3

我想构建一个分类器，在给定时间序列向量的情况下提供标签。我有基于静态 LSTM 的分类器的代码，但我不知道如何合并时间信息：

训练集：

time   = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10,11,12,13,14,15,16,17,18]
f1     = [1, 1, 1, 2, 2, 2, 1, 1, 1, 2, 2, 2, 1, 1, 1, 2, 2, 2]
f2     = [2, 1, 3, 2, 4, 2, 3, 1, 9, 2, 1, 2, 1, 6, 1, 8, 2, 2]
labels = [a, a, b, b, a, a, b, b, a, a, b, b, a, a, b, b, a, a]

测试集：

time   = [1, 2, 3, 4, 5, 6]
f1     = [2, 2, 2, 1, 1, 1]
f2     = [2, 1, 2, 1, 6, 1]
labels = [?, ?, ?, ?, ?, ?]

在这篇文章之后，我在 pybrain 中实现了以下内容：

from pybrain.datasets import SequentialDataSet
from itertools import cycle
import matplotlib.pyplot as plt
from pybrain.tools.shortcuts import buildNetwork
from pybrain.structure.modules import LSTMLayer
from pybrain.supervised import RPropMinusTrainer
from sys import stdout

data = [1,2,3,4,5,6,7]

ds = SequentialDataSet(1, 1)
for sample, next_sample in zip(data, cycle(data[1:])):
    ds.addSample(sample, next_sample)

print ds
net = buildNetwork(2, 5, 1, hiddenclass=LSTMLayer, outputbias=False, recurrent=True)


trainer = RPropMinusTrainer(net, dataset=ds)
train_errors = [] # save errors for plotting later
EPOCHS_PER_CYCLE = 5
CYCLES = 100
EPOCHS = EPOCHS_PER_CYCLE * CYCLES
for i in xrange(CYCLES):
    trainer.trainEpochs(EPOCHS_PER_CYCLE)
    train_errors.append(trainer.testOnData())
    epoch = (i+1) * EPOCHS_PER_CYCLE
    print("\r epoch {}/{}".format(epoch, EPOCHS))
    stdout.flush()

print()
print("final error =", train_errors[-1])

plt.plot(range(0, EPOCHS, EPOCHS_PER_CYCLE), train_errors)
plt.xlabel('epoch')
plt.ylabel('error')
plt.show()

for sample, target in ds.getSequenceIterator(0):
    print("               sample = %4.1f" % sample)
    print("predicted next sample = %4.1f" % net.activate(sample))
    print("   actual next sample = %4.1f" % target)
    print()

这训练了一个分类器，但我不知道如何合并时间信息。如何包含有关向量顺序的信息？

4

1 回答 1

0

这就是我实现序列标记的方式。我有六类标签。每个班级我有 20 个样本序列。每个序列由 100 个时间步长的数据点和 10 个变量组成。

input_variable = 10
output_class = 1
trndata = SequenceClassificationDataSet(input_variable,output_label, nb_classes=6)

 # input 1st sequence into dataset for class label 0
 for i in range(100):
     trndata.appendLinked(sequence1_class0[i,:], [0])
 trndata.newSequence()

 # input 2nd sequence into dataset for class label 0
 for i in range(100):
     trndata.appendLinked(sequence2_class0[i,:], [0])
 trndata.newSequence()
 ......
 ......

 # input 20th sequence into dataset for class label 5
 for i in range(100):
     trndata.appendLinked(sequence20_class5[i,:], [5])
 trndata.newSequence()

您最终可以将它们全部放入 for 循环中。每次将新的样本序列作为数据集给出时，都会调用 trndata.newSequence()。

网络的训练应该类似于你现有的代码。

于 2016-03-24T08:07:42.140 回答