Tensorflow ValueError：没有为任何变量提供梯度

Posted 2023-03-28

【中文标题】Tensorflow ValueError：没有为任何变量提供梯度

【英文标题】：Tensorflow ValueError: No gradients provided for any variable

【发布时间】：2020-10-31 21:06:26

【问题描述】：

我正在尝试让我的 tensorflow 模型在 2 类图像上进行训练，但我遇到了 ValueError 问题。有人可以帮忙吗。 以下是相关代码：

# Get image arrays and labels for all image files
images, labels = load_data(sys.argv[1])

# Split data into training and testing sets
x_train, x_test, y_train, y_test = train_test_split(
    images, labels, test_size=TEST_SIZE
)

# Get a compiled neural network
model = get_model()
model.summary()

# Fit model on training data
model.fit_generator(x_train, steps_per_epoch=128, epochs=EPOCHS,
                    validation_data=y_train, validation_steps=128)

def load_data(data_dir):
    image_generator = ImageDataGenerator(rescale=1. / 255)
    resized_imgs = image_generator.flow_from_directory(batch_size=128, directory=data_dir,
                              shuffle=True, target_size=dimensions,
       class_mode='binary')

    images, labels = next(resized_imgs)
    plotImages(images[:15])

    return images, labels


def get_model():
    # create a convolutional neural network
    model = tf.keras.models.Sequential([

        # convolutional layer. Learn 32 filters using 

a 3x3 kernel
        tf.keras.layers.Conv2D(
            32, (3, 3), activation="relu", input_shape=(IMG_WIDTH, IMG_HEIGHT, 3)
    ),

    tf.keras.layers.BatchNormalization(),

    # max-pooling layer, using 2x2 pool size
    tf.keras.layers.MaxPooling2D(pool_size=(2, 2)),

    # convolutional layer. Learn 32 filters using a 3x3 kernel
    tf.keras.layers.Conv2D(
        32, (3, 3), activation="relu", input_shape=(IMG_WIDTH, IMG_HEIGHT, 3)
    ),

    tf.keras.layers.BatchNormalization(),

    # max-pooling layer, using 2x2 pool size
    tf.keras.layers.MaxPooling2D(pool_size=(2, 2)),

    # flatten units
    tf.keras.layers.Flatten(),

    # add a hidden layer with dropout
    tf.keras.layers.Dense(128, activation="relu"),
    tf.keras.layers.Dropout(0.5),

    # add an output layer with NUM_CATEGORIES (43) units
    tf.keras.layers.Dense(NUM_CATEGORIES, activation="sigmoid")  # changed activation from softmax
    # to sigmoid whic is the proper activation for binary data
])

# train neural network
model.compile(
    optimizer="adam",
    loss=tf.keras.losses.BinaryCrossentropy(from_logits=True), metrics=["accuracy"]
)

return model

我最终得到以下错误： ValueError：没有为任何变量提供梯度：['conv2d/kernel:0', 'conv2d/bias:0', 'batch_normalization/gamma:0', 'batch_normalization/beta:0', 'conv2d_1/kernel:0', 'conv2d_1/bias:0'、'batch_normalization_1/gamma:0'、'batch_normalization_1/beta:0'、'dense/kernel:0'、'dense/bias:0'、'dense_1/kernel:0'、'dense_1 /bias:0']。

错误来自以下代码行，但不知道如何修复它：

model.fit_generator(x_train, steps_per_epoch=128, epochs=EPOCHS,
                        validation_data=y_train, validation_steps=128)

谢谢

【问题讨论】：

您需要将y_train 传递给.fit()。另外，使用.fit() 而不是.fit_generator()

谢谢。我将拟合线更新为： model.fit(x_train, y_train, epochs=EPOCHS) 但是当我这样做时出现错误： ValueError: logits and labels must have the same shape ((None, 2) vs (None, 1 ))

那么似乎是数据预处理错误

我认为我的预处理很简单。您发现以下预处理代码有什么问题吗？ image_generator = ImageDataGenerator(rescale=1./255) resized_imgs = image_generator.flow_from_directory(batch_size=128, directory=data_dir, shuffle=True, target_size=dimensions, class_mode='binary')

【参考方案1】：

想通了。由于我的 tf 模型中的最终输出层，我的 logits 与我的标签形状不匹配。

NUM_CATEGORIES = 2

tf.keras.layers.Dense(NUM_CATEGORIES, activation="sigmoid")

我将单位设置为 2 而不是 1，所以我的输出形状是 (None, 2) 而不是 (None, 1)