所有时期的变量都相同

Question

我正在使用带有 keras 的 CNN 试验图像分类器

我的代码 -

model = Sequential()

model.add(Conv2D(32, (3, 3), padding='same', input_shape=(224, 224, 3), activation="relu"))
model.add(Conv2D(32, (3, 3), padding='same', activation="relu"))
model.add(MaxPooling2D(pool_size=(2, 2)))

model.add(Flatten())
model.add(Dense(5, activation="softmax"))

model.compile(
    loss='categorical_crossentropy',
    optimizer='rmsprop',
    metrics=['accuracy']
)
model.summary()

check=ModelCheckpoint(filepath=r'C:\Users\softloft\AppData\Local\Programs\Python\Python37\Scripts\Untitled Folder\i_models.hdf5', verbose=1, save_best_only = True)

history=model.fit(
    x_train,
    y_train,
    batch_size=100,
    epochs=30,
    validation_data=(x_test, y_test),
    shuffle=True,
    callbacks=[check],
)

我用一个 CNN 层运行这段代码，它给出的 val_acc 约为 71。

然后，我添加另一个 CNN 层（即上面的代码）并给出结果 -

Train on 3242 samples, validate on 1081 samples
Epoch 1/30
3242/3242 [==============================] - 235s 73ms/step - loss: 13.0773 - acc: 0.1771 - val_loss: 13.8219 - val_acc: 0.1425
Epoch 00001: val_loss improved from inf to 13.82190, saving model to i_models.hdf5
Epoch 2/30
3242/3242 [==============================] - 235s 73ms/step - loss: 13.2345 - acc: 0.1789 - val_loss: 13.8219 - val_acc: 0.1425

Epoch 00002: val_loss did not improve from 13.82190
Epoch 3/30
3242/3242 [==============================] - 235s 72ms/step - loss: 13.2345 - acc: 0.1789 - val_loss: 13.8219 - val_acc: 0.1425

Epoch 00003: val_loss did not improve from 13.82190
Epoch 4/30
3242/3242 [==============================] - 235s 73ms/step - loss: 13.2345 - acc: 0.1789 - val_loss: 13.8219 - val_acc: 0.1425

Epoch 00004: val_loss did not improve from 13.82190
Epoch 5/30
3242/3242 [==============================] - 235s 72ms/step - loss: 13.2345 - acc: 0.1789 - val_loss: 13.8219 - val_acc: 0.1425

Epoch 00005: val_loss did not improve from 13.82190
Epoch 6/30
3242/3242 [==============================] - 234s 72ms/step - loss: 13.2345 - acc: 0.1789 - val_loss: 13.8219 - val_acc: 0.1425

Epoch 00006: val_loss did not improve from 13.82190
Epoch 7/30
3242/3242 [==============================] - 235s 72ms/step - loss: 13.2345 - acc: 0.1789 - val_loss: 13.8219 - val_acc: 0.1425

Epoch 00007: val_loss did not improve from 13.82190
Epoch 8/30
3242/3242 [==============================] - 235s 72ms/step - loss: 13.2345 - acc: 0.1789 - val_loss: 13.8219 - val_acc: 0.1425

Epoch 00008: val_loss did not improve from 13.82190
Epoch 9/30
3242/3242 [==============================] - 235s 72ms/step - loss: 13.2345 - acc: 0.1789 - val_loss: 13.8219 - val_acc: 0.1425

Epoch 00009: val_loss did not improve from 13.82190
Epoch 10/30
3242/3242 [==============================] - 236s 73ms/step - loss: 13.2345 - acc: 0.1789 - val_loss: 13.8219 - val_acc: 0.1425

Epoch 00010: val_loss did not improve from 13.82190
Epoch 11/30
3242/3242 [==============================] - 236s 73ms/step - loss: 13.2345 - acc: 0.1789 - val_loss: 13.8219 - val_acc: 0.1425

Epoch 00011: val_loss did not improve from 13.82190
Epoch 12/30
3242/3242 [==============================] - 236s 73ms/step - loss: 13.2345 - acc: 0.1789 - val_loss: 13.8219 - val_acc: 0.1425

Epoch 00012: val_loss did not improve from 13.82190
Epoch 13/30
3242/3242 [==============================] - 235s 73ms/step - loss: 13.2345 - acc: 0.1789 - val_loss: 13.8219 - val_acc: 0.1425

Epoch 00013: val_loss did not improve from 13.82190
Epoch 14/30
3242/3242 [==============================] - 236s 73ms/step - loss: 13.2345 - acc: 0.1789 - val_loss: 13.8219 - val_acc: 0.1425

Epoch 00014: val_loss did not improve from 13.82190
Epoch 15/30
3242/3242 [==============================] - 236s 73ms/step - loss: 13.2345 - acc: 0.1789 - val_loss: 13.8219 - val_acc: 0.1425

Epoch 00015: val_loss did not improve from 13.82190
Epoch 16/30
3242/3242 [==============================] - 235s 73ms/step - loss: 13.2345 - acc: 0.1789 - val_loss: 13.8219 - val_acc: 0.1425

Epoch 00016: val_loss did not improve from 13.82190
Epoch 17/30
3242/3242 [==============================] - 236s 73ms/step - loss: 13.2345 - acc: 0.1789 - val_loss: 13.8219 - val_acc: 0.1425

Epoch 00017: val_loss did not improve from 13.82190
Epoch 18/30
3242/3242 [==============================] - 235s 73ms/step - loss: 13.2345 - acc: 0.1789 - val_loss: 13.8219 - val_acc: 0.1425

Epoch 00018: val_loss did not improve from 13.82190
Epoch 19/30
3242/3242 [==============================] - 235s 73ms/step - loss: 13.2345 - acc: 0.1789 - val_loss: 13.8219 - val_acc: 0.1425

Epoch 00019: val_loss did not improve from 13.82190
Epoch 20/30
3242/3242 [==============================] - 236s 73ms/step - loss: 13.2345 - acc: 0.1789 - val_loss: 13.8219 - val_acc: 0.1425

Epoch 00020: val_loss did not improve from 13.82190
Epoch 21/30
3242/3242 [==============================] - 236s 73ms/step - loss: 13.2345 - acc: 0.1789 - val_loss: 13.8219 - val_acc: 0.1425

Epoch 00021: val_loss did not improve from 13.82190
Epoch 22/30
3242/3242 [==============================] - 235s 72ms/step - loss: 13.2345 - acc: 0.1789 - val_loss: 13.8219 - val_acc: 0.1425

Epoch 00022: val_loss did not improve from 13.82190
Epoch 23/30
3242/3242 [==============================] - 236s 73ms/step - loss: 13.2345 - acc: 0.1789 - val_loss: 13.8219 - val_acc: 0.1425

Epoch 00023: val_loss did not improve from 13.82190
Epoch 24/30
3242/3242 [==============================] - 236s 73ms/step - loss: 13.2345 - acc: 0.1789 - val_loss: 13.8219 - val_acc: 0.1425

Epoch 00024: val_loss did not improve from 13.82190
Epoch 25/30
3242/3242 [==============================] - 236s 73ms/step - loss: 13.2345 - acc: 0.1789 - val_loss: 13.8219 - val_acc: 0.1425

Epoch 00025: val_loss did not improve from 13.82190
Epoch 26/30
3242/3242 [==============================] - 235s 73ms/step - loss: 13.2345 - acc: 0.1789 - val_loss: 13.8219 - val_acc: 0.1425

Epoch 00026: val_loss did not improve from 13.82190
Epoch 27/30
3242/3242 [==============================] - 235s 73ms/step - loss: 13.2345 - acc: 0.1789 - val_loss: 13.8219 - val_acc: 0.1425

Epoch 00027: val_loss did not improve from 13.82190
Epoch 28/30
3242/3242 [==============================] - 236s 73ms/step - loss: 13.2345 - acc: 0.1789 - val_loss: 13.8219 - val_acc: 0.1425

Epoch 00028: val_loss did not improve from 13.82190
Epoch 29/30
3242/3242 [==============================] - 236s 73ms/step - loss: 13.2345 - acc: 0.1789 - val_loss: 13.8219 - val_acc: 0.1425

Epoch 00029: val_loss did not improve from 13.82190
Epoch 30/30
3242/3242 [==============================] - 236s 73ms/step - loss: 13.2345 - acc: 0.1789 - val_loss: 13.8219 - val_acc: 0.1425

Epoch 00030: val_loss did not improve from 13.82190

会发生什么，该怎么做..

所有时期的所有变量如何相同

谢谢

Answer 1

尝试更改此设置

model.compile(loss = "categorical_crossentropy", optimizer = "rmsprop")

到

model.compile(loss = "categorical_crossentropy", optimizer = 'adam')