标签 tensorflow 下的文章 - 白帽Wiki

假设给出如下需求:
我给你几个行为
A B C D E F
要求知道A B C D E 推测出F
这种使用场景就能使用LSTM,关于LSTM本文就不废话了,因为网上介绍一大堆了,直接上干货:

首先我们需要将数据 A B C D E F 转为编号:0 1 2 3 4 6
其次,对其进行扁平归一化,并且划分训练和测试数据：
```cpp
train_path = './result_list.csv'
data_frame = pd.read_csv(train_path)
data_frame['activity'] = data_frame['activity'].astype('float32')

scaler = StandardScaler()
data_frame['activity'] = scaler.fit_transform(
    data_frame['activity'].values.reshape(-1, 1), scaler.fit(data_frame['activity'].values.reshape(-1, 1)))

train_size = int(len(data_frame['activity']) * 0.75)
trainlist = data_frame['activity'][:train_size]
testlist = data_frame['activity'][train_size:]
```
读出来应该是:
0
1
2
3
4
5
....

然后构造滑块,成0 1 2 3 4(X), 5(Y)的样子：
```cpp
look_back = 64
trainX, trainY = create_dataset(trainlist, look_back, None)
testX, testY = create_dataset(testlist, look_back, train_size)
```
注意网上的create_dataset代码都过时了,大部分你直接抄就会报错,用我的就行:
```cpp
def create_dataset(dataset, look_back, start_index):
    dataX, dataY = [], []
    for i in range(len(dataset)-look_back-1):
        a = dataset[i:(i+look_back)]
        dataX.append(a)
        if start_index != None:
            dataY.append(dataset[start_index + i + look_back])
        else:
            dataY.append(dataset[i + look_back])

return np.array(dataX), np.array(dataY)
```
记得要reshap一下:
```cpp
trainX = trainX.reshape(trainX.shape[0], trainX.shape[1], 1)
testX = testX.reshape(testX.shape[0], testX.shape[1], 1)
```
之后直接训练即可:
```cpp
model = keras.Sequential()
model.add(keras.layers.LSTM(128, input_shape=(look_back, 1), return_sequences=True))
model.add(keras.layers.LSTM(256))
model.add(keras.layers.Dense(1))
model.compile(optimizer=keras.optimizers.Adam(), loss='mae', metrics=['MeanSquaredError'])
model.fit(trainX, trainY, epochs=26, batch_size=128)
model.save('./model_lstm.h5')
```
测试:
![](https://key08.com/usr/uploads/2021/08/1454514772.png)

白帽Wiki

一只鸭子

白帽Wiki - 一个简单的wiki

[2021]从0开始的tensorflow2.0 (三) LSTM

[2021]从0开始的tensorflow2.0 (二)