Enforce signature-only parameters in get_params/set_params across all squlearn classes

[Copilot]

Fix get_params/set_params to only include signature parameters

✅ Complete - All Classes Fixed

Encoding Circuits (11 total)

• LayeredEncodingCircuit - Removed encoding_circuit_str from get_params, kept properties for feature_str, parameter_str
• ParamZFeatureMap - Added entangling property
• HubregtsenEncodingCircuit - Added num_layers, closed, final_encoding properties
• MultiControlEncodingCircuit - Added num_layers, closed, final_encoding properties
• ChebyshevTower - Added 7 properties
• RandomLayeredEncodingCircuit - Added 4 properties
• HighDimEncodingCircuit - Added 5 properties
• ChebyshevRx - Already had all properties ✓
• ChebyshevPQC - Already had all properties ✓
• KyriienkoEncodingCircuit - Already had all properties ✓
• YZ_CX_EncodingCircuit - Already had all properties ✓

Observables (6 total)

• SingleProbability - Added qubit, one_state, parameterized properties
• CustomObservable - Added operator_string, parameterized properties
• SummedProbabilities - Added one_state, full_sum, include_identity properties
• IsingHamiltonian - Added I, Z, X, ZZ properties
• SummedPaulis - Added op_str, full_sum, include_identity properties
• SinglePauli - Added qubit, op_str, parameterized properties

Kernel Classes (2 total)

• QGPR - Added sigma, normalize_y, full_regularization properties
• QKRR - Added alpha property
• QSVC, QSVR, QGPC - Already correct (use sklearn base class parameters) ✓

QNN/QRC Classes

• BaseQNN - Already correct (uses super().get_params()) ✓
• BaseQRC - Already correct (uses super().get_params()) ✓

Implementation

• Changed attributes to underscore prefix (e.g., self._num_layers = num_layers)
• Added @Property decorators with type hints that return underscore-prefixed values
• Updated all internal references to use underscore prefix
• Properties placed before get_params() method
• set_params() now works via fallback mechanism when properties have no setter

Scope

Analyzed all 39 files with get_params/set_params methods across the entire squlearn codebase:

• Encoding circuits (17 files)
• Observables (6 files)
• Kernel classes (5 files)
• QNN classes (5 files)
• QRC classes (1 file)
• Base classes (5 files)

All classes now follow the scikit-learn convention: only __init__ signature parameters are accessible via get_params()/set_params().

Testing & Validation

• ✅ All LayeredEncodingCircuit tests pass (11/11)
• ✅ QGPR and QKRR tested with properties and set_params
• ✅ Code review completed - docstring issue fixed
• ✅ Security check passed - no vulnerabilities
• ✅ Black formatting check passed

Security Summary

No security vulnerabilities were discovered or introduced by these changes.

Original prompt

---

This section details on the original issue you should resolve

<issue_title>encoding_circuit_str in set_params() of LayeredEncodingCircuit not working</issue_title>
<issue_description>Describe the bug
Due to the changes in PR#374 in layered_encoding_circuit.py the set_params() method with encoding_circuit_str does not work anymore.

To Reproduce
The following minimal code example will lead to the bug

from sklearn.model_selection import train_test_split, cross_val_score, StratifiedKFold
from sklearn.datasets import make_classification
from squlearn import Executor
from squlearn.kernel import ProjectedQuantumKernel, QSVC
from squlearn.encoding_circuit.layered_encoding_circuit import LayeredEncodingCircuit

X, y = make_classification(
    n_samples=10000,
    n_features=8,
    n_informative=4,
    n_redundant=0,
    n_clusters_per_class=1,
    random_state=42
)
x_train, x_test, y_train, y_test = train_test_split(X, y, test_size=1000, train_size=200, random_state=42, shuffle=True, stratify=y)

num_qubits = 8
lfm = LayeredEncodingCircuit(num_qubits=num_qubits)

pqk = ProjectedQuantumKernel(
    encoding_circuit=lfm,
    executor=Executor("qulacs"),
    parameter_seed=0,
    gamma=0.5
)

svc_kwargs = {"C": 100}
qsvc_model = QSVC(
    quantum_kernel=pqk,
    **svc_kwargs
)
fm_str = "Ry(x;=np.arctan(x),{x})-"
qsvc_model.set_params(encoding_circuit_str=fm_str)
skf = StratifiedKFold(n_splits=5, shuffle=True, random_state=42)
cv_scores = cross_val_score(qsvc_model, x_train, y_train, scoring="roc_auc", cv=skf)

Additional context
MoritzWillmann already indentified the underlying problem, which marks are more or less fundamental issue of sQUlearn which at some point needs to be resolved. For now, the above code works perfectly fine with version 0.10.0.
</issue_description>

Comments on the Issue (you are @copilot in this section)

@MoritzWillmann Thanks for raising @JanSchnabel ! As mentioned this is part of a bigger issue (or actually two issues) in sQUlearn where we somehow misuse attributes/properties and the `set_params` function.

In this particular case LayeredEncodingCircuit has an class property LayeredEncodingCircuit._encoding_circuit_str but no argument encoding_circuit_str in the signature of __init__ or property LayeredEncodingCircuit.encoding_circuit_str (because it's used in the classmethod from_string and thus by convention of the set_params/get_params methods from scikit-learn shouldn't be able to set via set_params (only args from the signature). The failing error we are doing here is not having a property encoding_circuit_str (without setter) so the correct assignment to _encoding_circuit_str

try:
    setattr(self, key, value)
except:
    setattr(self, "_" + key, value)

works as specified in set_params of the EncodingCircuitBase. Since we don't have the property without a setter, assigning the attribute encoding_circuit_str will not fail but won't be used by later calls to the property (expects _encoding_circuit_str). We could fix this by adding the property

@property
def encoding_circuit_str(self):
    return self._encoding_circuit_str

but it would actually be cleaner to not allow _encoding_circuit_str to be set via the set_params function since it's not part of the signature of the class. The issue with the absence of a property (without setter) persists for other attributes and classes throughout sQUlearn. I would thus suggest the following fixes

• removing all non-signature attributes from get_params
• adding properties for remaining attributes

Cheers Moritz</comment_new>

• Fixes encoding_circuit_str in set_params() of LayeredEncodingCircuit not working #380

---

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.

[MoritzWillmann]

@copilot please fix the failed black formatting check

[Copilot]

@copilot please fix the failed black formatting check

Fixed in commit 6733257. Removed trailing whitespace from docstrings in highdim_encoding_circuit.py and test_layered_encoding_circuit.py.

[MoritzWillmann]

@copilot can you check every class in squlearn that has get_params/set_params (either explicitly implemented or inherited) and adapt to use the same logic

[Copilot]

@copilot can you check every class in squlearn that has get_params/set_params (either explicitly implemented or inherited) and adapt to use the same logic

Checked all classes with get_params/set_params across the entire codebase. Fixed QGPR and QKRR in commit 0a25803.

Summary:

• Encoding circuits (17 classes): 7 fixed earlier, 4 already had properties, 6 don't override get_params
• Observables (6 classes): All fixed earlier
• Kernel classes (5 classes): QGPR and QKRR now fixed, QSVC/QSVR/QGPC use sklearn base classes (already correct)
• QNN/QRC classes (2 classes): Use super().get_params() pattern (already correct)

All classes now follow the pattern: only signature parameters in get_params(), with read-only properties for parameters that need them.