How to predict past test set on time series data using LSTM

Question

I'm trying to do a regression on some inventory amounts with the following model using Keras:

model = Sequential()
model.add(LSTM(100, batch_input_shape=(BATCH_SIZE, TIME_STEPS, 
X_train_timeseries.shape[2]), dropout=0.0, 
           recurrent_dropout=0.0, stateful=True, 
kernel_initializer='random_uniform'))
model.add(Dropout(rate=0.5))
model.add(Dense(20, activation='relu'))
model.add(Dense(1, activation='sigmoid'))
optimizer = optimizers.RMSprop(lr=LEARNING_RATE)
model.compile(loss='mean_squared_error', optimizer=optimizer, metrics=['mae'])

I have a time step of 60, how do I predict past my test data to make actual predictions into the future?

score 1 · Answer 1 · answered Jul 31 '19 at 19:22

1

The best answer I found is here

Unfortunately I couldn't get it to work for my solution but reworked my solution and just incremented the sliding window, appended that answer to my test data, and iterated like that.

answered Jul 31 '19 at 19:22

Seth

31
2

How to predict past test set on time series data using LSTM

1 Answers1