Training hybrid models using the Pennylane backend¶
In this example, we will first train a pure quantum model using PennyLane and PyTorch to classify whether a sentence is about cooking or computing. We will then train a hybrid model that takes in pairs of sentences and determines whether they are talking about the same or different topics.
[1]:
BATCH_SIZE = 10
EPOCHS = 30
LEARNING_RATE = 0.1
SEED = 2
[2]:
import torch
import random
import numpy as np
torch.manual_seed(SEED)
random.seed(SEED)
np.random.seed(SEED)
Read in the data and create diagrams¶
[3]:
def read_data(filename):
labels, sentences = [], []
with open(filename) as f:
for line in f:
t = float(line[0])
labels.append([t, 1-t])
sentences.append(line[1:].strip())
return labels, sentences
train_labels, train_data = read_data('datasets/mc_train_data.txt')
dev_labels, dev_data = read_data('datasets/mc_dev_data.txt')
test_labels, test_data = read_data('datasets/mc_test_data.txt')
[4]:
from lambeq import BobcatParser
reader = BobcatParser(verbose='text')
raw_train_diagrams = reader.sentences2diagrams(train_data)
raw_dev_diagrams = reader.sentences2diagrams(dev_data)
raw_test_diagrams = reader.sentences2diagrams(test_data)
Tagging sentences.
Parsing tagged sentences.
Turning parse trees to diagrams.
Tagging sentences.
Parsing tagged sentences.
Turning parse trees to diagrams.
Tagging sentences.
Parsing tagged sentences.
Turning parse trees to diagrams.
Remove cups¶
[5]:
from lambeq import RemoveCupsRewriter
remove_cups = RemoveCupsRewriter()
train_diagrams = [remove_cups(diagram) for diagram in raw_train_diagrams]
dev_diagrams = [remove_cups(diagram) for diagram in raw_dev_diagrams]
test_diagrams = [remove_cups(diagram) for diagram in raw_test_diagrams]
train_diagrams[0].draw()
Create circuits¶
[6]:
from lambeq import AtomicType, IQPAnsatz
ansatz = IQPAnsatz({AtomicType.NOUN: 1, AtomicType.SENTENCE: 1},
n_layers=1, n_single_qubit_params=3)
train_circuits = [ansatz(diagram) for diagram in train_diagrams]
dev_circuits = [ansatz(diagram) for diagram in dev_diagrams]
test_circuits = [ansatz(diagram) for diagram in test_diagrams]
train_circuits[0].draw(figsize=(8, 8))
Create (pure quantum) model and initialise parameters¶
[7]:
from lambeq import PennyLaneModel
all_circuits = train_circuits + dev_circuits + test_circuits
model = PennyLaneModel.from_diagrams(all_circuits)
model.initialise_weights()
Prepare train dataset¶
[8]:
from lambeq import Dataset
train_dataset = Dataset(train_circuits,
train_labels,
batch_size=BATCH_SIZE)
val_dataset = Dataset(dev_circuits, dev_labels)
Training¶
Using PytorchTrainer¶
[9]:
def acc(y_hat, y):
return (torch.argmax(y_hat, dim=1) ==
torch.argmax(y, dim=1)).sum().item()/len(y)
def loss(y_hat, y):
return torch.nn.functional.mse_loss(y_hat, y)
[10]:
from lambeq import PytorchTrainer
trainer = PytorchTrainer(
model=model,
loss_function=loss,
optimizer=torch.optim.Adam,
learning_rate=LEARNING_RATE,
epochs=EPOCHS,
evaluate_functions={"acc": acc},
evaluate_on_train=True,
use_tensorboard=False,
verbose='text',
seed=SEED
)
trainer.fit(train_dataset, val_dataset)
Epoch 1: train/loss: 0.1542 valid/loss: 0.2271 train/time: 0.88s valid/time: 0.23s train/acc: 0.5571 valid/acc: 0.5333
Epoch 2: train/loss: 0.1318 valid/loss: 0.2877 train/time: 0.59s valid/time: 0.17s train/acc: 0.8571 valid/acc: 0.6000
Epoch 3: train/loss: 0.0677 valid/loss: 0.1879 train/time: 0.53s valid/time: 0.16s train/acc: 0.8429 valid/acc: 0.7333
Epoch 4: train/loss: 0.1274 valid/loss: 0.1289 train/time: 0.62s valid/time: 0.17s train/acc: 0.9000 valid/acc: 0.8333
Epoch 5: train/loss: 0.0604 valid/loss: 0.1909 train/time: 0.48s valid/time: 0.16s train/acc: 0.8571 valid/acc: 0.6667
Epoch 6: train/loss: 0.0572 valid/loss: 0.1599 train/time: 0.48s valid/time: 0.16s train/acc: 0.8857 valid/acc: 0.7333
Epoch 7: train/loss: 0.0147 valid/loss: 0.1156 train/time: 0.58s valid/time: 0.47s train/acc: 0.9286 valid/acc: 0.8000
Epoch 8: train/loss: 0.0057 valid/loss: 0.0661 train/time: 0.76s valid/time: 0.21s train/acc: 0.8857 valid/acc: 0.9333
Epoch 9: train/loss: 0.0988 valid/loss: 0.1100 train/time: 0.47s valid/time: 0.16s train/acc: 0.9429 valid/acc: 0.8667
Epoch 10: train/loss: 0.0067 valid/loss: 0.0928 train/time: 0.65s valid/time: 0.17s train/acc: 0.9714 valid/acc: 0.8667
Epoch 11: train/loss: 0.0854 valid/loss: 0.0410 train/time: 0.47s valid/time: 0.16s train/acc: 0.9714 valid/acc: 0.9667
Epoch 12: train/loss: 0.0434 valid/loss: 0.0416 train/time: 0.54s valid/time: 0.17s train/acc: 0.9714 valid/acc: 0.9333
Epoch 13: train/loss: 0.0368 valid/loss: 0.0262 train/time: 0.56s valid/time: 0.16s train/acc: 1.0000 valid/acc: 1.0000
Epoch 14: train/loss: 0.0007 valid/loss: 0.0239 train/time: 0.48s valid/time: 0.16s train/acc: 1.0000 valid/acc: 1.0000
Epoch 15: train/loss: 0.0002 valid/loss: 0.0109 train/time: 0.47s valid/time: 0.16s train/acc: 1.0000 valid/acc: 1.0000
Epoch 16: train/loss: 0.0002 valid/loss: 0.0057 train/time: 0.56s valid/time: 0.16s train/acc: 1.0000 valid/acc: 1.0000
Epoch 17: train/loss: 0.0014 valid/loss: 0.0078 train/time: 0.46s valid/time: 0.16s train/acc: 1.0000 valid/acc: 1.0000
Epoch 18: train/loss: 0.0048 valid/loss: 0.0071 train/time: 0.46s valid/time: 0.16s train/acc: 1.0000 valid/acc: 1.0000
Epoch 19: train/loss: 0.0021 valid/loss: 0.0060 train/time: 0.57s valid/time: 0.25s train/acc: 1.0000 valid/acc: 1.0000
Epoch 20: train/loss: 0.0007 valid/loss: 0.0051 train/time: 0.46s valid/time: 0.16s train/acc: 1.0000 valid/acc: 1.0000
Epoch 21: train/loss: 0.0002 valid/loss: 0.0046 train/time: 0.49s valid/time: 0.16s train/acc: 1.0000 valid/acc: 1.0000
Epoch 22: train/loss: 0.0001 valid/loss: 0.0054 train/time: 0.58s valid/time: 0.16s train/acc: 1.0000 valid/acc: 1.0000
Epoch 23: train/loss: 0.0001 valid/loss: 0.0057 train/time: 0.49s valid/time: 0.16s train/acc: 1.0000 valid/acc: 1.0000
Epoch 24: train/loss: 0.0000 valid/loss: 0.0058 train/time: 0.46s valid/time: 0.16s train/acc: 1.0000 valid/acc: 1.0000
Epoch 25: train/loss: 0.0000 valid/loss: 0.0058 train/time: 0.57s valid/time: 0.16s train/acc: 1.0000 valid/acc: 1.0000
Epoch 26: train/loss: 0.0000 valid/loss: 0.0058 train/time: 0.46s valid/time: 0.17s train/acc: 1.0000 valid/acc: 1.0000
Epoch 27: train/loss: 0.0000 valid/loss: 0.0059 train/time: 0.45s valid/time: 0.16s train/acc: 1.0000 valid/acc: 1.0000
Epoch 28: train/loss: 0.0000 valid/loss: 0.0059 train/time: 0.47s valid/time: 0.31s train/acc: 1.0000 valid/acc: 1.0000
Epoch 29: train/loss: 0.0000 valid/loss: 0.0059 train/time: 0.46s valid/time: 0.17s train/acc: 1.0000 valid/acc: 1.0000
Epoch 30: train/loss: 0.0000 valid/loss: 0.0059 train/time: 0.45s valid/time: 0.16s train/acc: 1.0000 valid/acc: 1.0000
Training completed!
train/time: 15.95s train/time_per_epoch: 0.53s train/time_per_step: 0.08s valid/time: 5.52s valid/time_per_eval: 0.18s
Determine test accuracy¶
[11]:
def accuracy(circs, labels):
probs = model(circs)
return (torch.argmax(probs, dim=1) ==
torch.argmax(torch.tensor(labels), dim=1)).sum().item()/len(circs)
accuracy(test_circuits, test_labels)
[11]:
1.0
Using standard PyTorch¶
As we have a small dataset, we can use early stopping to prevent overfitting to the training data.
[12]:
def accuracy(circs, labels):
probs = model(circs)
return (torch.argmax(probs, dim=1) ==
torch.argmax(torch.tensor(labels), dim=1)).sum().item()/len(circs)
[13]:
import pickle
model = PennyLaneModel.from_diagrams(all_circuits)
model.initialise_weights()
optimizer = torch.optim.Adam(model.parameters(), lr=LEARNING_RATE)
best = {'acc': 0, 'epoch': 0}
for i in range(EPOCHS):
epoch_loss = 0
for circuits, labels in train_dataset:
optimizer.zero_grad()
probs = model(circuits)
d_type = model.weights[0].dtype
probs = probs.to(d_type)
loss = torch.nn.functional.mse_loss(probs,
torch.tensor(labels))
epoch_loss += loss.item()
loss.backward()
optimizer.step()
if i % 5 == 0:
dev_acc = accuracy(dev_circuits, dev_labels)
print("Epoch: {}".format(i))
print("Train loss: {}".format(epoch_loss))
print("Dev acc: {}".format(dev_acc))
if dev_acc > best['acc']:
best['acc'] = dev_acc
best['epoch'] = i
model.save("model.lt")
elif i - best['epoch'] >= 10:
print("Early stopping")
break
if best["acc"] > accuracy(dev_circuits, dev_labels):
model.load("model.lt")
Epoch: 0
Train loss: 1.8844525068998337
Dev acc: 0.8
Epoch: 5
Train loss: 0.19278147350996733
Dev acc: 0.9666666666666667
Epoch: 10
Train loss: 0.014470159949269146
Dev acc: 0.9333333333333333
Epoch: 15
Train loss: 0.000635529821011005
Dev acc: 0.9666666666666667
Early stopping
Determine the test accuracy¶
[14]:
accuracy(test_circuits, test_labels)
[14]:
0.9666666666666667
Creating a hybrid model¶
This model will take in pairs of diagrams and attempt to determine whether they are talking about the same or different topics. It does this by first running the circuits to get a probability ouput on the open wire, and then passes this output to a simple neural network. We expect the circuits to learn to output [0, 1] or [1, 0] depending on the topic they are referring to (cooking or computing), and the neural network to learn to XOR these outputs to determine whether the topics are the same (in which case it should ouput 0) or different (in which case it should output 1). PennyLane allows us to train both the circuits and the NN simultaneously using PyTorch autograd.
[15]:
BATCH_SIZE = 50
EPOCHS = 100
LEARNING_RATE = 0.1
SEED = 2
[16]:
torch.manual_seed(SEED)
random.seed(SEED)
np.random.seed(SEED)
As the probability outputs from our circuits are guaranteed to be positive, we transform these outputs x by 2 * (x - 0.5), giving inputs to the neural network in the range [-1, 1]. This helps us to avoid “dying ReLUs”, which could otherwise occur if all the input weights to a given neuron were negative, leading to the gradient of all these weights being 0. (A couple of alternative approaches could also involve initialising all the neural network weights to be positive, or using
LeakyReLU as the activation function).
[17]:
from torch import nn
class XORSentenceModel(PennyLaneModel):
def __init__(self, **kwargs):
PennyLaneModel.__init__(self, **kwargs)
self.xor_net = nn.Sequential(
nn.Linear(4, 10),
nn.ReLU(),
nn.Linear(10, 1),
nn.Sigmoid()
)
def forward(self, diagram_pairs):
a, b = zip(*diagram_pairs)
evaluated_pairs = torch.cat((self.get_diagram_output(a),
self.get_diagram_output(b)),
dim=1)
evaluated_pairs = 2 * (evaluated_pairs - 0.5)
out = self.xor_net(evaluated_pairs)
return out
Make paired dataset¶
[18]:
from itertools import combinations
def make_pair_data(diagrams, labels):
pair_diags = list(combinations(diagrams, 2))
pair_labels = [int(x[0] == y[0]) for x, y in combinations(labels, 2)]
return pair_diags, pair_labels
train_pair_circuits, train_pair_labels = make_pair_data(train_circuits,
train_labels)
dev_pair_circuits, dev_pair_labels = make_pair_data(dev_circuits, dev_labels)
test_pair_circuits, test_pair_labels = make_pair_data(test_circuits,
test_labels)
There are lots of pairs (2415 train pairs), so we’ll sample a subset to make this example train more quickly.
[19]:
TRAIN_SAMPLES, DEV_SAMPLES, TEST_SAMPLES = 300, 200, 200
[20]:
train_pair_circuits, train_pair_labels = (
zip(*random.sample(list(zip(train_pair_circuits, train_pair_labels)),
TRAIN_SAMPLES)))
dev_pair_circuits, dev_pair_labels = (
zip(*random.sample(list(zip(dev_pair_circuits, dev_pair_labels)), DEV_SAMPLES)))
test_pair_circuits, test_pair_labels = (
zip(*random.sample(list(zip(test_pair_circuits, test_pair_labels)), TEST_SAMPLES)))
Initialise the model¶
[21]:
all_pair_circuits = (train_pair_circuits +
dev_pair_circuits +
test_pair_circuits)
a, b = zip(*all_pair_circuits)
model = XORSentenceModel.from_diagrams(a + b)
model.initialise_weights()
train_pair_dataset = Dataset(train_pair_circuits,
train_pair_labels,
batch_size=BATCH_SIZE)
optimizer = torch.optim.Adam(model.parameters(), lr=LEARNING_RATE)
Train the model and log accuracies¶
Only log every five epochs as evaluating is expensive.
[22]:
def accuracy(circs, labels):
predicted = model(circs)
return (torch.round(torch.flatten(predicted)) ==
torch.Tensor(labels)).sum().item()/len(circs)
[23]:
best = {'acc': 0, 'epoch': 0}
for i in range(EPOCHS):
epoch_loss = 0
for circuits, labels in train_pair_dataset:
optimizer.zero_grad()
predicted = model(circuits)
loss = torch.nn.functional.binary_cross_entropy(
torch.flatten(predicted), torch.Tensor(labels))
epoch_loss += loss.item()
loss.backward()
optimizer.step()
if i % 5 == 0:
dev_acc = accuracy(dev_pair_circuits, dev_pair_labels)
print("Epoch: {}".format(i))
print("Train loss: {}".format(epoch_loss))
print("Dev acc: {}".format(dev_acc))
if dev_acc > best['acc']:
best['acc'] = dev_acc
best['epoch'] = i
model.save("xor_model.lt")
elif i - best['epoch'] >= 10:
print("Early stopping")
break
if best["acc"] > accuracy(dev_pair_circuits, dev_pair_labels):
model.load("xor_model.lt")
model = model.double()
Epoch: 0
Train loss: 4.250532507896423
Dev acc: 0.53
Epoch: 5
Train loss: 1.186747632920742
Dev acc: 0.825
Epoch: 10
Train loss: 0.35468656790908426
Dev acc: 0.545
Epoch: 15
Train loss: 0.41043267399072647
Dev acc: 0.875
Epoch: 20
Train loss: 0.0038383470964618027
Dev acc: 0.88
Epoch: 25
Train loss: 0.0011570464266696945
Dev acc: 0.88
Epoch: 30
Train loss: 0.0007703642331762239
Dev acc: 0.88
Early stopping
[24]:
accuracy(test_pair_circuits, test_pair_labels)
[24]:
0.89