I reuse an old code who was perfectly functionnal 3 mount ago to train a new LSTM model on new set of data.
my data have these shapes:
X_train = [21500, 5, 4]
y_train = [21500, 1]
trainning model seem to work well, it finish with a single
tensor_input = Input(shape=(x_train.shape[1], x_train.shape[2]), name='main_inputs')
xy = LSTM(6, activation='softsign',
kernel_initializer=initializers.glorot_uniform(),
return_sequences=False, stateful=False,
name='Hlayer1')(tensor_input)
xy = Dropout(rate=.2)(xy)
Dense(y_train_shape[1], activation='linear')
but when I try to predict with
yhat = model.predict(x_test)
x_test = [1350,1]
I get
yhat = [1350, 5, 1]
I build my code this way since 5 years but first time I get this!??!
Never find the trouble, just reinstall Tensorflow and now it work!?