TensorFlow 非常高的损失值答案

【问题标题】：Tensorflow very high loss valueTensorFlow 非常高的损失值
【发布时间】：2021-07-26 07:02:57
【问题描述】：

import tensorflow as tf
import matplotlib.pyplot as plt
import os
import numpy as np
from tensorflow.keras.layers import Dense, Dropout, Activation, Flatten, DenseFeatures
from tensorflow.keras.layers import Conv2D, MaxPooling2D
from tensorflow.keras.preprocessing.image import ImageDataGenerator
from tensorflow.keras.preprocessing import image
import tensorflow.keras.optimizers as op
from tensorflow.keras.models import load_model

train = ImageDataGenerator(rescale=1. / 255,rotation_range=20,
    width_shift_range=0.2,
    height_shift_range=0.2,
    shear_range=0.2,
    zoom_range=0.2,
    horizontal_flip=True,
    vertical_flip=True,
    fill_mode='nearest')
validation = ImageDataGenerator(rescale=1. / 255)
test = ImageDataGenerator(rescale=1. / 255)
train_dataset = train.flow_from_directory('raw-img/training', batch_size=1,
                                          class_mode='categorical', subset="training")

validation_dataset = validation.flow_from_directory('raw-img/validation', target_size=(224, 224), batch_size=1,
                                                    class_mode='categorical')
testing_dataset = test.flow_from_directory('raw-img/testing', target_size=(224, 224), batch_size=1,
                                           class_mode='categorical')


model = tf.keras.models.Sequential([
    tf.keras.layers.Conv2D(32, (3,3), input_shape=(224, 224, 3), padding='same'),
    tf.keras.layers.BatchNormalization(),
    tf.keras.layers.LeakyReLU(alpha=0.3),
    tf.keras.layers.MaxPool2D(2, 2, padding='same'),
    tf.keras.layers.Dropout(0.01),

    #
    tf.keras.layers.Conv2D(32, (3,3), padding='same'),

    tf.keras.layers.LeakyReLU(alpha=0.3),
    tf.keras.layers.Dropout(0.01),
    tf.keras.layers.Conv2D(32, (3,3), padding='same'),

    tf.keras.layers.LeakyReLU(alpha=0.3),
    tf.keras.layers.MaxPool2D(2, 2, padding='same'),
    tf.keras.layers.Dropout(0.01),

    #
    tf.keras.layers.Conv2D(64, (3,3), padding='same'),

    tf.keras.layers.LeakyReLU(alpha=0.3),
    tf.keras.layers.Dropout(0.01),
    tf.keras.layers.Conv2D(64, (3,3), padding='same'),

    tf.keras.layers.LeakyReLU(alpha=0.3),
    tf.keras.layers.MaxPool2D(2, 2, padding='same'),
    tf.keras.layers.Dropout(0.01),
    #
    tf.keras.layers.Conv2D(128, (3,3), padding='same'),

    tf.keras.layers.LeakyReLU(alpha=0.3),
    tf.keras.layers.Dropout(0.01),
    tf.keras.layers.Conv2D(128, (3,3), padding='same'),

    tf.keras.layers.LeakyReLU(alpha=0.3),
    tf.keras.layers.MaxPool2D(2, 2, padding='same'),
    tf.keras.layers.Dropout(0.01),

    #
    tf.keras.layers.Conv2D(128, (3,3), padding='same'),

    tf.keras.layers.LeakyReLU(alpha=0.3),
    tf.keras.layers.Dropout(0.01),
    tf.keras.layers.Conv2D(128, (3,3), padding='same'),

    tf.keras.layers.LeakyReLU(alpha=0.3),
    tf.keras.layers.MaxPool2D(2, 2, padding='same'),
    tf.keras.layers.Dropout(0.01),

    #

    tf.keras.layers.Conv2D(256, (3,3), padding='same'),

    tf.keras.layers.LeakyReLU(alpha=0.3),
    tf.keras.layers.Dropout(0.01),
    tf.keras.layers.Conv2D(256, (3,3), padding='same'),

    tf.keras.layers.LeakyReLU(alpha=0.3),
    tf.keras.layers.MaxPool2D(2, 2, padding='same'),
    tf.keras.layers.Dropout(0.01),
    #
    tf.keras.layers.Conv2D(512, (3,3), padding='same'),

    tf.keras.layers.LeakyReLU(alpha=0.3),
    tf.keras.layers.Dropout(0.01),
    tf.keras.layers.Conv2D(512, (3,3), padding='same'),

    tf.keras.layers.LeakyReLU(alpha=0.3),
    tf.keras.layers.MaxPool2D(2, 2, padding='same'),
    tf.keras.layers.Dropout(0.01),

    #
    tf.keras.layers.Flatten(),
    #
    tf.keras.layers.Dense(1024),
    tf.keras.layers.LeakyReLU(alpha=0.3),
    tf.keras.layers.Dropout(0.01),

    tf.keras.layers.Dense(512),
    tf.keras.layers.LeakyReLU(alpha=0.3),
    tf.keras.layers.Dropout(0.01),

    tf.keras.layers.Dense(10, activation='softmax'),

])
# print the summary of the model architecture
model.summary()

rlronp = tf.keras.callbacks.ReduceLROnPlateau(monitor='val_loss', factor=.8, patience=3, verbose=1, mode='min')
es = tf.keras.callbacks.EarlyStopping(monitor="val_loss", patience=15, verbose=1,
                                      restore_best_weights=True)
model.compile(loss='categorical_crossentropy', optimizer=op.Adam(learning_rate=0.001), metrics=['accuracy'])
model_fit = model.fit(train_dataset, epochs=100, batch_size=1, validation_data=validation_dataset,
                      callbacks=[rlronp, es])

5865/26179 [=====>.......................] - ETA：12:23 - 损失：34992108.0000 - 准确度: 0.1589

您好，我有带有 adam 优化器的训练模型，但损失率非常高，我尝试将优化器更改为 SDG，但预测不正确。

如何更改代码以获得更好的结果和低损失值。

损失值 34992108.0000

【问题讨论】：

尝试在每个 CNN 块中添加 BatchNormalization() 层。 batch_size 是 1 尝试将其更改为 32 或 64。
@AdarshWase 的解决方案是否解决了您的问题？
@AdarshWase 我尝试在 Dropout 之后添加 BatchNormalization() ，然后在 Dropout 层损失减少到 2.200 之前添加。但是这个值上的损失值堆栈（2.200）
我应该在哪里添加 BatchNormalization() ？？
激活层之后。

标签： python tensorflow machine-learning keras deep-learning

【解决方案1】：

我复制了相同的代码并找到了解决此问题的方法，您可以通过在模型定义中进行以下更改来尝试：

您需要删除之前定义的所有Dropout layers Flatten() 模型中的层。
在训练数据集时还要定义batch_size=128。

如果问题仍然存在，请告诉我们。快乐编码！

【讨论】：