python-kerasHow do I use a Python Keras LSTM for a specific example?

The following example code shows how to use a Python Keras LSTM for a specific example. This example uses a simple dataset of 1000 samples consisting of two features and a label. The label is binary, and the features are randomly generated.

# import libraries
import numpy as np
from keras.models import Sequential
from keras.layers import Dense, LSTM

# generate dataset
data_dim = 2
timesteps = 8
num_classes = 2

# expected input data shape: (batch_size, timesteps, data_dim)
x_train = np.random.random((1000, timesteps, data_dim))
y_train = np.random.randint(2, size=(1000, 1))

# build model
model = Sequential()
model.add(LSTM(32, return_sequences=True,
               input_shape=(timesteps, data_dim)))
model.add(LSTM(32, return_sequences=True))
model.add(LSTM(32))
model.add(Dense(2, activation='softmax'))

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

# train model
model.fit(x_train, y_train,
          batch_size=64, epochs=5)

The code above builds a LSTM model with three LSTM layers and a Dense layer. The input shape is set to (batch_size, timesteps, data_dim). The model is then compiled with the categorical_crossentropy loss function and the rmsprop optimizer. Finally, the model is trained on the generated dataset.

Code explanation

1. import numpy as np: imports the NumPy library as np
2. from keras.models import Sequential: imports the Sequential model from Keras
3. from keras.layers import Dense, LSTM: imports the Dense and LSTM layers from Keras
4. data_dim = 2: sets the dimension of the data to 2
5. timesteps = 8: sets the number of timesteps to 8
6. num_classes = 2: sets the number of classes to 2
7. x_train = np.random.random((1000, timesteps, data_dim)): generates a dataset of 1000 samples with two features and a label
8. y_train = np.random.randint(2, size=(1000, 1)): generates a binary label for the dataset
9. model = Sequential(): creates a sequential model
10. model.add(LSTM(32, return_sequences=True, input_shape=(timesteps, data_dim))): adds a LSTM layer to the model with 32 units, return_sequences=True to return the output of each timestep, and input_shape set to the data shape
11. model.add(LSTM(32, return_sequences=True)): adds another LSTM layer to the model with 32 units and return_sequences=True
12. model.add(LSTM(32)): adds another LSTM layer to the model with 32 units
13. model.add(Dense(2, activation='softmax')): adds a Dense layer with 2 units and the softmax activation function
14. model.compile(loss='categorical_crossentropy', optimizer='rmsprop', metrics=['accuracy']): compiles the model with the categorical_crossentropy loss function, the rmsprop optimizer, and accuracy as the metric
15. model.fit(x_train, y_train, batch_size=64, epochs=5): trains the model on the dataset with a batch size of 64 and 5 epochs

Helpful links

• Keras Documentation - LSTM
• Keras Documentation - Sequential Model

Edit this code on GitHub