Data for "A Permutation-equivariant Deep Learning Model for Quantum State Characterization"

Data for "A Permutation-equivariant Deep Learning Model for Quantum State Characterization"

The repository contains the data supporting the findings contained in D. Maragnano, M. Liscidini, C. Cusano, "A Permutation-equivariant Deep Learning Model for Quantum State Characterization" (submitted to APL Machine Learning).

Data is stored in compressed NumPy archives.

The data is organized as follows.

2qubit_data

2qubit_features: it contains the input features to train, validate, and test the 2-qubit models.

The names of the files are structured according to the following convention.

features_{a}_2q_noiseless.npz: noiseless features for train, validation, and test.

features_{a}_2q_{b}{c}.npz: noisy features for train, validation, and test, for different noisy channels and different noise strengths.

a ={train, test, valid} for train, test, and validation, respectively.

b = {depol, stateimperf} for noiseless features, features affected by depolarizing or state-error imperfection, respectively.

c = {001, 005, 01} for noise strength equal to 0.01, 0.05, and 0.1, respectively.

Files where a = train contain an array with shape (21600, 16), where each row records the measurement outcomes of a single density matrix.

Files where a = validation, test contain an array with shape (1200, 16), where each row records the measurement outcomes of a single density matrix.

2qubit_predictions: it contains two subdirectories.

predictions_purity_2q: it contains the predictions of the purity of the Linear and PELinear models.

The names of the files are structured according to the following convention.

{m}_predictions_noiseless_purs_2qubit.npz: purity predictions when the model is trained on noiseless features.

{m}_predictions_{n}{c}_purs_2qubit.npz: purity predictions when the model is trained on noisy features.

m = {linear, pelinear}: Linear and PELinear modelm respectively.

n = {noisyd, noisysi}: the model is trained on features affected by depolarizing or state-error imperfection noise, respectively.

c = {001, 005, 01}: noise strength equal to 0.01, 0.05, and 0.1, respectively.

Each file contains an array with shape (1200, 1). Each row represents the purity prediction of a single density matrix.

predictions_tqst_2q: it contains the preditcions of the density matrices of the Linear and PELinear models.

The names of the files are structured as follows:

{m}_predictions_noiseless_qst_2qubit.npz: it contains the predictions of the density matrices when the model is trained on noiseless features.

{m}_predictions_{n}{c}_qst_2qubit.npz: it contains the predictions of the density matrices when the model is trained on noisy features.

m = {linear, pelinear}: Linear and PELinear model, respectively.

n = {noisyd, noisysi}: the model is trained on features affected by depolarizing or state-error imperfection noise, respectively.

c = {001, 005, 01}: noise strength equal to 0.01, 0.05, and 0.1, respectively.

Each file contains an array with shape (1200, 16). Each row represents the prediction of a single density matrix.

Three files named purs_{a}_2q.npz, with a ={train, test, valid}, containing the target purities for train, test, and validation, with shapes (21600,), (1200,), (1200,), respectively. Each element is the target purity of a single density matrix.

Three files named vecs_{a}_2q.npz, with a ={train, test, valid}, containing the target density matrices for train, test, and validation, with shapes (21600, 16), (1200, 16), (1200, 16), respectively. Each row of the array contains the real parameter describing a single density matrix. We provide a custom-made function to convert these arrays into density matrices in the file from_array_to_dms.py.

A file from_array_to_dms.py with a custom-made Python function to convert one-dimensional arrays into a density matrix.

4qubit_data

4qubit_features: it contains the input features to train, validate, and test the 2-qubit models.

The names of the files are structured according to the following convention.

features_{a}_4q_noiseless.npz: noiseless features for train, validation, and test.

features_{a}_4q_{b}{c}.npz: noisy features for train, validation, and test, for different noisy channels and different noise strengths.

a ={train, test, valid} for train, test, and validation, respectively.

b = {depol, stateimperf} for noiseless features, features affected by depolarizing or state-error imperfection, respectively.

c = {001, 005, 01, 02, 04, 06, 08, 1} for noise strength equal to 0.01, 0.05, 0.1, 0.2, 0.4, 0.6, 0.8, 1, respectively.

Files where a = train contain an array with shape (108000, 256), where each row records the measurement outcomes of a single density matrix.

Files where a = validation, test contain an array with shape (6000, 256), where each row records the measurement outcomes of a single density matrix.

4qubit_predictions: it contains two subdirectories.

predictions_purity_4q: it contains the predictions of the purity of the combined model.

The names of the files are structured according to the following convention.

predictions_purs_4qubit_noiseless.npz: purity predictions when the model is trained on noiseless features.

predictions_purs_4qubit_{n}{c}.npz: purity predictions when the model is trained on noisy features.

n = {noisyd, noisysi}: the model is trained on features affected by depolarizing or state-error imperfection noise, respectively.

c = {001, 005, 01, 02, 04, 06, 08, 1} for noise strength equal to 0.01, 0.05, 0.1, 0.2, 0.4, 0.6, 0.8, 1, respectively.

Each file contains an array with shape (6000, 1). Each row represents the purity prediction of a single density matrix.

predictions_tqst_4q: it contains the preditcions of the density matrices of the combined model.

The names of the files are structured as follows.

predictions_noiseless_qst_4qubit.npz: it contains the predictions of the density matrices when the model is trained on noiseless features.

predictions_{n}{c}_qst_4qubit.npz: it contains the predictions of the density matrices when the model is trained on noisy features.

n = {noisyd, noisysi}: the model is trained on features affected by depolarizing or state-error imperfection noise, respectively.

c = {001, 005, 01, 02, 04, 06, 08, 1} for noise strength equal to 0.01, 0.05, 0.1, 0.2, 0.4, 0.6, 0.8, 1, respectively.

Each file contains an array with shape (6000, 256). Each row represents the prediction of a single density matrix.

Three files named purs_{a}_4q.npz, with a ={train, test, valid}, containing the target purities for train, test, and validation, with shapes (108000,), (6000,), (6000,), respectively. Each element is the target purity of a single density matrix.

Three files named vecs_{a}_4q.npz, with a ={train, test, valid}, containing the target density matrices for train, test, and validation, with shapes (108000, 256),

(6000, 256), (6000, 256), respectively. Each row of the array contains the real parameter describing a single density matrix. We provide a custom-made function to

convert these arrays into density matrices in the file *from_array_to_dms.py*.

A file from_array_to_dms.py with a custom-made Python function to convert one-dimensional arrays into a density matrix.