MIOSR reproduces the Lorenz equations with target_sparsity, but not group_sparsity.

[scatr]

Hi again,

I've been using the MIOSR formulation recently, and I've noticed that when using target_sparsity to identify the Lorenz equations, it works fine, but when I try to use the group_sparsity option, the results do not agree. Further, if we perform the regression using the target_sparsity on one variable at a time, including the rest of the variables as control inputs, it does reproduce the each of the Lorenz equations individually. Have I used the group_sparsity option incorrectly? I would have expected that fitting each variable individually would produce an equivalent problem to the group_sparsity one.

I've included a short example to reproduce this

import pysindy as ps
from pysindy.utils import lorenz
from pysindy.differentiation import FiniteDifference
import numpy as np
from scipy.integrate import solve_ivp
from pysindy.optimizers import MIOSR

fd = FiniteDifference()

### create the data
integrator_keywords ={}
integrator_keywords["atol"] = 1e-12
integrator_keywords["rtol"] = 1e-12
dt = 0.0001
t_train = np.arange(0, 100, dt)
t_train_span = (t_train[0], t_train[-1])
x0_train = [-8, 8, 27]
x_train = solve_ivp(lorenz, t_train_span,
                    x0_train, t_eval=t_train, **integrator_keywords).y.T
poly_order = 2

#fit using target sparsity
print("Model with target sparsity")
model = ps.SINDy(
    optimizer=MIOSR(target_sparsity=7, alpha = 0.01, normalize_columns=False),
    feature_library=ps.PolynomialLibrary(degree=poly_order), feature_names = ["x","y","z"]
)
xdot = fd._differentiate(x_train, t_train)

model.fit(
    x_train,
    x_dot = xdot, 
    t=t_train,
)

model.print(precision = 10)

#fit using group sparsity
print("Model with group sparsity")
model = ps.SINDy(
    optimizer=MIOSR(group_sparsity = (4,4,4), alpha = 0.01, normalize_columns=False),
    feature_library=ps.PolynomialLibrary(degree=poly_order), feature_names = ["x","y","z"]
)


model.fit(
    x_train,
    x_dot = xdot,
    t=t_train,
)

model.print(precision = 20)


## fit each variable x,y,z one at a time and include the remainder as input variables
square_lib = ps.PolynomialLibrary(2)
generalized_lib = ps.GeneralizedLibrary([square_lib],
                                         inputs_per_library =np.array([[0,1,2]])
                                        )

labels = ["x", "y", "z"]
group_sparse = (2,3,2) 
print("Model fitting each variable separately")
for i in range(3): 
    control_indices = [jj for jj in range(3) if jj!=i]
    total_indices = [i] + control_indices
    lib_labels = [labels[jj] for jj in total_indices]

    opt = MIOSR(target_sparsity=group_sparse[i], alpha=0.01)
    model = ps.SINDy(optimizer=opt, feature_names=lib_labels, feature_library=generalized_lib)
    model.fit(x_train[:,i], x_dot=xdot[:,i], u= x_train[:,control_indices])
    model.print(precision=20)

This gives the output for me as

Model with target sparsity
(x)' = -9.9999977902 x + 9.9999977888 y
(y)' = 27.9999800480 x + -0.9999961070 y + -0.9999994417 x z
(z)' = -2.6666692197 z + 0.9999997068 x y
Model with group sparsity
(x)' = 2.27950419579689000926 y
(y)' = 25.54290423393682374353 x + -0.95131667283365073384 x z
(z)' = -2.66666921967712333696 z + 0.99999970675281002475 x y
Model fitting each variable separately
(x)' = -9.99999779017980650053 x + 9.99999778884162004999 y
(y)' = -0.99999610704016128615 y + 27.99998004798279893635 x + -0.99999944166784260347 x z
(z)' = -2.66666921967712333696 z + 0.99999970675281002475 x y

[scatr]

The issue is because target_sparsity has the default option 5 meaning that the group_sparsity only works if target_sparsity=None is passed as well.

[Jacob-Stevens-Haas]

I agree that the code should raise a ValueError here.