模型訓練問題

#train #ai #machine learning #accuracy

tc_89 2023-09-20 01:40:07 ‧ 1100 瀏覽

分享至

為什麼在training model的時候，accuracy一直都是0，要怎麼解決?

whoami iT邦新手 3 級 ‧ 2023-09-20 09:59:19 檢舉

可以提供更多資訊嘛?

tc_89 iT邦新手 5 級 ‧ 2023-09-20 14:02:56 檢舉

下面是我設定的模型
from keras.models import Sequential
from keras import layers
from tensorflow.keras import optimizers
from keras.layers import BatchNormalization,Dropout,Dense
from keras.callbacks import ModelCheckpoint

#建立model
def model():
model = Sequential()
model.add(layers.Dense(1024,kernel_initializer = 'random_normal',
activation = 'relu',
input_shape = (X_train_normal_data.shape[1],)))
model.add(Dropout(0.3))

model.add(layers.Dense(1024, kernel_initializer = 'random_normal',
activation = 'relu'))
model.add(Dropout(0.3))

model.add(layers.Dense(512, kernel_initializer = 'random_normal',
activation = 'relu'))
model.add(Dropout(0.3))

model.add(layers.Dense(256, kernel_initializer = 'random_normal',
activation = 'relu'))
model.add(Dropout(0.3))

model.add(layers.Dense(128, kernel_initializer = 'random_normal',
activation = 'relu'))
model.add(Dropout(0.3))

model.add(layers.Dense(32, kernel_initializer = 'random_normal',
activation = 'relu'))
model.add(Dropout(0.3))

model.add(layers.Dense(16, kernel_initializer = 'random_normal',
activation = 'relu'))
model.add(Dropout(0.3))

model.add(layers.Dense(1, kernel_initializer = 'random_normal',
activation = 'linear'))

adam = optimizers.Adam(lr=0.001)
#lr學習率
model.compile(optimizer = adam, loss = 'mae', metrics=['accuracy'])
return model

call = ModelCheckpoint('good.h5',
monitor = 'val_loss',
verbose = 0,
save_best_only = True,
save_weights_only = True,
mode = 'auto',
period = 1)

model = model()
history = model.fit(X_train_normal_data, Y_train,
validation_data = (X_validation_normal_data, Y_validation),
callbacks = [call],
epochs = 300,
batch_size = 250, verbose = 1)

#建立model
def model():
    model = Sequential()
    model.add(layers.Dense(1024,kernel_initializer = 'random_normal',
             activation = 'relu',
             input_shape = (X_train_normal_data.shape[1],)))
    model.add(Dropout(0.3))

model.add(layers.Dense(1024, kernel_initializer = 'random_normal',
                        activation = 'relu'))
    model.add(Dropout(0.3))

model.add(layers.Dense(512, kernel_initializer = 'random_normal',
                        activation = 'relu'))
    model.add(Dropout(0.3))

model.add(layers.Dense(256, kernel_initializer = 'random_normal',
                           activation = 'relu'))
    model.add(Dropout(0.3))

model.add(layers.Dense(128, kernel_initializer = 'random_normal',
                           activation = 'relu'))
    model.add(Dropout(0.3))

model.add(layers.Dense(32, kernel_initializer = 'random_normal',
                           activation = 'relu'))
    model.add(Dropout(0.3))

model.add(layers.Dense(16, kernel_initializer = 'random_normal',
                           activation = 'relu'))
    model.add(Dropout(0.3))

model.add(layers.Dense(1, kernel_initializer = 'random_normal',
                          activation = 'linear'))

adam = optimizers.Adam(lr=0.001)
    #lr學習率
    model.compile(optimizer = adam, loss = 'mae', metrics=['accuracy'])
    return model

call = ModelCheckpoint('good.h5',
                     monitor = 'val_loss',
                     verbose = 0,
                     save_best_only = True,
                     save_weights_only = True,
                     mode = 'auto',
                     period = 1)

model = model()
history = model.fit(X_train_normal_data, Y_train,
                    validation_data = (X_validation_normal_data, Y_validation),
                    callbacks = [call],
                    epochs = 300,
                    batch_size = 250, verbose = 1)

修改

登入發表討論