Python 如何在Keras LSTM网络中使批量大小大于1？

我一直在使用批量大小为1的Keras训练LSTM，速度非常慢。我想增加批量，以加快培训时间，但我不知道怎么做

下面是显示我的问题的代码（我的最小、可复制的示例）。它的批处理大小为1，但如何使用批处理大小为2

import pandas as pd
import numpy as np

from keras.models import Sequential
from keras.layers import Dense
from keras.layers import LSTM
from keras.layers import TimeDistributed

#the demo data contains 10 instances each with 7 features and 3 targets
#the features are 0 or 1, the targets are the sum of the features in binary (with 3 bits)
raw_data = [[1,1,1,0,0,0,1,1,0,0], #has features 1,1,1,0,0,0,1, which contains 4 (100 in binary) 1's
            [0,0,0,0,0,0,0,0,0,0],[0,1,1,0,0,0,0,0,1,0],[1,0,0,1,1,1,1,1,0,1],[0,0,0,0,0,1,0,0,0,1],
            [1,1,1,1,1,1,0,1,1,0],[1,1,1,0,0,0,0,0,1,1],[1,1,1,1,1,1,1,1,1,1],[0,1,0,0,1,0,1,0,1,1],
            [1,1,1,1,0,1,1,1,1,0]]

#how can I use a batch_size of 2?
batch_size = 1
epochs = 10

df = pd.DataFrame(raw_data)
train_x, train_y =  df.values[:,0:-3], df.values[:,-3:]

#reshape <batch_size, time_steps, seq_len>  https://mc.ai/understanding-input-and-output-shape-in-lstm-keras/
#but can't reshape to a batch size of 2 as I get the ValueError below,
#ValueError: cannot reshape array of size 70 into shape (2,10,7) - which makes sense
#if I remove the batch_size from the reshape I get the ValueError below,
#ValueError: Error when checking input: expected lstm_1_input to have 3 dimensions, but got array with shape (10, 7)
train_x = np.reshape(train_x, (batch_size, train_x.shape[0], train_x.shape[1]))
train_y = np.reshape(train_y, (batch_size, train_y.shape[0], train_y.shape[1]))

model = Sequential()
model.add(LSTM(batch_input_shape=(batch_size, 10, 7), return_sequences=True, units=7))
model.add(TimeDistributed(Dense(activation='linear', units=3)))
model.compile(loss='mse', optimizer='rmsprop')

#training and testing on the same data here, but it's only example code to demonstrate my batch_size problem
history = model.fit(train_x, train_y, epochs=epochs, batch_size=batch_size, validation_data=(train_x, train_y))

yhat = model.predict(train_x, batch_size)
print(yhat)

但这给出了一个错误的值

ValueError: Error when checking input: expected lstm_1_input to have shape (1, 7) but got array with shape (5, 7)

应用

model.fit时

---

在我上面的最小可复制示例中，我如何修改它以使用2的批量

仅在

型号中使用
批量大小
。以下代码适用于我：

import pandas as pd
import numpy as np

from keras.models import Sequential
from keras.layers import Dense
from keras.layers import LSTM
from keras.layers import TimeDistributed

#the demo data contains 10 instances each with 7 features and 3 targets
#the features are 0 or 1, the targets are the sum of the features in binary (with 3 bits)
raw_data = [[1,1,1,0,0,0,1,1,0,0], #has features 1,1,1,0,0,0,1, which contains 4 (100 in binary) 1's
            [0,0,0,0,0,0,0,0,0,0],[0,1,1,0,0,0,0,0,1,0],[1,0,0,1,1,1,1,1,0,1],[0,0,0,0,0,1,0,0,0,1],
            [1,1,1,1,1,1,0,1,1,0],[1,1,1,0,0,0,0,0,1,1],[1,1,1,1,1,1,1,1,1,1],[0,1,0,0,1,0,1,0,1,1],
            [1,1,1,1,0,1,1,1,1,0]]

#how can I use a batch_size of 2?
batch_size = 2
epochs = 10

df = pd.DataFrame(raw_data)
train_x, train_y =  df.values[:,0:-3], df.values[:,-3:]

#reshape <batch_size, time_steps, seq_len>  https://mc.ai/understanding-input-and-output-shape-in-lstm-keras/
#but can't reshape to a batch size of 2 as I get the ValueError below,
#ValueError: cannot reshape array of size 70 into shape (2,10,7) - which makes sense
#if I remove the batch_size from the reshape I get the ValueError below,
#ValueError: Error when checking input: expected lstm_1_input to have 3 dimensions, but got array with shape (10, 7)
train_x = np.reshape(train_x, (1, train_x.shape[0], train_x.shape[1]))
train_y = np.reshape(train_y, (1, train_y.shape[0], train_y.shape[1]))

model = Sequential()
model.add(LSTM(batch_input_shape=(1, 10, 7), return_sequences=True, units=7))
model.add(TimeDistributed(Dense(activation='linear', units=3)))
model.compile(loss='mse', optimizer='rmsprop')

#training and testing on the same data here, but it's only example code to demonstrate my batch_size problem
history = model.fit(train_x, train_y, epochs=epochs, batch_size=batch_size, validation_data=(train_x, train_y))

yhat = model.predict(train_x, 1)
print(yhat)

将熊猫作为pd导入
将numpy作为np导入
从keras.models导入顺序
从keras.layers导入稠密
从keras.layers导入LSTM
从keras.layers导入TimeDistributed
#演示数据包含10个实例，每个实例具有7个功能和3个目标
#特征为0或1，目标为二进制特征的总和（3位）
原始#数据=[[1,1,1,0,0,1,1,0,0]，#具有特征1,1,1,0,0,1，其中包含4个（二进制为100个）1
[0,0,0,0,0,0,0,0,0,0],[0,1,1,0,0,0,0,0,1,0],[1,0,0,1,1,1,1,1,0,1],[0,0,0,0,0,1,0,0,0,1],
[1,1,1,1,1,1,0,1,1,0],[1,1,1,0,0,0,0,0,1,1],[1,1,1,1,1,1,1,1,1,1],[0,1,0,0,1,0,1,0,1,1],
[1,1,1,1,0,1,1,1,1,0]]
#如何使用2号批次？
批量大小=2
纪元=10
df=pd.DataFrame（原始数据）
列x，列y=df.值[：，0:-3]，df.值[：，-3:]
#重塑https://mc.ai/understanding-input-and-output-shape-in-lstm-keras/
#但无法将批量大小改为2，因为我得到下面的ValueError，
#ValueError:无法将大小为70的数组重塑为形状（2,10,7）-这是有意义的
#如果我从重塑中删除批次大小，我会得到下面的ValueError，
#ValueError：检查输入时出错：预期lstm_1_输入有3个维度，但得到了形状为（10,7）的数组
列车x=np.整形（列车x，（1，列车x.形状[0]，列车x.形状[1]））
序列y=np.整形（序列y，（1，序列y.形状[0]，序列y.形状[1]））
模型=顺序（）
添加（LSTM（批处理输入形状=（1,10,7），返回序列=真，单位=7））
添加（时间分布（密集（激活=线性），单位=3）））
compile（loss='mse'，optimizer='rmsprop'）
#在这里对相同的数据进行培训和测试，但这只是演示批量大小问题的示例代码
历史=模型.fit（序列x，序列y，年代=年代，批量大小=批量大小，验证数据=（序列x，序列y））
yhat=模型预测（列车×1）
印刷品（yhat）
仅在
型号中使用
批量大小。以下代码适用于我：

import pandas as pd
import numpy as np

from keras.models import Sequential
from keras.layers import Dense
from keras.layers import LSTM
from keras.layers import TimeDistributed

#the demo data contains 10 instances each with 7 features and 3 targets
#the features are 0 or 1, the targets are the sum of the features in binary (with 3 bits)
raw_data = [[1,1,1,0,0,0,1,1,0,0], #has features 1,1,1,0,0,0,1, which contains 4 (100 in binary) 1's
            [0,0,0,0,0,0,0,0,0,0],[0,1,1,0,0,0,0,0,1,0],[1,0,0,1,1,1,1,1,0,1],[0,0,0,0,0,1,0,0,0,1],
            [1,1,1,1,1,1,0,1,1,0],[1,1,1,0,0,0,0,0,1,1],[1,1,1,1,1,1,1,1,1,1],[0,1,0,0,1,0,1,0,1,1],
            [1,1,1,1,0,1,1,1,1,0]]

#how can I use a batch_size of 2?
batch_size = 2
epochs = 10

df = pd.DataFrame(raw_data)
train_x, train_y =  df.values[:,0:-3], df.values[:,-3:]

#reshape <batch_size, time_steps, seq_len>  https://mc.ai/understanding-input-and-output-shape-in-lstm-keras/
#but can't reshape to a batch size of 2 as I get the ValueError below,
#ValueError: cannot reshape array of size 70 into shape (2,10,7) - which makes sense
#if I remove the batch_size from the reshape I get the ValueError below,
#ValueError: Error when checking input: expected lstm_1_input to have 3 dimensions, but got array with shape (10, 7)
train_x = np.reshape(train_x, (1, train_x.shape[0], train_x.shape[1]))
train_y = np.reshape(train_y, (1, train_y.shape[0], train_y.shape[1]))

model = Sequential()
model.add(LSTM(batch_input_shape=(1, 10, 7), return_sequences=True, units=7))
model.add(TimeDistributed(Dense(activation='linear', units=3)))
model.compile(loss='mse', optimizer='rmsprop')

#training and testing on the same data here, but it's only example code to demonstrate my batch_size problem
history = model.fit(train_x, train_y, epochs=epochs, batch_size=batch_size, validation_data=(train_x, train_y))

yhat = model.predict(train_x, 1)
print(yhat)

将熊猫作为pd导入
将numpy作为np导入
从keras.models导入顺序
从keras.layers导入稠密
从keras.layers导入LSTM
从keras.layers导入TimeDistributed
#演示数据包含10个实例，每个实例具有7个功能和3个目标
#特征为0或1，目标为二进制特征的总和（3位）
原始#数据=[[1,1,1,0,0,1,1,0,0]，#具有特征1,1,1,0,0,1，其中包含4个（二进制为100个）1
[0,0,0,0,0,0,0,0,0,0],[0,1,1,0,0,0,0,0,1,0],[1,0,0,1,1,1,1,1,0,1],[0,0,0,0,0,1,0,0,0,1],
[1,1,1,1,1,1,0,1,1,0],[1,1,1,0,0,0,0,0,1,1],[1,1,1,1,1,1,1,1,1,1],[0,1,0,0,1,0,1,0,1,1],
[1,1,1,1,0,1,1,1,1,0]]
#如何使用2号批次？
批量大小=2
纪元=10
df=pd.DataFrame（原始数据）
列x，列y=df.值[：，0:-3]，df.值[：，-3:]
#重塑https://mc.ai/understanding-input-and-output-shape-in-lstm-keras/
#但无法将批量大小改为2，因为我得到下面的ValueError，
#ValueError:无法将大小为70的数组重塑为形状（2,10,7）-这是有意义的
#如果我从重塑中删除批次大小，我会得到下面的ValueError，
#ValueError：检查输入时出错：预期lstm_1_输入有3个维度，但得到了形状为（10,7）的数组
列车x=np.整形（列车x，（1，列车x.形状[0]，列车x.形状[1]））
序列y=np.整形（序列y，（1，序列y.形状[0]，序列y.形状[1]））
模型=顺序（）
添加（LSTM（批处理输入形状=（1,10,7），返回序列=真，单位=7））
添加（时间分布（密集（激活=线性），单位=3）））
compile（loss='mse'，optimizer='rmsprop'）
#在这里对相同的数据进行培训和测试，但这只是演示批量大小问题的示例代码
历史=模型.fit（序列x，序列y，年代=年代，批量大小=批量大小，验证数据=（序列x，序列y））
yhat=模型预测（列车×1）
印刷品（yhat）
太好了，很乐意帮忙！太好了，很高兴帮忙！