crazyleg / time-dependant-rnn-embeddings-keras Goto Github PK

Great attempt for rnns
But would be for use of
Usual classifier
Did you tried his concatenate data in sliding time window

I've read your entry and the original article several times. But I'm not clear about the pre-processing of the data to adapt it to your layer implemented in Keras.

For example, I have stock price entries (entire values of only one company). There are entries in the format of every 1 minute, every 5 minutes and every 10 minutes. It would be a matter of training an RNN with the stock quotes available so far with the 1, 5 and 10 minutes entries to predict the next value (10 minutes). How could this information be represented? The classic model based on RNN would only train with the 10 minute entries to predict the next value.

Regards,