运行环境
- pytorch=1.13.1
- pennylane=0.23.0
# install the latest released version of PennyLane
pip install pennylane --upgrade
# Visit https://pytorch.org/get-started/locally/ for instructions on installing PyTorch.
文件结构目录
|
——models
| ____qmodel.py
|
——run.py
创建文件夹models，并进入文件夹创建qmodel.py，文件内容如下
qmodel.py
import torch
import torch.nn as nn
import pennylane as qml
n_qubits = 4 # Number of qubits
dev = qml.device("default.qubit", wires=n_qubits)
#device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
device = torch.device("cpu")
q_depth = 1
q_delta = 0.01
#定义量子网络
def H_layer(nqubits):
"""Layer of single-qubit Hadamard gates.
"""
for idx in range(nqubits):
qml.Hadamard(wires=idx)
def RY_layer(w):
"""Layer of parametrized qubit rotations around the y axis.
"""
for idx, element in enumerate(w):
qml.RY(element, wires=idx)
def entangling_layer(nqubits):
"""Layer of CNOTs followed by another shifted layer of CNOT.
"""
# In other words it should apply something like :
# CNOT CNOT CNOT CNOT... CNOT
# CNOT CNOT CNOT... CNOT
for i in range(0, nqubits - 1, 2): # Loop over even indices: i=0,2,...N-2
qml.CNOT(wires=[i, i + 1])
for i in range(1, nqubits - 1, 2): # Loop over odd indices: i=1,3,...N-3
qml.CNOT(wires=[i, i + 1])
@qml.qnode(dev, interface="torch")
def quantum_net(q_input_features, q_weights_flat):
"""
The variational quantum circuit.
"""
# Reshape weights
q_weights = q_weights_flat.reshape(q_depth, n_qubits)
# Start from state |+> , unbiased w.r.t. |0> and |1>
H_layer(n_qubits)
# Embed features in the quantum node
RY_layer(q_input_features)
# Sequence of trainable variational layers
for k in range(q_depth):
entangling_layer(n_qubits)
RY_layer(q_weights[k])
# Expectation values in the Z basis
# exp_vals = [qml.expval(qml.PauliZ(position)) for position in range(n_qubits)]
exp_vals = [qml.expval(qml.PauliZ(position)) for position in range(n_qubits)]
return exp_vals
class DressedQuantumNet(nn.Module):
"""
Torch module implementing the *dressed* quantum net.
"""
def __init__(self):
"""
Definition of the *dressed* layout.
"""
super().__init__()
self.flat_net = nn.Flatten()
self.pre_net = nn.Linear(784, n_qubits)
self.q_params = nn.Parameter(q_delta * torch.randn(q_depth * n_qubits))
self.post_net = nn.Linear(n_qubits, 10)
def forward(self, input_features):
"""
Defining how tensors are supposed to move through the *dressed* quantum
net.
"""
# obtain the input features for the quantum circuit
# by reducing the feature dimension from 784 to 4
input_features = self.flat_net(input_features)
pre_out = self.pre_net(input_features)
pre_out = torch.arctan(pre_out)
q_out = torch.Tensor(0, n_qubits)
q_out = q_out.to(device)
for elem in pre_out:
q_out_elem = torch.Tensor(quantum_net(elem, self.q_params)).float().unsqueeze(0)
q_out = torch.cat((q_out, q_out_elem))
return self.post_net(q_out)
run.py
import time
import os
# PyTorch
import torch
import torch.nn as nn
import torch.optim as optim
import torchvision
from torchvision import transforms
from torch.utils.data import DataLoader
# Pennylane
import pennylane as qml
from pennylane import numpy as np
torch.manual_seed(42)
np.random.seed(42)
os.environ["OMP_NUM_THREADS"] = "1"
n_qubits = 4 # Number of qubits
step = 0.003 # Learning rate
batch_size = 32 # Number of samples for each training step
num_epochs = 50 # Number of training epochs
q_depth = 3 # Depth of the quantum circuit (number of variational layers)
gamma_lr_scheduler = 0.1 # Learning rate reduction applied every 10 epochs.
q_delta = 0.01 # Initial spread of random quantum weights
start_time = time.time() # Start of the computation timer
dev = qml.device("default.qubit", wires=n_qubits)
#device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
device = torch.device("cpu")
#load Mnist dataset
# 载入训练集
train_dataset = torchvision.datasets.MNIST(
root = 'dataset/',
train = True,
transform = transforms.ToTensor(),
download=True
)
# 生成测试集
test_dataset = torchvision.datasets.MNIST(
root = 'dataset/',
train = False,
transform = transforms.ToTensor(),
download=True
)
train_data_size = len(train_dataset)
valid_data_size = len(test_dataset)
train_dataloader = DataLoader(dataset=train_dataset,batch_size=batch_size,shuffle=True)
test_dataloader = DataLoader(dataset=test_dataset,batch_size=batch_size,shuffle=True)
from models.qmodel import DressedQuantumNet
model= DressedQuantumNet()
criterion = nn.CrossEntropyLoss()
optimizer= optim.Adam(model.parameters(), lr=step)
def train_model(model, criterion, optimizer, num_epochs):
print("Training started:")
best_acc = 0
for epoch in range(num_epochs):
epoch_start = time.time()
print("Epoch: {}/{}".format(epoch + 1, num_epochs))
model.train()
train_loss = 0.0
train_acc = 0.0
valid_loss = 0.0
valid_acc = 0.0
for i,(inputs, labels) in enumerate(train_dataloader):
inputs = inputs.to(device)
labels = labels.to(device)
optimizer.zero_grad()
outputs = model(inputs)
loss = criterion(outputs, labels)
loss.backward()
optimizer.step()
train_loss += loss.item() * inputs.size(0)
ret, predictions = torch.max(outputs.data, 1)
# print(ret)
correct_counts = predictions.eq(labels.data.view_as(predictions))
acc = torch.mean(correct_counts.type(torch.FloatTensor))
train_acc += acc.item() * inputs.size(0)
with torch.no_grad():
model.eval()
y_true=[]
y_pred=[]
for j, (inputs, labels) in enumerate(test_dataloader):
inputs = inputs.to(device)
labels = labels.to(device)
outputs = model(inputs)
loss = criterion(outputs, labels)
valid_loss += loss.item() * inputs.size(0)
ret, predictions = torch.max(outputs.data, 1)
correct_counts = predictions.eq(labels.data.view_as(predictions))
acc = torch.mean(correct_counts.type(torch.FloatTensor))
valid_acc += acc.item() * inputs.size(0)
y_true.extend(labels.cpu().numpy().tolist())
y_pred.extend(predictions.cpu().numpy().tolist())
avg_train_loss = train_loss / train_data_size
avg_train_acc = train_acc / train_data_size
avg_valid_loss = valid_loss / valid_data_size
avg_valid_acc = valid_acc / valid_data_size
if best_acc < avg_valid_acc:
best_acc = avg_valid_acc
best_epoch = epoch + 1
epoch_end = time.time()
print(
"\t Training: Loss: {:.4f}, Accuracy: {:.4f}%, "
"\n\t Validation: Loss: {:.4f}, Accuracy: {:.4f}%, Time: {:.3f}s".format(
avg_train_loss, avg_train_acc * 100, avg_valid_loss, avg_valid_acc * 100,
epoch_end - epoch_start
))
print("\t Best Accuracy for validation : {:.4f} at epoch {:03d}".format(best_acc, best_epoch))
train_model(model, criterion, optimizer, num_epochs)
训练结果，相比于纯经典的网络（94.49%）还是差一点
Training started:
Epoch: 1/50
Training: Loss: 1.1719, Accuracy: 55.0467%,
Validation: Loss: 0.9824, Accuracy: 62.8900%, Time: 612.915s
Best Accuracy for validation : 0.6289 at epoch 001
Epoch: 2/50
Training: Loss: 0.8633, Accuracy: 70.1850%,
Validation: Loss: 0.7958, Accuracy: 73.6900%, Time: 609.395s
Best Accuracy for validation : 0.7369 at epoch 002
Epoch: 3/50
Training: Loss: 0.7467, Accuracy: 76.1867%,
Validation: Loss: 0.7379, Accuracy: 77.0200%, Time: 603.005s
Best Accuracy for validation : 0.7702 at epoch 003
Epoch: 4/50
Training: Loss: 0.6965, Accuracy: 78.6633%,
Validation: Loss: 0.6963, Accuracy: 78.9100%, Time: 607.274s
Best Accuracy for validation : 0.7891 at epoch 004
Epoch: 5/50
Training: Loss: 0.6713, Accuracy: 79.6633%,
Validation: Loss: 0.7107, Accuracy: 78.0700%, Time: 615.727s
Best Accuracy for validation : 0.7891 at epoch 004
Epoch: 6/50
Training: Loss: 0.6517, Accuracy: 80.3983%,
Validation: Loss: 0.6842, Accuracy: 79.3700%, Time: 613.060s
Best Accuracy for validation : 0.7937 at epoch 006
Epoch: 7/50
Training: Loss: 0.6397, Accuracy: 80.5467%,
Validation: Loss: 0.6647, Accuracy: 80.3300%, Time: 609.802s
Best Accuracy for validation : 0.8033 at epoch 007
Epoch: 8/50
Training: Loss: 0.6246, Accuracy: 81.0650%,
Validation: Loss: 0.6638, Accuracy: 80.9200%, Time: 619.219s
Best Accuracy for validation : 0.8092 at epoch 008
Epoch: 9/50
Training: Loss: 0.6183, Accuracy: 81.3633%,
Validation: Loss: 0.6591, Accuracy: 80.8900%, Time: 613.014s
Best Accuracy for validation : 0.8092 at epoch 008
Epoch: 10/50
Training: Loss: 0.6067, Accuracy: 81.9067%,
Validation: Loss: 0.6495, Accuracy: 81.0200%, Time: 609.195s
Best Accuracy for validation : 0.8102 at epoch 010
Epoch: 11/50
Training: Loss: 0.6004, Accuracy: 82.1517%,
Validation: Loss: 0.6396, Accuracy: 81.1100%, Time: 609.755s
Best Accuracy for validation : 0.8111 at epoch 011
Epoch: 12/50
Training: Loss: 0.5946, Accuracy: 82.2650%,
Validation: Loss: 0.6305, Accuracy: 82.2600%, Time: 621.299s
Best Accuracy for validation : 0.8226 at epoch 012
Epoch: 13/50
Training: Loss: 0.5880, Accuracy: 82.7517%,
Validation: Loss: 0.6329, Accuracy: 81.1800%, Time: 609.291s
Best Accuracy for validation : 0.8226 at epoch 012
Epoch: 14/50
Training: Loss: 0.5826, Accuracy: 82.8700%,
Validation: Loss: 0.6086, Accuracy: 82.6100%, Time: 607.807s
Best Accuracy for validation : 0.8261 at epoch 014
Epoch: 15/50
Training: Loss: 0.5814, Accuracy: 82.9350%,
Validation: Loss: 0.6105, Accuracy: 82.3100%, Time: 614.727s
Best Accuracy for validation : 0.8261 at epoch 014
Epoch: 16/50
Training: Loss: 0.5743, Accuracy: 83.2600%,
Validation: Loss: 0.6109, Accuracy: 82.8700%, Time: 609.623s
Best Accuracy for validation : 0.8287 at epoch 016
Epoch: 17/50
Training: Loss: 0.5740, Accuracy: 83.2433%,
Validation: Loss: 0.6122, Accuracy: 82.7100%, Time: 615.076s
Best Accuracy for validation : 0.8287 at epoch 016
Epoch: 18/50
Training: Loss: 0.5691, Accuracy: 83.4367%,
Validation: Loss: 0.6257, Accuracy: 82.3700%, Time: 612.821s
Best Accuracy for validation : 0.8287 at epoch 016
Epoch: 19/50
Training: Loss: 0.5698, Accuracy: 83.5633%,
Validation: Loss: 0.5976, Accuracy: 83.3600%, Time: 612.744s
Best Accuracy for validation : 0.8336 at epoch 019
Epoch: 20/50
Training: Loss: 0.5672, Accuracy: 83.4583%,
Validation: Loss: 0.5923, Accuracy: 83.7300%, Time: 608.514s
Best Accuracy for validation : 0.8373 at epoch 020
Epoch: 21/50
Training: Loss: 0.5664, Accuracy: 83.6250%,
Validation: Loss: 0.5903, Accuracy: 83.4800%, Time: 616.796s
Best Accuracy for validation : 0.8373 at epoch 020
Epoch: 22/50
Training: Loss: 0.5647, Accuracy: 83.7300%,
Validation: Loss: 0.6010, Accuracy: 83.0700%, Time: 620.391s
Best Accuracy for validation : 0.8373 at epoch 020
Epoch: 23/50
Training: Loss: 0.5647, Accuracy: 83.7650%,
Validation: Loss: 0.6070, Accuracy: 83.0500%, Time: 611.572s
Best Accuracy for validation : 0.8373 at epoch 020
Epoch: 24/50
Training: Loss: 0.5643, Accuracy: 83.7450%,
Validation: Loss: 0.5994, Accuracy: 83.0500%, Time: 616.740s
Best Accuracy for validation : 0.8373 at epoch 020
Epoch: 25/50
Training: Loss: 0.5608, Accuracy: 83.8667%,
Validation: Loss: 0.5872, Accuracy: 83.9300%, Time: 574.860s
Best Accuracy for validation : 0.8393 at epoch 025
Epoch: 26/50
Training: Loss: 0.5619, Accuracy: 83.8317%,
Validation: Loss: 0.5995, Accuracy: 83.4800%, Time: 580.807s
Best Accuracy for validation : 0.8393 at epoch 025
Epoch: 27/50
Training: Loss: 0.5597, Accuracy: 83.9667%,
Validation: Loss: 0.5981, Accuracy: 83.9100%, Time: 573.785s
Best Accuracy for validation : 0.8393 at epoch 025
Epoch: 28/50
Training: Loss: 0.5609, Accuracy: 83.9600%,
Validation: Loss: 0.5985, Accuracy: 83.0100%, Time: 581.108s
Best Accuracy for validation : 0.8393 at epoch 025
Epoch: 29/50
Training: Loss: 0.5582, Accuracy: 84.0417%,
Validation: Loss: 0.6095, Accuracy: 83.1200%, Time: 555.763s
Best Accuracy for validation : 0.8393 at epoch 025
Epoch: 30/50
Training: Loss: 0.5573, Accuracy: 84.0017%,
Validation: Loss: 0.6013, Accuracy: 83.2900%, Time: 553.698s
Best Accuracy for validation : 0.8393 at epoch 025
Epoch: 31/50
Training: Loss: 0.5597, Accuracy: 83.9700%,
Validation: Loss: 0.6226, Accuracy: 82.9000%, Time: 549.435s
Best Accuracy for validation : 0.8393 at epoch 025
Epoch: 32/50
Training: Loss: 0.5583, Accuracy: 83.9933%,
Validation: Loss: 0.6041, Accuracy: 83.4200%, Time: 546.375s
Best Accuracy for validation : 0.8393 at epoch 025
Epoch: 33/50
Training: Loss: 0.5558, Accuracy: 84.0100%,
Validation: Loss: 0.6008, Accuracy: 83.5900%, Time: 546.361s
Best Accuracy for validation : 0.8393 at epoch 025
Epoch: 34/50
Training: Loss: 0.5553, Accuracy: 84.0717%,
Validation: Loss: 0.6344, Accuracy: 82.0200%, Time: 545.937s
Best Accuracy for validation : 0.8393 at epoch 025
Epoch: 35/50
Training: Loss: 0.5572, Accuracy: 83.9150%,
Validation: Loss: 0.5921, Accuracy: 83.7000%, Time: 546.308s
Best Accuracy for validation : 0.8393 at epoch 025
Epoch: 36/50
Training: Loss: 0.5554, Accuracy: 84.1750%,
Validation: Loss: 0.5932, Accuracy: 83.5300%, Time: 546.791s
Best Accuracy for validation : 0.8393 at epoch 025
Epoch: 37/50
Training: Loss: 0.5566, Accuracy: 84.0600%,
Validation: Loss: 0.5954, Accuracy: 83.0300%, Time: 546.066s
Best Accuracy for validation : 0.8393 at epoch 025
Epoch: 38/50
Training: Loss: 0.5540, Accuracy: 84.1700%,
Validation: Loss: 0.5945, Accuracy: 83.8800%, Time: 545.714s
Best Accuracy for validation : 0.8393 at epoch 025
Epoch: 39/50
Training: Loss: 0.5549, Accuracy: 84.2417%,
Validation: Loss: 0.6054, Accuracy: 82.8800%, Time: 544.491s
Best Accuracy for validation : 0.8393 at epoch 025
Epoch: 40/50
Training: Loss: 0.5560, Accuracy: 83.9600%,
Validation: Loss: 0.6189, Accuracy: 82.7500%, Time: 541.625s
Best Accuracy for validation : 0.8393 at epoch 025
Epoch: 41/50
Training: Loss: 0.5546, Accuracy: 84.1667%,
Validation: Loss: 0.6133, Accuracy: 83.2700%, Time: 553.133s
Best Accuracy for validation : 0.8393 at epoch 025
Epoch: 42/50
Training: Loss: 0.5548, Accuracy: 84.0717%,
Validation: Loss: 0.5927, Accuracy: 83.4300%, Time: 576.668s
Best Accuracy for validation : 0.8393 at epoch 025
Epoch: 43/50
Training: Loss: 0.5531, Accuracy: 84.1867%,
Validation: Loss: 0.6155, Accuracy: 82.1500%, Time: 615.985s
Best Accuracy for validation : 0.8393 at epoch 025
Epoch: 44/50
Training: Loss: 0.5525, Accuracy: 84.2217%,
Validation: Loss: 0.5902, Accuracy: 83.9600%, Time: 616.727s
Best Accuracy for validation : 0.8396 at epoch 044
Epoch: 45/50
Training: Loss: 0.5525, Accuracy: 84.1167%,
Validation: Loss: 0.5843, Accuracy: 84.0900%, Time: 618.055s
Best Accuracy for validation : 0.8409 at epoch 045
Epoch: 46/50
Training: Loss: 0.5532, Accuracy: 84.2133%,
Validation: Loss: 0.6115, Accuracy: 82.8800%, Time: 616.387s
Best Accuracy for validation : 0.8409 at epoch 045
Epoch: 47/50
Training: Loss: 0.5529, Accuracy: 84.0767%,
Validation: Loss: 0.6046, Accuracy: 83.3400%, Time: 613.826s
Best Accuracy for validation : 0.8409 at epoch 045
Epoch: 48/50
Training: Loss: 0.5507, Accuracy: 84.3400%,
Validation: Loss: 0.6280, Accuracy: 82.3800%, Time: 614.894s
Best Accuracy for validation : 0.8409 at epoch 045
Epoch: 49/50
Training: Loss: 0.5494, Accuracy: 84.2983%,
Validation: Loss: 0.6100, Accuracy: 82.1000%, Time: 594.077s
Best Accuracy for validation : 0.8409 at epoch 045
Epoch: 50/50
Training: Loss: 0.5510, Accuracy: 84.3800%,
Validation: Loss: 0.5850, Accuracy: 84.2000%, Time: 553.280s
Best Accuracy for validation : 0.8420 at epoch 050
附经典训练结果：
Training started:
Phase: train Epoch: 1/50 Loss: 1.7329 Acc: 0.3983
Phase: train Epoch: 2/50 Loss: 1.1598 Acc: 0.5933
Phase: train Epoch: 3/50 Loss: 0.9086 Acc: 0.7049
Phase: train Epoch: 4/50 Loss: 0.7694 Acc: 0.7510
Phase: train Epoch: 5/50 Loss: 0.6658 Acc: 0.7857
Phase: train Epoch: 6/50 Loss: 0.6100 Acc: 0.8001
Phase: train Epoch: 7/50 Loss: 0.5628 Acc: 0.8229
Phase: train Epoch: 8/50 Loss: 0.5416 Acc: 0.8289
Phase: train Epoch: 9/50 Loss: 0.5088 Acc: 0.8348
Phase: train Epoch: 10/50 Loss: 0.4896 Acc: 0.8423
Phase: train Epoch: 11/50 Loss: 0.4722 Acc: 0.8447
Phase: train Epoch: 12/50 Loss: 0.4476 Acc: 0.8502
Phase: train Epoch: 13/50 Loss: 0.4314 Acc: 0.8576
Phase: train Epoch: 14/50 Loss: 0.4223 Acc: 0.8606
Phase: train Epoch: 15/50 Loss: 0.4111 Acc: 0.8636
Phase: train Epoch: 16/50 Loss: 0.3913 Acc: 0.8686
Phase: train Epoch: 17/50 Loss: 0.3753 Acc: 0.8770
Phase: train Epoch: 18/50 Loss: 0.3662 Acc: 0.8800
Phase: train Epoch: 19/50 Loss: 0.3604 Acc: 0.8755
Phase: train Epoch: 20/50 Loss: 0.3464 Acc: 0.8854
Phase: train Epoch: 21/50 Loss: 0.3391 Acc: 0.8844
Phase: train Epoch: 22/50 Loss: 0.3262 Acc: 0.8938
Phase: train Epoch: 23/50 Loss: 0.3265 Acc: 0.8874
Phase: train Epoch: 24/50 Loss: 0.3079 Acc: 0.9058
Phase: train Epoch: 25/50 Loss: 0.3049 Acc: 0.9008
Phase: train Epoch: 26/50 Loss: 0.2941 Acc: 0.9028
Phase: train Epoch: 27/50 Loss: 0.2876 Acc: 0.9082
Phase: train Epoch: 28/50 Loss: 0.2831 Acc: 0.9043
Phase: train Epoch: 29/50 Loss: 0.2684 Acc: 0.9072
Phase: train Epoch: 30/50 Loss: 0.2616 Acc: 0.9152
Phase: train Epoch: 31/50 Loss: 0.2676 Acc: 0.9072
Phase: train Epoch: 32/50 Loss: 0.2544 Acc: 0.9092
Phase: train Epoch: 33/50 Loss: 0.2480 Acc: 0.9147
Phase: train Epoch: 34/50 Loss: 0.2429 Acc: 0.9216
Phase: train Epoch: 35/50 Loss: 0.2351 Acc: 0.9187
Phase: train Epoch: 36/50 Loss: 0.2238 Acc: 0.9231
Phase: train Epoch: 37/50 Loss: 0.2375 Acc: 0.9191
Phase: train Epoch: 38/50 Loss: 0.2153 Acc: 0.9286
Phase: train Epoch: 39/50 Loss: 0.2098 Acc: 0.9276
Phase: train Epoch: 40/50 Loss: 0.2016 Acc: 0.9330
Phase: train Epoch: 41/50 Loss: 0.1988 Acc: 0.9330
Phase: train Epoch: 42/50 Loss: 0.2062 Acc: 0.9271
Phase: train Epoch: 43/50 Loss: 0.1990 Acc: 0.9301
Phase: train Epoch: 44/50 Loss: 0.1907 Acc: 0.9320
Phase: train Epoch: 45/50 Loss: 0.1798 Acc: 0.9375
Phase: train Epoch: 46/50 Loss: 0.1777 Acc: 0.9325
Phase: train Epoch: 47/50 Loss: 0.1770 Acc: 0.9365
Phase: train Epoch: 48/50 Loss: 0.1668 Acc: 0.9390
Phase: train Epoch: 49/50 Loss: 0.1683 Acc: 0.9395
Phase: train Epoch: 50/50 Loss: 0.1582 Acc: 0.9449
1万+
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
