Datasets

Pytorch provides two classes: DataSet and DataLoader. If you're not familiar with them, please review them here

Classiq provides two simple datasets, which are mainly used in our examples. They can be found in classiq/applications/qnn/data_sets.py.

DatasetNot

This dataset is used for training a network to learn the "NOT" operation. More specifically, learning the angle to a parametrized Rx gate (spoiler - the correct answer is pi).

This dataset has 2 items:

1. the data is |00...0>, the label is |11...1>.
2. the data is |11...1>, the label is |00...0>.

The amount of qubits is specified in the constructor (def DatasetNot.__init__(self, n: int, ...)).

Transformers

Additionally, we provide 2 Transforms for this class. (Their PyTorch documentation can be found here), which serve our example.

Note that these transformers are automatically used. Feel free to keep reading about them, though this is not mandatory.

In our example, we wish to A) encode the input state onto the quantum circle B) execute the PQC C) measure the PQC D) post-process the measurement results

Our transformers help steps A and D.

First, state_to_weights transforms the state, into parameters for the encoding Rx gates. This is used for step A In other words, the state |0> is transformed into the angle 0, and the state |1> is transformed into the angle pi.

Second, state_to_label transforms the expected output state into a single float number. More specifically, since we generate this data, we set the expected output state to be a pure state (in the Z basis). Additionally, we define the post-processing to return the probability of measuring |00...0> in the output state. Thus, for the output |00...0>, the post processed output is 1, corresponding to 100%, and for the output |11...1>, the post processed output is 0.

DatasetXor

This dataset is used for training a network to learn the "XOR" operation. This may take, similar to DatasetNot, the amount of qubits in the constructor. The "XOR" operation on more than 2 inputs is defined as "a quantum program that outputs a 1 when the number of 1s at its inputs is odd, and a 0 when the number of incoming 1s is even" (credit: wikipedia)

Usage examples

Using pre-configured DataLoaders

from classiq.applications.qnn.datasets import DATALOADER_NOT

for data, label in DATALOADER_NOT:
    print(f"Training the following data: {data}")
    print(f"with the following labels: {label}")

Using pre-configures Datasets

from classiq.applications.qnn.datasets import DATASET_NOT
from torch.utils.data import DataLoader

DATALOADER_NOT = DataLoader(DATASET_NOT, batch_size=2, shuffle=True)

for data, label in DATALOADER_NOT:
    print(f"Training the following data: {data}")
    print(f"with the following labels: {label}")

Using the DatasetNot class

without using our pre-defined transformers

from classiq.applications.qnn.datasets import DatasetNot
from torch.utils.data import DataLoader

NUM_QUBITS = 1

DATASET_NOT = DatasetNot(NUM_QUBITS)

DATALOADER_NOT = DataLoader(DATASET_NOT, batch_size=2, shuffle=True)

for data, label in DATALOADER_NOT:
    print(f"Training the following data: {data}")
    print(f"with the following labels: {label}")

with our pre-defined transformers

from classiq.applications.qnn.datasets import (
    DatasetNot,
    state_to_weights,
    state_to_label,
)
from torch.utils.data import DataLoader
from torchvision.transforms import Lambda

NUM_QUBITS = 1

DATASET_NOT = DatasetNot(
    1, transform=Lambda(state_to_weights), target_transform=Lambda(state_to_label)
)

DATALOADER_NOT = DataLoader(DATASET_NOT, batch_size=2, shuffle=True)

for data, label in DATALOADER_NOT:
    print(f"Training the following data: {data}")
    print(f"with the following labels: {label}")