`39`	`38`	`self.gama = gama`
`40`	`39`	`self.weight = weight # <0 e <1`
`41`		`- self.ridge_param = ridge_param`
`42`	`40`	`self.xi = None`
`43`	`41`	`self.theta_evolution = None`
`44`	`42`	`self.basis_function = basis_function`
`@@ -121,7 +119,7 @@ def least_squares(self, psi, y):`
`121`	`119`	`theta = np.linalg.lstsq(psi, y, rcond=None)[0]`
`122`	`120`	`return theta`
`123`	`121`
`124`		`- def ridge_regression(self, psi, y):`
	`122`	`+ def ridge_regression_classic(self, psi, y):`
`125`	`123`	`"""Estimate the model parameters using the regularized least squares method`
`126`	`124`	`known as ridge regression. Based on the least_squares module and uses`
`127`	`125`	`the same data format but you need to pass ridge_param in the call to`
`@@ -970,10 +968,16 @@ def ridge_regression(self, psi, y):`
`970`	`968`
`971`	`969`	`y = y[self.max_lag :, 0].reshape(-1, 1)`
`972`	`970`
`973`		`- U, d, Vt = np.linalg.svd(psi, full_matrices=False)`
`974`		`- D = np.diag(d)`
`975`		`- I = np.identity(len(D))`
`976`		`-`
`977`		`- theta = Vt.T @ np.linalg.inv(D*2 + self.ridge_paramI) @ D @ U.T @ y`
	`971`	`+ try:`
	`972`	`+ U, d, Vt = np.linalg.svd(psi, full_matrices=False)`
	`973`	`+ D = np.diag(d)`
	`974`	`+ I = np.identity(len(D))`
	`975`	`+`
	`976`	`+ theta = Vt.T @ np.linalg.inv(D*2 + self.ridge_paramI) @ D @ U.T @ y`
	`977`	`+ except:`
	`978`	`+ warnings.warn("The SVD computation does not converge. Value calculated with the classic algorithm",`
	`979`	`+ stacklevel=2)`
	`980`	`+`
	`981`	`+ theta = self.ridge_regression_classic(psi, y)`
`978`	`982`
`979`	`983`	`return theta`

`‎sysidentpy/parameter_estimation/tests/test_estimators.py`

Lines changed: 0 additions & 16 deletions

Original file line number	Diff line number	Diff line change
`@@ -259,21 +259,5 @@ def test_least_mean_squares_mixed_norm():`
`259`	`259`	`assert_almost_equal(model.theta, theta, decimal=2)`
`260`	`260`
`261`	`261`
`262`		`-def test_ridge_regression():`
`263`		`- x, y, theta = create_test_data()`
`264`		`- basis_function = Polynomial(degree=2)`
`265`		`- model = FROLS(`
`266`		`- n_terms=5,`
`267`		`- extended_least_squares=False,`
`268`		`- ylag=[1, 2],`
`269`		`- xlag=2,`
`270`		`- estimator="ridge_regression",`
`271`		`- ridge_param=0.01,`
`272`		`- basis_function=basis_function,`
`273`		`- )`
`274`		`- model.fit(X=x, y=y)`
`275`		`- assert_almost_equal(model.theta, theta, decimal=2)`
`276`		`-`
`277`		`-`
`278`	`262`	`def test_model_order_selection():`
`279`	`263`	`assert_raises(ValueError, Estimators, max_lag=-1)`

0 commit comments

Comments

(0)

Navigation Menu

Search code, repositories, users, issues, pull requests...

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Commit 23c0356

File tree

3 files changed

3 files changed

`‎sysidentpy/model_structure_selection/forward_regression_orthogonal_least_squares.py`

`‎sysidentpy/parameter_estimation/estimators.py`

`‎sysidentpy/parameter_estimation/tests/test_estimators.py`

0 commit comments