LSTM encoder-decoder reshape data - keras

Question

I am experimenting with LSTM encoder-decoder. It is not clear to me who should I reshape the input data.
I used the following code:
import keras
import random
import numpy as np
from random import randint
from numpy import array
from numpy import argmax
from pandas import DataFrame
from pandas import concat
from keras.models import Sequential
from keras.layers import LSTM
from keras.layers import Dense
from keras.layers import TimeDistributed
from keras.layers import RepeatVector
cardinality= 10
n_steps=10
n_steps_y=3
n_features=1
def getRandomInt():
return getOneHotEncoded(random.randint(1,cardinality),cardinality)
def getOneHotEncoded(value, cardinality):
encoded = [0 for _ in range(cardinality+1)]
encoded[value] = 1
return encoded
def generateXY():
X, y = list(), list()
for q in range(100):
x_temp = [getRandomInt() for _ in range(10)]
y_temp = x_temp[-3:]
X.append(x_temp)
y.append(y_temp)
return np.array(X), np.array(y)
def getModel(n_steps=n_steps,n_features=n_features):
model = Sequential()
model.add(LSTM(12, input_shape=(n_steps,n_features)))
model.add(RepeatVector(n_steps_y))
model.add(LSTM(5, return_sequences=True))
model.add(TimeDistributed(Dense(1)))
model.compile(loss='categorical_crossentropy',optimizer='adam')
print(model.summary())
return model
X,y = generateXY()
model=getModel()
model.fit(X,y, epochs=10, batch_size=10,verbose=1)
and got error about the shape of the input.
ValueError: Error when checking input: expected lstm_1_input to have
shape (10, 1) but got array with shape (10, 11)
how should I reshape the input appropriately for this code?

Answer 1

I think what you are trying to do is passing a sequence of arrays of one-hot-encoded random numbers. Your sequences are 10 long and your arrays are 11 long.
To represent that, you need to set n_steps = 10 and n_features = 11
By the way: In encoded = [0 for _ in range(cardinality+1)], I don't quite understand the reasoning behind cardinality+1. You don't need to add 1 to represent the numbers from 0 to 9. If you change it to encoded = [0 for _ in range(cardinality)], you can set n_features = 10.
I hope this helped.