Search code examples
pythonkeraslstm

How do I train an LSTM architecture to predict number sequence?


I have implemented the following LSTM architecture. I am tring to train it to predict the number sequence but when I test it is not working. I think Iam giving the wrong input and the wrong test data.


import numpy as np
import tensorflow as tf
import keras
from keras.models import Sequential
from keras.layers import LSTM,Dense

X_train = np.array([
    [[1, 2, 3], [4, 5, 6], [7, 8, 9]],
    [[10, 11, 12], [13, 14, 15], [16, 17, 18]],
])

y_train = np.array([
    [4, 5, 6],
    [13, 14, 15],
])

#X_train = X_train.reshape((X_train.shape[0], 5, 5))

model = keras.Sequential()
model.add(keras.layers.LSTM(3,input_shape =(3, 3)))  #### The input_shape has to correspond to the input data
model.compile(loss="mean_squared_error", optimizer="adam")
model.fit(X_train, y_train, epochs=100)


X_new = np.array([[1,2,3]])
X_new = np.reshape(X_new, (1,3))

y_pred = model.predict(X_new)
print(y_pred)

Someone can give me the correct input data and test data to train this architecture?


Solution

  • One way to set the input and output sizes for your problem is as follows (not necessarily the only way):

    import numpy as np
    import tensorflow as tf
    from tensorflow import keras
    from keras.models import Sequential
    from keras.layers import LSTM, Dense
    
    # train input
    X_train = np.array([
        [[1], [2], [3]],
        [[4], [5], [6]],
        [[7], [8], [9]],
        [[10], [11], [12]],
        [[13], [14], [15]],
        [[16], [17], [18]],
    ])
    
    y_train = np.array([
        [4],
        [5],
        [6],
        [13],
        [14],
        [15],
    ])
    
    # Build the model
    model = Sequential()
    model.add(LSTM(3, input_shape=(3, 1)))  # Input shape should be (sequence_length, input_dimension)
    model.add(Dense(1))  # just one neuron for regression
    model.compile(loss="mean_squared_error", optimizer="adam")
    
    # Reshape the input data to match the model's input shape
    X_train = X_train.reshape((X_train.shape[0], 3, 1))
    
    
    model.fit(X_train, y_train, epochs=100)
    
    # Test the model
    X_new = np.array([[1], [2], [3]])  # Input a sequence of numbers
    X_new = X_new.reshape((1, 3, 1))
    y_pred = model.predict(X_new)
    print(y_pred)