 Ver en TensorFlow.org |  Ejecutar en Google Colab |  Ver fuente en GitHub |  Descargar libreta |
El ruido está presente en las computadoras cuánticas modernas. Los qubits son susceptibles a la interferencia del entorno circundante, la fabricación imperfecta, TLS y, a veces, incluso los rayos gamma . Hasta que se alcance la corrección de errores a gran escala, los algoritmos de hoy deben poder permanecer funcionales en presencia de ruido. Esto hace que probar algoritmos bajo ruido sea un paso importante para validar que los algoritmos/modelos cuánticos funcionarán en las computadoras cuánticas de hoy.
En este tutorial, explorará los conceptos básicos de la simulación de circuitos ruidosos en TFQ a través de la API tfq.layers de alto nivel.
Configuración
pip install tensorflow==2.7.0 tensorflow-quantum
pip install -q git+https://github.com/tensorflow/docs
# Update package resources to account for version changes.
import importlib, pkg_resources
importlib.reload(pkg_resources)
<module 'pkg_resources' from '/tmpfs/src/tf_docs_env/lib/python3.7/site-packages/pkg_resources/__init__.py'>
import random
import cirq
import sympy
import tensorflow_quantum as tfq
import tensorflow as tf
import numpy as np
# Plotting
import matplotlib.pyplot as plt
import tensorflow_docs as tfdocs
import tensorflow_docs.plots
2022-02-04 12:35:30.853880: E tensorflow/stream_executor/cuda/cuda_driver.cc:271] failed call to cuInit: CUDA_ERROR_NO_DEVICE: no CUDA-capable device is detected
1. Comprender el ruido cuántico
1.1 Ruido de circuito básico
El ruido en una computadora cuántica afecta las muestras de cadenas de bits que puede medir a partir de ella. Una forma intuitiva de comenzar a pensar en esto es que una computadora cuántica ruidosa "insertará", "eliminará" o "reemplazará" puertas en lugares aleatorios como el diagrama a continuación:
A partir de esta intuición, cuando se trata de ruido, ya no se utiliza un único \(|\psi \rangle\) estado puro, sino que se trata de un conjunto de todas las realizaciones ruidosas posibles del circuito deseado: \(\rho = \sum_j p_j |\psi_j \rangle \langle \psi_j |\) . Donde \(p_j\) da la probabilidad de que el sistema esté en \(|\psi_j \rangle\) .
Revisando la imagen de arriba, si supiéramos de antemano que el 90% de las veces nuestro sistema se ejecutó perfectamente, o tuvo errores el 10% de las veces con solo este modo de falla, entonces nuestro conjunto sería:
\(\rho = 0.9 |\psi_\text{desired} \rangle \langle \psi_\text{desired}| + 0.1 |\psi_\text{noisy} \rangle \langle \psi_\text{noisy}| \)
Si hubiera más de una forma en que nuestro circuito pudiera fallar, entonces el conjunto \(\rho\) contendría más de dos términos (uno para cada nueva realización ruidosa que podría ocurrir). \(\rho\) se conoce como la matriz de densidad que describe su sistema ruidoso.
1.2 Uso de canales para modelar el ruido del circuito
Desafortunadamente, en la práctica, es casi imposible conocer todas las formas en que su circuito podría fallar y sus probabilidades exactas. Una suposición simplificadora que puede hacer es que después de cada operación en su circuito hay algún tipo de canal que captura aproximadamente cómo podría fallar esa operación. Puede crear rápidamente un circuito con algo de ruido:
def x_circuit(qubits):
"""Produces an X wall circuit on `qubits`."""
return cirq.Circuit(cirq.X.on_each(*qubits))
def make_noisy(circuit, p):
"""Add a depolarization channel to all qubits in `circuit` before measurement."""
return circuit + cirq.Circuit(cirq.depolarize(p).on_each(*circuit.all_qubits()))
my_qubits = cirq.GridQubit.rect(1, 2)
my_circuit = x_circuit(my_qubits)
my_noisy_circuit = make_noisy(my_circuit, 0.5)
my_circuit
my_noisy_circuit
Puede examinar la matriz de densidad silenciosa \(\rho\) con:
rho = cirq.final_density_matrix(my_circuit)
np.round(rho, 3)
array([[0.+0.j, 0.+0.j, 0.+0.j, 0.+0.j],
[0.+0.j, 0.+0.j, 0.+0.j, 0.+0.j],
[0.+0.j, 0.+0.j, 0.+0.j, 0.+0.j],
[0.+0.j, 0.+0.j, 0.+0.j, 1.+0.j]], dtype=complex64)
Y la matriz de densidad ruidosa \(\rho\) con:
rho = cirq.final_density_matrix(my_noisy_circuit)
np.round(rho, 3)
array([[0.111+0.j, 0. +0.j, 0. +0.j, 0. +0.j],
[0. +0.j, 0.222+0.j, 0. +0.j, 0. +0.j],
[0. +0.j, 0. +0.j, 0.222+0.j, 0. +0.j],
[0. +0.j, 0. +0.j, 0. +0.j, 0.444+0.j]], dtype=complex64)
Al comparar los dos \( \rho \) diferentes, puede ver que el ruido ha afectado las amplitudes del estado (y, en consecuencia, las probabilidades de muestreo). En el caso sin ruido, siempre esperaría muestrear el estado \( |11\rangle \) . Pero en el estado ruidoso ahora hay una probabilidad distinta de cero de muestrear \( |00\rangle \) o \( |01\rangle \) o \( |10\rangle \) también:
"""Sample from my_noisy_circuit."""
def plot_samples(circuit):
samples = cirq.sample(circuit + cirq.measure(*circuit.all_qubits(), key='bits'), repetitions=1000)
freqs, _ = np.histogram(samples.data['bits'], bins=[i+0.01 for i in range(-1,2** len(my_qubits))])
plt.figure(figsize=(10,5))
plt.title('Noisy Circuit Sampling')
plt.xlabel('Bitstring')
plt.ylabel('Frequency')
plt.bar([i for i in range(2** len(my_qubits))], freqs, tick_label=['00','01','10','11'])
plot_samples(my_noisy_circuit)
Sin ningún ruido, siempre obtendrá \(|11\rangle\):
"""Sample from my_circuit."""
plot_samples(my_circuit)
Si aumenta el ruido un poco más, será cada vez más difícil distinguir el comportamiento deseado (muestreo \(|11\rangle\) ) del ruido:
my_really_noisy_circuit = make_noisy(my_circuit, 0.75)
plot_samples(my_really_noisy_circuit)
2. Ruido básico en TFQ
Con esta comprensión de cómo el ruido puede afectar la ejecución del circuito, puede explorar cómo funciona el ruido en TFQ. TensorFlow Quantum utiliza simulación basada en monte-carlo/trayectoria como alternativa a la simulación de matriz de densidad. Esto se debe a que la complejidad de la memoria de la simulación de matriz de densidad limita las simulaciones grandes a <= 20 qubits con los métodos tradicionales de simulación de matriz de densidad completa. Monte-carlo/trayectoria intercambia este costo en memoria por un costo adicional en el tiempo. La opción backend='noisy' está disponible para todos los tfq.layers.Sample , tfq.layers.SampledExpectation y tfq.layers.Expectation (en el caso de Expectation , esto agrega un parámetro de repetitions requerido).
2.1 Muestreo ruidoso en TFQ
Para recrear los gráficos anteriores usando TFQ y simulación de trayectoria, puede usar tfq.layers.Sample
"""Draw bitstring samples from `my_noisy_circuit`"""
bitstrings = tfq.layers.Sample(backend='noisy')(my_noisy_circuit, repetitions=1000)
numeric_values = np.einsum('ijk,k->ij', bitstrings.to_tensor().numpy(), [1, 2])[0]
freqs, _ = np.histogram(numeric_values, bins=[i+0.01 for i in range(-1,2** len(my_qubits))])
plt.figure(figsize=(10,5))
plt.title('Noisy Circuit Sampling')
plt.xlabel('Bitstring')
plt.ylabel('Frequency')
plt.bar([i for i in range(2** len(my_qubits))], freqs, tick_label=['00','01','10','11'])
<BarContainer object of 4 artists>
2.2 Expectativa basada en muestras ruidosas
Para hacer un cálculo de expectativa basado en muestras ruidosas, puede usar tfq.layers.SampleExpectation :
some_observables = [cirq.X(my_qubits[0]), cirq.Z(my_qubits[0]), 3.0 * cirq.Y(my_qubits[1]) + 1]
some_observables
[cirq.X(cirq.GridQubit(0, 0)),
cirq.Z(cirq.GridQubit(0, 0)),
cirq.PauliSum(cirq.LinearDict({frozenset({(cirq.GridQubit(0, 1), cirq.Y)}): (3+0j), frozenset(): (1+0j)}))]
Calcule las estimaciones de expectativa sin ruido a través del muestreo del circuito:
noiseless_sampled_expectation = tfq.layers.SampledExpectation(backend='noiseless')(
my_circuit, operators=some_observables, repetitions=10000
)
noiseless_sampled_expectation.numpy()
array([[-0.0028, -1. , 1.0264]], dtype=float32)
Compáralos con las versiones ruidosas:
noisy_sampled_expectation = tfq.layers.SampledExpectation(backend='noisy')(
[my_noisy_circuit, my_really_noisy_circuit], operators=some_observables, repetitions=10000
)
noisy_sampled_expectation.numpy()
array([[ 0.0242 , -0.33200002, 1.0138001 ],
[ 0.0108 , -0.0012 , 0.9502 ]], dtype=float32)
Puede ver que el ruido ha afectado particularmente la precisión \(\langle \psi | Z | \psi \rangle\) , con my_really_noisy_circuit concentrándose muy rápidamente hacia 0.
2.3 Cálculo de expectativa analítica ruidoso
Hacer cálculos de expectativas analíticas ruidosos es casi idéntico al anterior:
noiseless_analytic_expectation = tfq.layers.Expectation(backend='noiseless')(
my_circuit, operators=some_observables
)
noiseless_analytic_expectation.numpy()
array([[ 1.9106853e-15, -1.0000000e+00, 1.0000002e+00]], dtype=float32)
noisy_analytic_expectation = tfq.layers.Expectation(backend='noisy')(
[my_noisy_circuit, my_really_noisy_circuit], operators=some_observables, repetitions=10000
)
noisy_analytic_expectation.numpy()
array([[ 1.9106850e-15, -3.3359998e-01, 1.0000000e+00],
[ 1.9106857e-15, -3.8000005e-03, 1.0000001e+00]], dtype=float32)
3. Modelos híbridos y ruido de datos cuánticos
Ahora que ha implementado algunas simulaciones de circuitos ruidosos en TFQ, puede experimentar cómo el ruido afecta a los modelos clásicos cuánticos e híbridos comparando y contrastando su rendimiento ruidoso frente a silencioso. Una buena primera verificación para ver si un modelo o algoritmo es resistente al ruido es probar bajo un modelo de despolarización de todo el circuito que se parece a esto:
Donde cada segmento de tiempo del circuito (a veces denominado momento) tiene un canal de despolarización adjunto después de cada operación de compuerta en ese segmento de tiempo. El canal de despolarización aplicará uno de \(\{X, Y, Z \}\) con probabilidad \(p\) o no aplicará nada (mantendrá la operación original) con probabilidad \(1-p\).
3.1 Datos
Para este ejemplo, puede usar algunos circuitos preparados en el módulo tfq.datasets como datos de entrenamiento:
qubits = cirq.GridQubit.rect(1, 8)
circuits, labels, pauli_sums, _ = tfq.datasets.xxz_chain(qubits, 'closed')
circuits[0]
Downloading data from https://storage.googleapis.com/download.tensorflow.org/data/quantum/spin_systems/XXZ_chain.zip 184451072/184449737 [==============================] - 2s 0us/step 184459264/184449737 [==============================] - 2s 0us/step
Escribir una pequeña función de ayuda ayudará a generar los datos para el caso ruidoso vs silencioso:
def get_data(qubits, depolarize_p=0.):
"""Return quantum data circuits and labels in `tf.Tensor` form."""
circuits, labels, pauli_sums, _ = tfq.datasets.xxz_chain(qubits, 'closed')
if depolarize_p >= 1e-5:
circuits = [circuit.with_noise(cirq.depolarize(depolarize_p)) for circuit in circuits]
tmp = list(zip(circuits, labels))
random.shuffle(tmp)
circuits_tensor = tfq.convert_to_tensor([x[0] for x in tmp])
labels_tensor = tf.convert_to_tensor([x[1] for x in tmp])
return circuits_tensor, labels_tensor
3.2 Definir un circuito modelo
Ahora que tiene datos cuánticos en forma de circuitos, necesitará un circuito para modelar estos datos, como con los datos, puede escribir una función auxiliar para generar este circuito que opcionalmente contiene ruido:
def modelling_circuit(qubits, depth, depolarize_p=0.):
"""A simple classifier circuit."""
dim = len(qubits)
ret = cirq.Circuit(cirq.H.on_each(*qubits))
for i in range(depth):
# Entangle layer.
ret += cirq.Circuit(cirq.CX(q1, q2) for (q1, q2) in zip(qubits[::2], qubits[1::2]))
ret += cirq.Circuit(cirq.CX(q1, q2) for (q1, q2) in zip(qubits[1::2], qubits[2::2]))
# Learnable rotation layer.
# i_params = sympy.symbols(f'layer-{i}-0:{dim}')
param = sympy.Symbol(f'layer-{i}')
single_qb = cirq.X
if i % 2 == 1:
single_qb = cirq.Y
ret += cirq.Circuit(single_qb(q) ** param for q in qubits)
if depolarize_p >= 1e-5:
ret = ret.with_noise(cirq.depolarize(depolarize_p))
return ret, [op(q) for q in qubits for op in [cirq.X, cirq.Y, cirq.Z]]
modelling_circuit(qubits, 3)[0]
3.3 Construcción de modelos y entrenamiento
Con los datos y el circuito del modelo creados, la función de ayuda final que necesitará es una que pueda ensamblar un tf.keras.Model híbrido cuántico ruidoso o silencioso:
def build_keras_model(qubits, depolarize_p=0.):
"""Prepare a noisy hybrid quantum classical Keras model."""
spin_input = tf.keras.Input(shape=(), dtype=tf.dtypes.string)
circuit_and_readout = modelling_circuit(qubits, 4, depolarize_p)
if depolarize_p >= 1e-5:
quantum_model = tfq.layers.NoisyPQC(*circuit_and_readout, sample_based=False, repetitions=10)(spin_input)
else:
quantum_model = tfq.layers.PQC(*circuit_and_readout)(spin_input)
intermediate = tf.keras.layers.Dense(4, activation='sigmoid')(quantum_model)
post_process = tf.keras.layers.Dense(1)(intermediate)
return tf.keras.Model(inputs=[spin_input], outputs=[post_process])
4. Compara el rendimiento
4.1 Línea de base silenciosa
Con la generación de datos y el código de construcción del modelo, ahora puede comparar y contrastar el rendimiento del modelo en la configuración silenciosa y ruidosa, primero puede ejecutar un entrenamiento silencioso de referencia:
training_histories = dict()
depolarize_p = 0.
n_epochs = 50
phase_classifier = build_keras_model(qubits, depolarize_p)
phase_classifier.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=0.02),
loss=tf.keras.losses.BinaryCrossentropy(from_logits=True),
metrics=['accuracy'])
# Show the keras plot of the model
tf.keras.utils.plot_model(phase_classifier, show_shapes=True, dpi=70)
noiseless_data, noiseless_labels = get_data(qubits, depolarize_p)
training_histories['noiseless'] = phase_classifier.fit(x=noiseless_data,
y=noiseless_labels,
batch_size=16,
epochs=n_epochs,
validation_split=0.15,
verbose=1)
Epoch 1/50 4/4 [==============================] - 1s 133ms/step - loss: 0.7212 - accuracy: 0.4688 - val_loss: 0.6834 - val_accuracy: 0.5000 Epoch 2/50 4/4 [==============================] - 0s 80ms/step - loss: 0.6787 - accuracy: 0.4688 - val_loss: 0.6640 - val_accuracy: 0.5000 Epoch 3/50 4/4 [==============================] - 0s 76ms/step - loss: 0.6637 - accuracy: 0.4688 - val_loss: 0.6529 - val_accuracy: 0.5000 Epoch 4/50 4/4 [==============================] - 0s 78ms/step - loss: 0.6505 - accuracy: 0.4688 - val_loss: 0.6423 - val_accuracy: 0.5000 Epoch 5/50 4/4 [==============================] - 0s 77ms/step - loss: 0.6409 - accuracy: 0.4688 - val_loss: 0.6322 - val_accuracy: 0.5000 Epoch 6/50 4/4 [==============================] - 0s 77ms/step - loss: 0.6300 - accuracy: 0.4844 - val_loss: 0.6187 - val_accuracy: 0.5000 Epoch 7/50 4/4 [==============================] - 0s 77ms/step - loss: 0.6171 - accuracy: 0.5781 - val_loss: 0.6007 - val_accuracy: 0.5000 Epoch 8/50 4/4 [==============================] - 0s 79ms/step - loss: 0.6008 - accuracy: 0.6250 - val_loss: 0.5825 - val_accuracy: 0.5833 Epoch 9/50 4/4 [==============================] - 0s 76ms/step - loss: 0.5864 - accuracy: 0.6406 - val_loss: 0.5610 - val_accuracy: 0.6667 Epoch 10/50 4/4 [==============================] - 0s 77ms/step - loss: 0.5670 - accuracy: 0.6719 - val_loss: 0.5406 - val_accuracy: 0.8333 Epoch 11/50 4/4 [==============================] - 0s 79ms/step - loss: 0.5474 - accuracy: 0.6875 - val_loss: 0.5173 - val_accuracy: 0.9167 Epoch 12/50 4/4 [==============================] - 0s 77ms/step - loss: 0.5276 - accuracy: 0.7188 - val_loss: 0.4941 - val_accuracy: 0.9167 Epoch 13/50 4/4 [==============================] - 0s 75ms/step - loss: 0.5066 - accuracy: 0.7500 - val_loss: 0.4686 - val_accuracy: 0.9167 Epoch 14/50 4/4 [==============================] - 0s 76ms/step - loss: 0.4838 - accuracy: 0.7812 - val_loss: 0.4437 - val_accuracy: 0.9167 Epoch 15/50 4/4 [==============================] - 0s 76ms/step - loss: 0.4618 - accuracy: 0.8281 - val_loss: 0.4182 - val_accuracy: 0.9167 Epoch 16/50 4/4 [==============================] - 0s 76ms/step - loss: 0.4386 - accuracy: 0.8281 - val_loss: 0.3930 - val_accuracy: 1.0000 Epoch 17/50 4/4 [==============================] - 0s 79ms/step - loss: 0.4158 - accuracy: 0.8438 - val_loss: 0.3673 - val_accuracy: 1.0000 Epoch 18/50 4/4 [==============================] - 0s 79ms/step - loss: 0.3944 - accuracy: 0.8438 - val_loss: 0.3429 - val_accuracy: 1.0000 Epoch 19/50 4/4 [==============================] - 0s 77ms/step - loss: 0.3735 - accuracy: 0.8594 - val_loss: 0.3203 - val_accuracy: 1.0000 Epoch 20/50 4/4 [==============================] - 0s 77ms/step - loss: 0.3535 - accuracy: 0.8750 - val_loss: 0.2998 - val_accuracy: 1.0000 Epoch 21/50 4/4 [==============================] - 0s 78ms/step - loss: 0.3345 - accuracy: 0.8906 - val_loss: 0.2815 - val_accuracy: 1.0000 Epoch 22/50 4/4 [==============================] - 0s 76ms/step - loss: 0.3168 - accuracy: 0.8906 - val_loss: 0.2640 - val_accuracy: 1.0000 Epoch 23/50 4/4 [==============================] - 0s 76ms/step - loss: 0.3017 - accuracy: 0.9062 - val_loss: 0.2465 - val_accuracy: 1.0000 Epoch 24/50 4/4 [==============================] - 0s 76ms/step - loss: 0.2840 - accuracy: 0.9219 - val_loss: 0.2328 - val_accuracy: 1.0000 Epoch 25/50 4/4 [==============================] - 0s 76ms/step - loss: 0.2700 - accuracy: 0.9219 - val_loss: 0.2181 - val_accuracy: 1.0000 Epoch 26/50 4/4 [==============================] - 0s 76ms/step - loss: 0.2566 - accuracy: 0.9219 - val_loss: 0.2053 - val_accuracy: 1.0000 Epoch 27/50 4/4 [==============================] - 0s 77ms/step - loss: 0.2445 - accuracy: 0.9375 - val_loss: 0.1935 - val_accuracy: 1.0000 Epoch 28/50 4/4 [==============================] - 0s 76ms/step - loss: 0.2332 - accuracy: 0.9375 - val_loss: 0.1839 - val_accuracy: 1.0000 Epoch 29/50 4/4 [==============================] - 0s 78ms/step - loss: 0.2227 - accuracy: 0.9375 - val_loss: 0.1734 - val_accuracy: 1.0000 Epoch 30/50 4/4 [==============================] - 0s 76ms/step - loss: 0.2145 - accuracy: 0.9375 - val_loss: 0.1630 - val_accuracy: 1.0000 Epoch 31/50 4/4 [==============================] - 0s 76ms/step - loss: 0.2047 - accuracy: 0.9375 - val_loss: 0.1564 - val_accuracy: 1.0000 Epoch 32/50 4/4 [==============================] - 0s 76ms/step - loss: 0.1971 - accuracy: 0.9375 - val_loss: 0.1525 - val_accuracy: 1.0000 Epoch 33/50 4/4 [==============================] - 0s 75ms/step - loss: 0.1894 - accuracy: 0.9531 - val_loss: 0.1464 - val_accuracy: 1.0000 Epoch 34/50 4/4 [==============================] - 0s 74ms/step - loss: 0.1825 - accuracy: 0.9531 - val_loss: 0.1407 - val_accuracy: 1.0000 Epoch 35/50 4/4 [==============================] - 0s 77ms/step - loss: 0.1771 - accuracy: 0.9531 - val_loss: 0.1330 - val_accuracy: 1.0000 Epoch 36/50 4/4 [==============================] - 0s 75ms/step - loss: 0.1704 - accuracy: 0.9531 - val_loss: 0.1288 - val_accuracy: 1.0000 Epoch 37/50 4/4 [==============================] - 0s 76ms/step - loss: 0.1647 - accuracy: 0.9531 - val_loss: 0.1237 - val_accuracy: 1.0000 Epoch 38/50 4/4 [==============================] - 0s 80ms/step - loss: 0.1603 - accuracy: 0.9531 - val_loss: 0.1221 - val_accuracy: 1.0000 Epoch 39/50 4/4 [==============================] - 0s 76ms/step - loss: 0.1551 - accuracy: 0.9688 - val_loss: 0.1177 - val_accuracy: 1.0000 Epoch 40/50 4/4 [==============================] - 0s 75ms/step - loss: 0.1509 - accuracy: 0.9688 - val_loss: 0.1136 - val_accuracy: 1.0000 Epoch 41/50 4/4 [==============================] - 0s 76ms/step - loss: 0.1466 - accuracy: 0.9688 - val_loss: 0.1110 - val_accuracy: 1.0000 Epoch 42/50 4/4 [==============================] - 0s 76ms/step - loss: 0.1426 - accuracy: 0.9688 - val_loss: 0.1083 - val_accuracy: 1.0000 Epoch 43/50 4/4 [==============================] - 0s 75ms/step - loss: 0.1386 - accuracy: 0.9688 - val_loss: 0.1050 - val_accuracy: 1.0000 Epoch 44/50 4/4 [==============================] - 0s 83ms/step - loss: 0.1362 - accuracy: 0.9688 - val_loss: 0.0989 - val_accuracy: 1.0000 Epoch 45/50 4/4 [==============================] - 0s 78ms/step - loss: 0.1324 - accuracy: 0.9688 - val_loss: 0.0978 - val_accuracy: 1.0000 Epoch 46/50 4/4 [==============================] - 0s 77ms/step - loss: 0.1290 - accuracy: 0.9688 - val_loss: 0.0964 - val_accuracy: 1.0000 Epoch 47/50 4/4 [==============================] - 0s 75ms/step - loss: 0.1265 - accuracy: 0.9688 - val_loss: 0.0929 - val_accuracy: 1.0000 Epoch 48/50 4/4 [==============================] - 0s 77ms/step - loss: 0.1234 - accuracy: 0.9688 - val_loss: 0.0923 - val_accuracy: 1.0000 Epoch 49/50 4/4 [==============================] - 0s 77ms/step - loss: 0.1213 - accuracy: 0.9688 - val_loss: 0.0903 - val_accuracy: 1.0000 Epoch 50/50 4/4 [==============================] - 0s 77ms/step - loss: 0.1182 - accuracy: 0.9688 - val_loss: 0.0885 - val_accuracy: 1.0000
Y explore los resultados y la precisión:
loss_plotter = tfdocs.plots.HistoryPlotter(metric = 'loss', smoothing_std=10)
loss_plotter.plot(training_histories)
acc_plotter = tfdocs.plots.HistoryPlotter(metric = 'accuracy', smoothing_std=10)
acc_plotter.plot(training_histories)
4.2 Comparación ruidosa
Ahora puede construir un nuevo modelo con una estructura ruidosa y compararlo con el anterior, el código es casi idéntico:
depolarize_p = 0.001
n_epochs = 50
noisy_phase_classifier = build_keras_model(qubits, depolarize_p)
noisy_phase_classifier.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=0.02),
loss=tf.keras.losses.BinaryCrossentropy(from_logits=True),
metrics=['accuracy'])
# Show the keras plot of the model
tf.keras.utils.plot_model(noisy_phase_classifier, show_shapes=True, dpi=70)
noisy_data, noisy_labels = get_data(qubits, depolarize_p)
training_histories['noisy'] = noisy_phase_classifier.fit(x=noisy_data,
y=noisy_labels,
batch_size=16,
epochs=n_epochs,
validation_split=0.15,
verbose=1)
Epoch 1/50 4/4 [==============================] - 8s 2s/step - loss: 0.8265 - accuracy: 0.4844 - val_loss: 0.8369 - val_accuracy: 0.4167 Epoch 2/50 4/4 [==============================] - 7s 2s/step - loss: 0.7613 - accuracy: 0.4844 - val_loss: 0.7695 - val_accuracy: 0.4167 Epoch 3/50 4/4 [==============================] - 7s 2s/step - loss: 0.7151 - accuracy: 0.4844 - val_loss: 0.7290 - val_accuracy: 0.4167 Epoch 4/50 4/4 [==============================] - 7s 2s/step - loss: 0.6915 - accuracy: 0.4844 - val_loss: 0.7014 - val_accuracy: 0.4167 Epoch 5/50 4/4 [==============================] - 7s 2s/step - loss: 0.6837 - accuracy: 0.4844 - val_loss: 0.6811 - val_accuracy: 0.4167 Epoch 6/50 4/4 [==============================] - 7s 2s/step - loss: 0.6717 - accuracy: 0.4844 - val_loss: 0.6801 - val_accuracy: 0.4167 Epoch 7/50 4/4 [==============================] - 7s 2s/step - loss: 0.6739 - accuracy: 0.4844 - val_loss: 0.6726 - val_accuracy: 0.4167 Epoch 8/50 4/4 [==============================] - 7s 2s/step - loss: 0.6713 - accuracy: 0.4844 - val_loss: 0.6661 - val_accuracy: 0.4167 Epoch 9/50 4/4 [==============================] - 7s 2s/step - loss: 0.6710 - accuracy: 0.4844 - val_loss: 0.6667 - val_accuracy: 0.4167 Epoch 10/50 4/4 [==============================] - 7s 2s/step - loss: 0.6669 - accuracy: 0.4844 - val_loss: 0.6627 - val_accuracy: 0.4167 Epoch 11/50 4/4 [==============================] - 7s 2s/step - loss: 0.6637 - accuracy: 0.4844 - val_loss: 0.6550 - val_accuracy: 0.4167 Epoch 12/50 4/4 [==============================] - 7s 2s/step - loss: 0.6616 - accuracy: 0.4844 - val_loss: 0.6593 - val_accuracy: 0.4167 Epoch 13/50 4/4 [==============================] - 7s 2s/step - loss: 0.6536 - accuracy: 0.4844 - val_loss: 0.6514 - val_accuracy: 0.4167 Epoch 14/50 4/4 [==============================] - 7s 2s/step - loss: 0.6489 - accuracy: 0.4844 - val_loss: 0.6481 - val_accuracy: 0.4167 Epoch 15/50 4/4 [==============================] - 7s 2s/step - loss: 0.6491 - accuracy: 0.4844 - val_loss: 0.6484 - val_accuracy: 0.4167 Epoch 16/50 4/4 [==============================] - 7s 2s/step - loss: 0.6389 - accuracy: 0.4844 - val_loss: 0.6396 - val_accuracy: 0.4167 Epoch 17/50 4/4 [==============================] - 7s 2s/step - loss: 0.6307 - accuracy: 0.4844 - val_loss: 0.6337 - val_accuracy: 0.4167 Epoch 18/50 4/4 [==============================] - 7s 2s/step - loss: 0.6296 - accuracy: 0.4844 - val_loss: 0.6260 - val_accuracy: 0.4167 Epoch 19/50 4/4 [==============================] - 7s 2s/step - loss: 0.6194 - accuracy: 0.4844 - val_loss: 0.6282 - val_accuracy: 0.4167 Epoch 20/50 4/4 [==============================] - 7s 2s/step - loss: 0.6095 - accuracy: 0.4844 - val_loss: 0.6138 - val_accuracy: 0.4167 Epoch 21/50 4/4 [==============================] - 7s 2s/step - loss: 0.6075 - accuracy: 0.4844 - val_loss: 0.5874 - val_accuracy: 0.4167 Epoch 22/50 4/4 [==============================] - 7s 2s/step - loss: 0.5981 - accuracy: 0.4844 - val_loss: 0.5981 - val_accuracy: 0.4167 Epoch 23/50 4/4 [==============================] - 7s 2s/step - loss: 0.5823 - accuracy: 0.4844 - val_loss: 0.5818 - val_accuracy: 0.4167 Epoch 24/50 4/4 [==============================] - 7s 2s/step - loss: 0.5768 - accuracy: 0.4844 - val_loss: 0.5617 - val_accuracy: 0.4167 Epoch 25/50 4/4 [==============================] - 7s 2s/step - loss: 0.5651 - accuracy: 0.4844 - val_loss: 0.5638 - val_accuracy: 0.4167 Epoch 26/50 4/4 [==============================] - 7s 2s/step - loss: 0.5496 - accuracy: 0.4844 - val_loss: 0.5532 - val_accuracy: 0.4167 Epoch 27/50 4/4 [==============================] - 7s 2s/step - loss: 0.5340 - accuracy: 0.5000 - val_loss: 0.5345 - val_accuracy: 0.4167 Epoch 28/50 4/4 [==============================] - 7s 2s/step - loss: 0.5297 - accuracy: 0.5156 - val_loss: 0.5308 - val_accuracy: 0.4167 Epoch 29/50 4/4 [==============================] - 7s 2s/step - loss: 0.5120 - accuracy: 0.5312 - val_loss: 0.5224 - val_accuracy: 0.5000 Epoch 30/50 4/4 [==============================] - 7s 2s/step - loss: 0.4992 - accuracy: 0.5781 - val_loss: 0.4921 - val_accuracy: 0.5833 Epoch 31/50 4/4 [==============================] - 7s 2s/step - loss: 0.4823 - accuracy: 0.5938 - val_loss: 0.4975 - val_accuracy: 0.5000 Epoch 32/50 4/4 [==============================] - 7s 2s/step - loss: 0.5025 - accuracy: 0.5781 - val_loss: 0.4814 - val_accuracy: 0.5000 Epoch 33/50 4/4 [==============================] - 7s 2s/step - loss: 0.4655 - accuracy: 0.6562 - val_loss: 0.4391 - val_accuracy: 0.6667 Epoch 34/50 4/4 [==============================] - 7s 2s/step - loss: 0.4552 - accuracy: 0.7031 - val_loss: 0.4528 - val_accuracy: 0.5833 Epoch 35/50 4/4 [==============================] - 7s 2s/step - loss: 0.4516 - accuracy: 0.6719 - val_loss: 0.3993 - val_accuracy: 0.8333 Epoch 36/50 4/4 [==============================] - 7s 2s/step - loss: 0.4320 - accuracy: 0.7656 - val_loss: 0.4225 - val_accuracy: 0.6667 Epoch 37/50 4/4 [==============================] - 7s 2s/step - loss: 0.4060 - accuracy: 0.7656 - val_loss: 0.4001 - val_accuracy: 0.9167 Epoch 38/50 4/4 [==============================] - 7s 2s/step - loss: 0.3858 - accuracy: 0.7812 - val_loss: 0.4152 - val_accuracy: 0.8333 Epoch 39/50 4/4 [==============================] - 7s 2s/step - loss: 0.3964 - accuracy: 0.7656 - val_loss: 0.3899 - val_accuracy: 0.7500 Epoch 40/50 4/4 [==============================] - 7s 2s/step - loss: 0.3640 - accuracy: 0.8125 - val_loss: 0.3689 - val_accuracy: 0.7500 Epoch 41/50 4/4 [==============================] - 7s 2s/step - loss: 0.3676 - accuracy: 0.7812 - val_loss: 0.3786 - val_accuracy: 0.7500 Epoch 42/50 4/4 [==============================] - 7s 2s/step - loss: 0.3466 - accuracy: 0.8281 - val_loss: 0.3313 - val_accuracy: 0.8333 Epoch 43/50 4/4 [==============================] - 7s 2s/step - loss: 0.3520 - accuracy: 0.8594 - val_loss: 0.3398 - val_accuracy: 0.8333 Epoch 44/50 4/4 [==============================] - 7s 2s/step - loss: 0.3402 - accuracy: 0.8438 - val_loss: 0.3135 - val_accuracy: 0.9167 Epoch 45/50 4/4 [==============================] - 7s 2s/step - loss: 0.3253 - accuracy: 0.8281 - val_loss: 0.3469 - val_accuracy: 0.8333 Epoch 46/50 4/4 [==============================] - 7s 2s/step - loss: 0.3239 - accuracy: 0.8281 - val_loss: 0.3038 - val_accuracy: 0.9167 Epoch 47/50 4/4 [==============================] - 7s 2s/step - loss: 0.2948 - accuracy: 0.8594 - val_loss: 0.3056 - val_accuracy: 0.9167 Epoch 48/50 4/4 [==============================] - 7s 2s/step - loss: 0.2972 - accuracy: 0.9219 - val_loss: 0.2699 - val_accuracy: 0.9167 Epoch 49/50 4/4 [==============================] - 7s 2s/step - loss: 0.3041 - accuracy: 0.8281 - val_loss: 0.2754 - val_accuracy: 0.9167 Epoch 50/50 4/4 [==============================] - 7s 2s/step - loss: 0.2944 - accuracy: 0.8750 - val_loss: 0.2988 - val_accuracy: 0.9167
loss_plotter.plot(training_histories)
acc_plotter.plot(training_histories)
