如何使用 Keras TimeseriesGenerator

Question

我在实施 Keras TimeseriesGenerator 时遇到问题。我想要的是为look_back 尝试不同的值，这是一个变量，它根据每个 y 确定 X 的滞后长度。现在，我将其设置为 3，但希望能够测试多个值。本质上，我想看看使用最后 n 行来预测一个值是否会提高准确性。这是我的代码：

### trying with timeseries generator
from keras.preprocessing.sequence import TimeseriesGenerator

look_back = 3

train_data_gen = TimeseriesGenerator(X_train, X_train,
    length=look_back, sampling_rate=1,stride=1,
    batch_size=3)
test_data_gen = TimeseriesGenerator(X_test, X_test,
    length=look_back, sampling_rate=1,stride=1,
    batch_size=1)

### Bi_LSTM
Bi_LSTM = Sequential()
Bi_LSTM.add(layers.Bidirectional(layers.LSTM(512, input_shape=(look_back, 11))))
Bi_LSTM.add(layers.Dropout(.5))
# Bi_LSTM.add(layers.Flatten())
Bi_LSTM.add(Dense(11, activation='softmax'))
Bi_LSTM.compile(optimizer='rmsprop',
              loss='categorical_crossentropy',
              metrics=['accuracy'])
### fitting a small normal model seems to be necessary for compile
Bi_LSTM.fit(X_train[:1],
              y_train[:1],
              epochs=1,
              batch_size=32,
              validation_data=(X_test[:1], y_test[:1]),
              class_weight=class_weights)
print('ignore above, necessary to run custom generator...')
Bi_LSTM_history = Bi_LSTM.fit_generator(Bi_LSTM.fit_generator(generator,
                                                    steps_per_epoch=1,
                                                    epochs=20,
                                                    verbose=0,
                                                    class_weight=class_weights))

这会产生以下错误：

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-35-11561ec7fb92> in <module>()
     26               batch_size=32,
     27               validation_data=(X_test[:1], y_test[:1]),
---> 28               class_weight=class_weights)
     29 print('ignore above, necessary to run custom generator...')
     30 Bi_LSTM_history = Bi_LSTM.fit_generator(Bi_LSTM.fit_generator(generator,

2 frames
/usr/local/lib/python3.6/dist-packages/keras/engine/training_utils.py in standardize_input_data(data, names, shapes, check_batch_axis, exception_prefix)
    143                             ': expected ' + names[i] + ' to have shape ' +
    144                             str(shape) + ' but got array with shape ' +
--> 145                             str(data_shape))
    146     return data
    147 

ValueError: Error when checking input: expected lstm_16_input to have shape (3, 11) but got array with shape (1, 11)

如果我将 BiLSTM 输入形状更改为上面列出的 (1,11)，则会收到此错误：

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-36-7360e3790518> in <module>()
     31                                                     epochs=20,
     32                                                     verbose=0,
---> 33                                                     class_weight=class_weights))
     34 

5 frames
/usr/local/lib/python3.6/dist-packages/keras/engine/training_utils.py in standardize_input_data(data, names, shapes, check_batch_axis, exception_prefix)
    143                             ': expected ' + names[i] + ' to have shape ' +
    144                             str(shape) + ' but got array with shape ' +
--> 145                             str(data_shape))
    146     return data
    147 

ValueError: Error when checking input: expected lstm_17_input to have shape (1, 11) but got array with shape (3, 11)

这里发生了什么？

如果需要，我的数据从 df 中读取，其中每一行（观察）是一个 (1,11) 浮点向量，每个标签是一个 int，我将其转换为 1 个热向量形状 (1,11)。

Answer 1

我在代码中发现了很多错误...因此，我想提供一个虚拟示例，您可以按照它来执行任务。请注意您的数据的原始维度和 TimeSeriesGenerator 生成的数据的维度。这对于了解如何构建网络很重要

# utility variable
look_back = 3
batch_size = 3
n_feat = 11
n_class = 11
n_train = 200
n_test = 60

# data simulation
X_train = np.random.uniform(0,1, (n_train,n_feat)) # 2D!
X_test = np.random.uniform(0,1, (n_test,n_feat)) # 2D!
y_train = np.random.randint(0,2, (n_train,n_class)) # 2D!
y_test = np.random.randint(0,2, (n_test,n_class)) # 2D!


train_data_gen = TimeseriesGenerator(X_train, y_train, length=look_back, batch_size=batch_size)
test_data_gen = TimeseriesGenerator(X_test, y_test, length=look_back, batch_size=batch_size)

# check generator dimensions
for i in range(len(train_data_gen)):
    x, y = train_data_gen[i]
    print(x.shape, y.shape)

Bi_LSTM = Sequential()
Bi_LSTM.add(Bidirectional(LSTM(512), input_shape=(look_back, n_feat)))
Bi_LSTM.add(Dropout(.5))
Bi_LSTM.add(Dense(n_class, activation='softmax'))
print(Bi_LSTM.summary())

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

Bi_LSTM_history = Bi_LSTM.fit_generator(train_data_gen,
                                        steps_per_epoch=50,
                                        epochs=3,
                                        verbose=1,
                                        validation_data=test_data_gen) # class_weight=class_weights)