python - 使用 Keras 进行整数序列预测

Question

我正在尝试编写一个 RNN 模型来预测整数系列中的下一个数字。每个时期的模型损失都会变小，但预测永远不会变得非常准确。我尝试过许多训练集的大小和纪元数，但我的预测值总是与预期相差几位数。你能给我一些建议，告诉我哪些地方需要改进，或者我做错了什么吗？这是代码:

from keras.models import Sequential
from keras.layers import Dense, Dropout, LSTM
from keras.callbacks import ModelCheckpoint
from keras.utils import np_utils
from keras import metrics
import numpy as np

training_length = 10000
rnn_size = 512
hm_epochs = 30

def generate_sequence(length=10):
    step = np.random.randint(0,50)
    first_element = np.random.randint(0,10)
    first_element = 0
    l_ist = [(first_element + (step*i)) for i in range(length)]
    return l_ist

training_set = []

for _ in range(training_length):
    training_set.append(generate_sequence(10))

feature_set = [i[:-1] for i in training_set]

label_set = [i[-1:] for i in training_set]

X = np.reshape(feature_set,(training_length, 9, 1))
y = np.array(label_set)


model = Sequential()
model.add(LSTM(rnn_size, input_shape = (X.shape[1], X.shape[2]), return_sequences = True))
model.add(Dropout(0.2))
model.add(LSTM(rnn_size))
model.add(Dropout(0.2))
model.add(Dense(y.shape[1], activation='linear'))
model.compile(loss='mse', optimizer='rmsprop', metrics=['accuracy'])

filepath="checkpoint_folder/weights-improvement.hdf5"
checkpoint = ModelCheckpoint(filepath, monitor='loss', verbose=1, save_best_only=True, mode='min')
callbacks_list = [checkpoint]

model.fit(X,y,epochs=hm_epochs, callbacks=callbacks_list)

效果:

30 个 epoch 后(损失:66.39):

1 序列:[0, 20, 40, 60, 80, 100, 120, 140, 160]预期:[180] ||得到:[181.86118]

2 序列:[0, 11, 22, 33, 44, 55, 66, 77, 88]预期:[99] ||得到:[102.17369]

3 序列:[0, 47, 94, 141, 188, 235, 282, 329, 376]预期:[423] ||得到:[419.1763]

4 序列:[0, 47, 94, 141, 188, 235, 282, 329, 376]预期:[423] ||得到:[419.1763]

5 序列:[0, 4, 8, 12, 16, 20, 24, 28, 32] 预期:[36] ||得到:[37.506496]

6 序列:[0, 48, 96, 144, 192, 240, 288, 336, 384]预期:[432] ||得到:[425.0569]

7 序列:[0, 28, 56, 84, 112, 140, 168, 196, 224]预期:[252] ||得到:[253.60233]

8 序列:[0, 18, 36, 54, 72, 90, 108, 126, 144]预期:[162] ||得到:[163.538]

9 序列:[0, 19, 38, 57, 76, 95, 114, 133, 152]预期:[171] ||得到:[173.77933]

10 序列:[0, 1, 2, 3, 4, 5, 6, 7, 8]预期:[9] ||得到:[9.577981]

...

100 个 epoch 后(损失:54.81):

1 序列:[0, 20, 40, 60, 80, 100, 120, 140, 160] 预期:[180] ||得到:[181.03535]

2 序列:[0, 11, 22, 33, 44, 55, 66, 77, 88] 预期:[99] ||得到:[99.15022]

3 序列:[0, 47, 94, 141, 188, 235, 282, 329, 376] 预期:[423] ||得到:[423.7969]

4 序列:[0, 47, 94, 141, 188, 235, 282, 329, 376] 预期:[423] ||得到:[423.7969]

5 序列:[0, 4, 8, 12, 16, 20, 24, 28, 32] 预期:[36] ||得到:[34.47298]

6 序列:[0, 48, 96, 144, 192, 240, 288, 336, 384] 预期:[432] ||得到:[432.73163]

7 序列:[0, 28, 56, 84, 112, 140, 168, 196, 224] 预期:[252] ||得到:[251.55792]

8 序列:[0, 18, 36, 54, 72, 90, 108, 126, 144] 预期:[162] ||得到:[164.81227]

9 序列:[0, 19, 38, 57, 76, 95, 114, 133, 152] 预期:[171] ||得到:[172.6425]

10 序列:[0, 1, 2, 3, 4, 5, 6, 7, 8] 预期:[9] ||得到:[8.837313]

Answer 1

你尝试过更长的序列吗？ LSTM不是必需的，因为依赖关系不是很长。您可以尝试使用 RNN 的另一种变体。