iT邦幫忙

0

tensorflow 載入自己訓練好的 model 錯誤

我如果直接訓練的話,把 model 輸出後再用 model = load_model('./my_model.h5') 讀取是正常的

但如果把註解的部分拿掉,model.fit 換成使用註解的,訓練完後再把 model 載入就會出問題?

請問怎麼解決? 用了半天用不出來QQ

1_mnist.py

import tensorflow as tf
from tensorflow.keras.preprocessing.image import ImageDataGenerator

mnist = tf.keras.datasets.mnist
(x_train, y_train), (x_test, y_test) = mnist.load_data()
x_train, x_test = x_train/255.0, x_test/255.0
# x_train = x_train.reshape(x_train.shape[0], 28, 28, 1)
# x_test = x_test.reshape(x_test.shape[0], 28, 28, 1)

# image_gen_train = ImageDataGenerator(
#     rescale=1./1.,
#     rotation_range=45,
#     width_shift_range=15,
#     height_shift_range=15,
#     horizontal_flip=True,
#     zoom_range=0.5
# )
# image_gen_train.fit(x_train)

model = tf.keras.models.Sequential([
    tf.keras.layers.Flatten(),
    tf.keras.layers.Dense(128, activation='relu'),
    tf.keras.layers.Dense(10, activation='softmax')
])

model.compile(optimizer='adam',
                loss = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=False),
                metrics=['sparse_categorical_accuracy'])

# model.fit(image_gen_train.flow(x_train, y_train, batch_size=32), epochs=5)
model.fit(x_train, y_train, batch_size=32, epochs=5)
model.summary()
model.save('my_model.h5')

使用註解掉的程式碼會出現以下錯誤

raise ValueError('The last dimension of the inputs to `Dense` '
ValueError: The last dimension of the inputs to `Dense` should be defined. Found `None`.

2 個回答

0
I code so I am
iT邦研究生 3 級 ‧ 2021-02-04 10:39:45
最佳解答

Flatten 要加 input_shape 如下:

tf.keras.layers.Flatten(input_shape=(28, 28))

可以了,謝謝
不過為什麼要加 input_shape 呢?
Flatten 不是單純把資料展開嗎?

模型一定要設定input的規格,因此,第一層需有 input_shape,或者

model = tf.keras.models.Sequential([
  tf.keras.layers.Flatten(),
  tf.keras.layers.Dense(128, activation='relu'),
  tf.keras.layers.Dropout(0.2),
  tf.keras.layers.Dense(10, activation='softmax')
])

x = tf.keras.layers.Input(shape=(28, 28))
# 或 x = tf.Variable(tf.random.truncated_normal([28, 28]))
y = model(x)
0
hokou
iT邦新手 4 級 ‧ 2021-02-03 08:31:00

我單純讀存起來的 Model 沒問題
但是 Model 怪怪的,不知道是不是沒 INPUT 或是 fit_generator 的關係
可能要請其他人解答

可以參考下面的官方連結

一開始的

Epoch 1/5
60000/60000 [==============================] - 3s 56us/sample - loss: 0.2595 - sparse_categorical_accuracy: 0.9258
Epoch 2/5
60000/60000 [==============================] - 3s 53us/sample - loss: 0.1157 - sparse_categorical_accuracy: 0.9660
Epoch 3/5
60000/60000 [==============================] - 3s 54us/sample - loss: 0.0799 - sparse_categorical_accuracy: 0.9757
Epoch 4/5
60000/60000 [==============================] - 3s 54us/sample - loss: 0.0594 - sparse_categorical_accuracy: 0.9819
Epoch 5/5
60000/60000 [==============================] - 3s 53us/sample - loss: 0.0457 - sparse_categorical_accuracy: 0.9861

Model: "sequential"
_________________________________________________________________
Layer (type)                 Output Shape              Param #
=================================================================
flatten (Flatten)            multiple                  0
_________________________________________________________________
dense (Dense)                multiple                  100480
_________________________________________________________________
dense_1 (Dense)              multiple                  1290
=================================================================
Total params: 101,770
Trainable params: 101,770
Non-trainable params: 0

用 ImageDataGenerator

1875/1875 [==============================] - 24s 13ms/step - loss: 2.1724 - sparse_categorical_accuracy: 0.1695
Epoch 2/5
1875/1875 [==============================] - 23s 12ms/step - loss: 2.0163 - sparse_categorical_accuracy: 0.2663
Epoch 3/5
1875/1875 [==============================] - 22s 12ms/step - loss: 1.9251 - sparse_categorical_accuracy: 0.3191
Epoch 4/5
1875/1875 [==============================] - 22s 12ms/step - loss: 1.8789 - sparse_categorical_accuracy: 0.3411
Epoch 5/5
1875/1875 [==============================] - 22s 12ms/step - loss: 1.8455 - sparse_categorical_accuracy: 0.3585

Model: "sequential"
_________________________________________________________________
Layer (type)                 Output Shape              Param #
=================================================================
flatten (Flatten)            multiple                  0
_________________________________________________________________
dense (Dense)                multiple                  100480
_________________________________________________________________
dense_1 (Dense)              multiple                  1290
=================================================================
Total params: 101,770
Trainable params: 101,770
Non-trainable params: 0
_________________________________________________________________

參考官方範例
开始使用 Keras Sequential 顺序模型

from keras.models import Sequential
from keras.layers import Dense, Activation

model = Sequential([
    Dense(32, input_shape=(784,)),
    Activation('relu'),
    Dense(10),
    Activation('softmax'),
])

ImageDataGenerator 类

(x_train, y_train), (x_test, y_test) = cifar10.load_data()
y_train = np_utils.to_categorical(y_train, num_classes)
y_test = np_utils.to_categorical(y_test, num_classes)

datagen = ImageDataGenerator(
    featurewise_center=True,
    featurewise_std_normalization=True,
    rotation_range=20,
    width_shift_range=0.2,
    height_shift_range=0.2,
    horizontal_flip=True)

# 计算特征归一化所需的数量
# (如果应用 ZCA 白化,将计算标准差,均值,主成分)
datagen.fit(x_train)

# 使用实时数据增益的批数据对模型进行拟合:
model.fit_generator(datagen.flow(x_train, y_train, batch_size=32),
                    steps_per_epoch=len(x_train) / 32, epochs=epochs)

# 这里有一个更 「手动」的例子
for e in range(epochs):
    print('Epoch', e)
    batches = 0
    for x_batch, y_batch in datagen.flow(x_train, y_train, batch_size=32):
        model.fit(x_batch, y_batch)
        batches += 1
        if batches >= len(x_train) / 32:
            # 我们需要手动打破循环,
            # 因为生成器会无限循环
            break

我有個好奇的點,為什麼我在我電腦上兩種方法的數值都是 1875/1875,而妳的會有 60000/60000 跟 1875/1875 兩種呢?
是我的問題嗎XD

hokou iT邦新手 4 級 ‧ 2021-02-04 08:32:30 檢舉

不確定是不是顯示的差別

基本上是 1875*32 = 60000 images
batch_size 預設是 32
所以 1875 是 1875 batches
batch_size 每次 32張圖

參考
Why the model is training on only 1875 training set images if there are 60000 images in the MNIST dataset?

我要發表回答

立即登入回答