Keras (3) implementa una red neuronal profunda

Este artículo agregará un modelo de aprendizaje profundo de DNN basado en la sección anterior.

Uno, imprima la información de la versión de la biblioteca de Python utilizada

import matplotlib as mpl
import matplotlib.pyplot as plt
import numpy as np
import sklearn
import pandas as pd
import os
import sys
import time
import tensorflow as tf
from tensorflow import keras

# 1,打印使用的python库的版本信息
print(sys.version_info)
for module in mpl, np, pd, sklearn, tf, keras:
    print(module.__name__, module.__version__)

#---output-----------
sys.version_info(major=3, minor=8, micro=5, releaselevel='final', serial=0)
matplotlib 3.3.2
numpy 1.19.1
pandas 1.0.3
sklearn 0.23.2
tensorflow 2.2.0
tensorflow.keras 2.3.0-tf

En segundo lugar, extraiga el conjunto de datos de imagen: "datos de entrenamiento", "datos de validación", "datos de prueba" de keras.datasets e imprima sus dimensiones.

fashion_mnist = keras.datasets.fashion_mnist
(x_train_all, y_train_all), (x_test, y_test) = fashion_mnist.load_data()
x_valid, x_train = x_train_all[:5000], x_train_all[5000:]
y_valid, y_train = y_train_all[:5000], y_train_all[5000:]

print(x_valid.shape, y_valid.shape)
print(x_train.shape, y_train.shape)
print(x_test.shape, y_test.shape)

#---------output--------------------
(5000, 28, 28) (5000,)
(55000, 28, 28) (55000,)
(10000, 28, 28) (10000,)

3. Estandarizar el procesamiento de datos

# x = (x - u) / std
from sklearn.preprocessing import StandardScaler

scaler = StandardScaler()
# x_train: [None, 28, 28] -> [None, 784]
x_train_scaled = scaler.fit_transform(
    x_train.astype(np.float32).reshape(-1, 1)).reshape(-1, 28, 28)
x_valid_scaled = scaler.transform(
    x_valid.astype(np.float32).reshape(-1, 1)).reshape(-1, 28, 28)
x_test_scaled = scaler.transform(
    x_test.astype(np.float32).reshape(-1, 1)).reshape(-1, 28, 28)

Cuarto, construir un modelo de red neuronal profunda DNN

# tf.keras.models.Sequential()
model = keras.models.Sequential()
model.add(keras.layers.Flatten(input_shape=[28, 28]))
for _ in range(50):
    model.add(keras.layers.Dense(100, activation="relu"))
model.add(keras.layers.Dense(10, activation="softmax"))

model.compile(loss="sparse_categorical_crossentropy",
              optimizer = keras.optimizers.SGD(0.01),
              metrics = ["accuracy"])

Cinco, ver la estructura general del modelo y los parámetros

print(model.summary())

Seis, agregar función de devolución de llamada, modelo de entrenamiento

# Tensorboard, earlystopping, ModelCheckpoint
logdir = './dnn-callbacks'
if not os.path.exists(logdir):
    os.mkdir(logdir)
output_model_file = os.path.join(logdir,"fashion_mnist_model.h5")

callbacks = [
    keras.callbacks.TensorBoard(logdir),
    keras.callbacks.ModelCheckpoint(output_model_file,save_best_only = True),
    keras.callbacks.EarlyStopping(patience=10, min_delta=1e-3),
]
history = model.fit(x_train_scaled, y_train, epochs=20,
                    validation_data=(x_valid_scaled, y_valid),
                    callbacks = callbacks)

Siete, dibujando el proceso de cambio en el modelo de formación

Durante el proceso de entrenamiento, el gradiente puede explotar o desaparecer. Por razones específicas, consulte: Explicación detallada de la desaparición del gradiente, razones de explosión y soluciones en el aprendizaje automático.
Aquí está el enlace a la derivación detallada del autor y al proceso de valor extremo de la función sigmoide de la siguiente manera:
derivación de la función sigmoidea, valor extremo (el más detallado en la historia)

def plot_learning_curves(history):
    pd.DataFrame(history.history).plot(figsize=(8, 5))
    plt.grid(True)
    plt.gca().set_ylim(0, 3)
    plt.show()

plot_learning_curves(history)

Ocho, llama al estimador

print(model.evaluate(x_test_scaled, y_test, verbose=0))

Nueve, resumen de código

#!/usr/bin/env python3
# -*- coding: utf-8 -*-

import matplotlib as mpl
import matplotlib.pyplot as plt
import numpy as np
import sklearn
import pandas as pd
import os
import sys
import time
import tensorflow as tf
from tensorflow import keras

# 1,打印使用的python库的版本信息
print(sys.version_info)
for module in mpl, np, pd, sklearn, tf, keras:
    print(module.__name__, module.__version__)


# 2,从keras.datasets中提取图片数据集-“训练数据”、“验证数据”、“测试数据”，并打印其维度
fashion_mnist = keras.datasets.fashion_mnist
(x_train_all, y_train_all), (x_test, y_test) = fashion_mnist.load_data()
x_valid, x_train = x_train_all[:5000], x_train_all[5000:]
y_valid, y_train = y_train_all[:5000], y_train_all[5000:]

print(x_valid.shape, y_valid.shape)
print(x_train.shape, y_train.shape)
print(x_test.shape, y_test.shape)

# 3,对数据标准化处理
# x = (x - u) / std
from sklearn.preprocessing import StandardScaler

scaler = StandardScaler()
# x_train: [None, 28, 28] -> [None, 784]
x_train_scaled = scaler.fit_transform(
    x_train.astype(np.float32).reshape(-1, 1)).reshape(-1, 28, 28)
x_valid_scaled = scaler.transform(
    x_valid.astype(np.float32).reshape(-1, 1)).reshape(-1, 28, 28)
x_test_scaled = scaler.transform(
    x_test.astype(np.float32).reshape(-1, 1)).reshape(-1, 28, 28)

# 4,构建DNN深度神经网络模型
# tf.keras.models.Sequential()
model = keras.models.Sequential()
model.add(keras.layers.Flatten(input_shape=[28, 28]))
for _ in range(30):
    model.add(keras.layers.Dense(100, activation="relu"))
model.add(keras.layers.Dense(10, activation="softmax"))

model.compile(loss="sparse_categorical_crossentropy",
              optimizer = keras.optimizers.SGD(0.01),
              metrics = ["accuracy"])

# 5,查看模型整体结构-及参数
print(model.summary())

# 6,添加回调函数、训练模型
# Tensorboard, earlystopping, ModelCheckpoint
logdir = './dnn-callbacks'
if not os.path.exists(logdir):
    os.mkdir(logdir)
output_model_file = os.path.join(logdir,"fashion_mnist_model.h5")

callbacks = [
    keras.callbacks.TensorBoard(logdir),
    keras.callbacks.ModelCheckpoint(output_model_file,save_best_only = True),
    keras.callbacks.EarlyStopping(patience=10, min_delta=1e-3),
]
history = model.fit(x_train_scaled, y_train, epochs=20,
                    validation_data=(x_valid_scaled, y_valid),
                    callbacks = callbacks)

# 7,绘制训练模型中，变化过程
def plot_learning_curves(history):
    pd.DataFrame(history.history).plot(figsize=(8, 5))
    plt.grid(True)
    plt.gca().set_ylim(0, 3)
    plt.show()

plot_learning_curves(history)

# 1. 参数众多，训练不充分
# 2. 梯度消失 -> 链式法则 -> 复合函数f(g(x))

# 8,调用估计器
print(model.evaluate(x_test_scaled, y_test, verbose=0))